Recurrent Neural Networks

Long-sequence memory tracking places two opposing demands on a recurrent state: near-lossless retention of stored bindings over long horizons, and active overwriting of stale ones. In our diagnostic suite, the strongest efficient baselines tend to solve only one side well. Continuous-time-parameterized state-space models (SSMs) such as Mamba obtain their discrete recurrence by zero-order-hold discretization of a continuous-time system; we argue that this detour is unnecessary for memory tracking and parameterize the discrete transition directly. Naju (Native Adaptive Junction Unit) factorizes the recurrent update, schematically $x_n = f_n\odot x_{n-1} + i_n\odot(B_n u_n)$, into an explicit discrete pole (a learned forget gate $f_n$), an independent write gain $i_n$, and input-dependent write/read maps. Since the sigmoid pole satisfies $0<f_n<1$, each frozen local coordinate is Schur-stable by construction, and the full time-varying recurrence satisfies a fading-memory/BIBO bound under uniform boundedness assumptions, with no stability regularizer. We formalize the key structural limitation of coupled designs: any non-expansive complementary single-gate recurrence ties the effective retention $r$ and write gain $w$ through $|r|+w\le 1$, so near-complete retention forces weak writing; decoupling $f_n$ from $i_n$ removes this constraint. Empirically, Naju is the only evaluated model that remains strong on both retention and overwriting at 4x the training length. Beyond the diagnostic suite, we evaluate Naju on WikiText-103 language modeling, Long Range Arena, and multi-query associative recall. Across these settings, Naju consistently combines strong long-range memory with competitive or superior performance, outperforming the Mamba baselines in the principal comparisons while remaining competitive with the Transformer and preserving linear-time, linear-memory scaling.

When does a weight-tied looped transformer -- one block applied T times -- implement an actual algorithm? We answer with four findings from controlled populations on group word problems. (1) The budget law: free training installs a linear computation frontier, a mechanism that solves v positions per loop, whose speed is priced by the training contract: v ~ n_train/T_train (exponent 0.98 +/- 0.04, R^2=0.99), exactly unity under T=n training. SGD selects a frontier matching the minimum the contract demands; granting more test-time loops than ever trained rescues late positions at fixed input length, yielding a principled halting rule T* = ceil(n / v-hat). (2) Architecture prior, not expressivity, picks the algorithm: standard-depth transformers learn parallel scans on this family; weight tying flips the selection to the serial frontier, even when positional addressing for a log-depth scan is supplied. At matched depth and parameters, untied models extrapolate worst and fail to learn A5 at all. (3) The walls are not where circuit complexity says: NC1-completeness costs nothing (A5 generalizes fully), while group order does (S5's 120x120 operator deadlocks joint learning) -- and an operator-first curriculum dissolves the wall in every seed. (4) Mechanisms are portable, not mandatable: warm-starting across budget contracts transfers the algorithm in every seed, re-pricing its speed, while imposing seriality through the input schedule fails where free training succeeds. These results are invisible to standard instruments, which provably saturate at the fixed points trained loops converge to. We introduce a head instrument, the convergence-time scaling tau(n,i), validate it causally via damage cones whose slope reproduces v, and show in-distribution head measurements predict out-of-distribution fate where tail metrics do not. Results replicate on the public easy-to-hard benchmark.

We propose a hybrid approach for user-centric modeling of transactional event sequences that combines contrastive representation learning (CoLES) with State Space Models (SSMs). While contrastive methods yield high-quality compressed user representations, existing encoders -- RNNs and Transformers -- suffer from vanishing gradients or quadratic complexity, respectively. Mamba, a selective SSM, efficiently handles long-range dependencies but remains underexplored for personalized user analysis. We investigate two integration strategies: (1)~initializing the Mamba hidden state with a CoLES embedding, and (2)~prepending the projected CoLES embedding as a prefix token to the input sequence. Both approaches supply the model with an informative user prior from the first step. Experiments on three public datasets -- Age (multiclass age-group prediction), MBD (multi-label product acquisition), and Taobao (binary purchase prediction) -- demonstrate consistent improvements over standalone Mamba and CoLES with a linear classifier, with the hybrid models converging 2--3$\times$ faster than the plain SSM baseline. Explainability analysis via discretization-step maps and Integrated Gradients reveals selective event filtering on behavior-rich datasets and identifies the most informative transaction features.

Modeling galaxy-galaxy strong gravitational lenses to infer the brightness of the source galaxy and the mass distribution of the foreground galaxy is computationally challenging, particularly for high-resolution, high signal-to-noise ratio observations. In this regime, high-dimensional representations of both the source and the foreground mass distribution are necessary to model the data down to the noise level. This inference problem has been challenging for both traditional and machine learning-based methods because of its high dimensionality and its non-linearity in the foreground mass distribution. We present a method to generate joint posterior samples of the source galaxy and foreground mass distribution as pixelated images conditioned on observations. The method combines diffusion-based generative modeling and recurrent inference machines. It can model realistic gravitational lensing simulations with background and foreground galaxies drawn from cosmological hydrodynamical simulations down to the noise level.

Recent advances in deep-learning-based nonlinear system identification have led to encoder-based estimation of neural state-space (ANN-SS) models that achieve state-of-the-art performance in offline settings by estimating initial model states from past input-output data. These methods are typically used in multiple-shooting-based offline identification, and online learning of these models remains largely unexplored. This paper presents a batch-wise learning pipeline and a direct recursive identification algorithm for subspace encoder-based ANN-SS models. We provide convergence analysis of the recursive formulation and validate its performance through extensive simulation studies. The results demonstrate that the proposed approach enables computationally efficient online adaptation with high model accuracy.

Bladder cancer treatment requires personalized and adaptive decision-making, particularly for recurrent disease, where treatment effectiveness changes across successive clinical episodes. Conventional clinical decision support systems typically rely on static treatment guidelines or single-step predictive models, limiting their ability to capture disease progression over time. This paper presents a recurrent patient state-transition simulation framework for bladder cancer treatment planning that integrates predictive state-transition modeling with a Markov Decision Process (MDP) and a Deep Q-Network (DQN) reinforcement learning environment. The predictive module estimates changes in tumor characteristics following treatment, while the reinforcement learning agent sequentially optimizes treatment decisions by interacting with simulated patient trajectories. This framework enables dynamic, patient-specific treatment planning by continuously adapting recommendations to evolving clinical states. It also generates interpretable treatment trajectories and detailed simulation logs to improve transparency and support clinical decision-making. The proposed framework was evaluated against existing reinforcement learning-based treatment planning approaches. It achieved a cumulative reward of 63,918.87, an average training loss per episode of 0.0056, and a policy improvement score of 6.62%, demonstrating effective sequential learning and robust treatment optimization in a simulated recurrent treatment environment. These findings highlight the potential of recurrent patient state-transition simulation with reinforcement learning as a flexible decision-support framework for personalized bladder cancer treatment planning and AI-assisted precision oncology.

Linear-attention models replace the growing KV cache with recurrent states, but autoregressive decoding still reads, updates, and writes these states one token at a time. Speculative decoding can reduce this cost by verifying several draft tokens in one target pass, yet existing speculative systems are designed for Transformer KV caches. For stateful linear-attention targets, verification must follow recurrent dependencies across chains and branches, acceptance must update only the accepted state trajectory, and the drafter must avoid submitting candidates that waste stateful verification work. This paper presents SpecLA, a speculative decoding runtime for stateful linear-attention models. SpecLA verifies chains and trees with topology-aware kernels, stores compact factors produced during verification to recover accepted states, and uses confidence pruning plus a target-aligned EAGLE-style drafter to feed useful candidates to the verifier. On an NVIDIA H100 with a public GDN-1.3B target, SpecLA achieves up to 1.70x end-to-end speedup over autoregressive decoding.

Transformer reasoning is limited by autoregressive decoding, which repeat edly compresses rich hidden computation through token space and makes it difficult for intermediate reasoning states to persist across time. We in troduce Transformers with Temporal Middle-Layer Recurrence (T2MLR), a transformers-based latent reasoning architecture that fuses a cached middle layer representation from the previous token directly into an earlier layer of the current token position, enabling abstract intermediate computation to persist across decoding steps with little inference overhead. Across natural-language pretraining and multi-hop reasoning finetuning, T2MLR consistently outperforms data- and parameter-matched Transformer base lines. Moreover, applying recurrence to only a localized middle-layer block (as little as 20% of the network) often outperforms full-layer recurrence. Im portantly, T2MLR does not require pretraining from scratch: retrofitting the recurrent pathway into an existing pretrained 1.7B Transformer and briefly finetuning substantially improves math reasoning, lowering the barrier to practical adoption. These results suggest that effective latent reasoning in Transformers does not require looping over all layers as in previous works, but can instead emerge more strongly from targeted middle-layer recurrence.

A growing gap separates inference context lengths from RL post-training: inference systems are approaching million-token contexts, while post-training workloads often remain at 256K tokens or below and rely on length generalization at deployment. The gap is especially important for AI agents, whose observations, tool outputs, documents, and prior decisions accumulate over long trajectories. LongStraw is an architecture-aware execution stack for million-token RL post-training under a fixed GPU budget, instantiated with Group Relative Policy Optimization (GRPO). It evaluates the shared prompt without autograd, retains only model-specific state needed by later tokens, and replays short response branches one at a time, reducing the live training graph at the cost of additional replay time. We implement it for the hybrid recurrent and full-attention Qwen3.6-27B and the compressed-attention mixture-of-experts GLM-5.2. On eight H20 GPUs, LongStraw completes grouped Qwen scoring and response backward at 2.1M positions for groups of 2 and 8; increasing the group size adds only 0.21 GB of peak allocated memory, while a separate stress test reaches 4.46M positions. On 32 H20 GPUs, we validate the end-to-end LongStraw execution path for a 2.1M-token prompt across all 78 layers of GLM-5.2. These experiments establish execution capacity rather than complete training correctness because the captured prompt state is detached and some distributed forward and gradient composition paths remain incomplete.

A depth-recurrent transformer applies a weight-tied core a variable number of times, and prior work has shown that training with a randomized recursion count yields one checkpoint usable across a range of inference depths. We ask what such a model actually computes per token, and measure it directly. On a 135M-class model trained on FineWeb-Edu, the recurrent state converges to a per-token fixed point: mean successive-output KL divergence falls from 3.9e-1 at the second loop to 8.5e-6 by the sixteenth, and per-token state change decays in step. Crucially, this convergence is not uniform across tokens. The median token converges by loop six, while approximately 10 percent of tokens continue to update at the training-mean depth of eight, and mean convergence depth is ordered by token type (whitespace shallowest, content words deepest). This per-token variation is the central object of the paper. We show it is directly readable and that reading it outperforms learning to predict it: a training-free rule that halts each token once its output stabilizes attains uniform depth-8 quality at 4.94 average loops (a 38 percent reduction in average depth) and matches uniform depth across the average-depth range, whereas a linear router trained on convergence labels harvested from the same model requires nearly full depth and yields no reduction. The elasticity that makes this possible reproduces here as background (validation loss decreases monotonically from 3.80 at one loop to 3.20 at eight and remains stable to 32 loops). We report average depth as a FLOP proxy with a three-point wall-clock bracket rather than a realized speedup, make no FLOP-matched parity claim, and note that the allocation results are established at a single scale and seed. The complete study runs on a single RTX 4090 in approximately 100 GPU-hours.

Incorporating hysteresis and eddy currents into finite element simulations of laminated-core electrical machines is computationally challenging. Resolving the fields inside the laminations at each integration point and at every nonlinear iteration leads to computational costs several orders of magnitude higher than anhysteretic simulations, making such approaches impractical for design applications. Conversely, simplified models accounting only for magnetic saturation are becoming increasingly inadequate as electrical machine topologies and operating conditions grow in complexity. In this context, machine learning surrogate modeling has emerged as a promising alternative, offering efficient and accurate approximations of complex electromagnetic behaviors. In this paper, a recurrent neural network is trained as a surrogate of a laminated-core material model for an isotropic laminated core, and is integrated into realistic two-dimensional magnetodynamic finite element simulations based on a magnetic vector potential formulation. The proposed approach achieves excellent agreement with the reference laminated-core model while limiting the computational cost to about twice that of an anhysteretic simulation. By training the recurrent neural network on a sufficiently diverse set of artificially generated magnetic field sequences designed to mimic those encountered in electrical machine simulations, the proposed approach can be readily applied across a wide range of finite element simulations. Furthermore, the trained surrogate model is provided as a standalone component that can be easily incorporated into existing computational frameworks. It is publicly available at https://gitlab.onelab.info/getdp/lamnet.

Building interactive worlds that respond coherently to player actions has long been a shared goal of computer graphics, games, and artificial intelligence. Recent video generative models provide a data-driven route toward this goal by predicting future observations conditioned on user actions, and are increasingly regarded as potential next-generation game engines. Realizing a genuinely interactive game world, however, requires interaction outcomes that follow rules over evolving game conditions, consequences that persist over long horizons, and a generation loop that operates in real time. Conventional game engines realize these properties through a recurrent action-state-observation loop, in which player actions update an explicit game state according to predefined rules and observations are rendered from the resulting state. Taking this loop as an organizing lens, this paper examines interactive game world modeling along four dimensions: player action control, game state dynamics, state-observation persistence, and real-time interactive generation. For each dimension, we start from the capabilities required by an interactive game world, group existing approaches into representative families, and discuss the strengths and trade-offs of each family. Complementing this analysis, we present a scalable data engine for Black Myth: Wukong that collects over 90 hours of gameplay with frame-aligned player actions, ground-truth game states, and visual observations, together with structured and semantic annotations, as a resource for state-aware game world modeling. We hope this paper offers a clear picture of where the field stands and fosters progress toward interactive game worlds.

Smart-building load forecasters are often trained offline on dense, multivariate, high-frequency data, but deployment may provide only hourly, feature-limited inputs. Missing features must then be reconstructed, and their errors can propagate through the model. If this input uncertainty is not reflected, prediction intervals may become miscalibrated, affecting demand-response scheduling. Our work examines where uncertainty should be placed once inference inputs are reconstructed. We develop a unified one-day-ahead probabilistic forecasting framework that aligns temporal resolution, reconstructs the unavailable inputs, and derives causal features, and we compare a modular post-hoc residual-quantile scheme with an integrated in-model quantile-learning scheme. The comparison uses three mid-scale Deep Learning (DL) backbones: recurrent, hybrid recurrent, and attention-based Temporal Fusion Transformer (TFT) models, under identical inputs, forecasting horizon, preprocessing rules, and training budgets. Results show that uncertainty placement is backbone-dependent. Integrated quantile learning is most reliable with the TFT, yielding 2.2-3.6% MAPE and 28-83W RMSE on the labeled test window, while producing intervals about 5x narrower than the modular intervals at the closest-to-nominal coverage level. Diebold-Mariano tests support the TFT ranking and the mixed behavior of the recurrent backbones. A reconstruction-sensitivity test shows that reconstructed inputs increase the Quantile Score (QS) by 106% while interval width remains nearly unchanged, indicating that the model does not automatically absorb reconstruction-induced uncertainty. Robustness checks against non-DL baselines and seasonal hold-out weeks support this ranking. Our results expose the limits of post-hoc residual quantiles when inference depends on reconstructed inputs.

Growing Neural Cellular Automata (GNCA) are capable of robust self-maintenance and self-repair, yet the internal dynamical mechanisms that support these capabilities remain poorly understood. Here, we investigate the role of internal fluctuations--temporal micro-variability of hidden channel states--in a trained GNCA model, challenging the assumption that such variability is merely residual stochastic noise. Through systematic analysis spanning update-rate sweeps, spatial correlation measurements, dimensionality reduction of collective state trajectories, localized damage experiments, transfer entropy vector field estimation, and partial information decomposition, we show that internal fluctuations are spatially structured, dynamically coupled to an attracting collective state, and associated with distributed small-magnitude updates that contribute to damage recovery. Damage induces a global deviation in latent state space followed by gradual re-convergence, and suppressing distributed small-magnitude updates associated with baseline fluctuation dynamics outside a permissive radius that encompasses the majority of the cells significantly impairs recovery. Transfer entropy analysis characterizes a spatially differentiated repair response: corrective inward flow near the damage site coexists with outward perturbation propagation at greater distances. Partial information decomposition further suggests a regime shift from synergy-dominant resting computation to redundancy-increased coordination during recovery. These findings indicate that GNCA self-repair emerges from high-dimensional nonlinear collective dynamics in which internal fluctuations serve as a functional component supporting information flow, coordination, and return toward an attracting recurrent state.

A substantial number of patients experience diminished mobility due to disabilities, diseases, or accidents. Although modern prostheses, powered by deep neural networks, hold the promise of significantly enhancing the quality of life for these individuals, their widespread adoption is hindered by significant latency, energy consumption, and spatial requirements. Wired connections to external high-performance processors restrict patient mobility, while wireless connections limit the volume of information that can be transmitted to these processors. Spiking neural networks offer the potential for compressed communication and low-power inference, yet they often lag behind state-of-the-art deep learning models in various applications. In this study, we propose a high-performance neural decoding method that effectively balances task performance and efficiency. An eventbased gated recurrent unit generates a sparse communication pattern with graded spikes, surpassing classical spiking neural networks in terms of task performance. Utilising an efficient training method and sparse inference, our model presents new opportunities for on-device neural decoding.

In pre-LayerNorm looped transformers, LayerNorm inside the recurrent block acts as an implicit gain controller: by coupling the block's local Lipschitz constant inversely to the activation scale, it renders the recurrence Jacobian non-normal -- asymptotically contractive at every verified fixed point even where its operator norm exceeds 1 -- so the true stability budget is the spectral margin, not an operator-norm bound. That margin depletes as the carry $ρ\to 1$, and a minority of initializations never converge to a fixed point at all, so the diagonal carry constraint $ρ(\bar{A}) < 1$ is necessary but not sufficient for convergence of the full recurrence. Training experiments across six tasks, including a controlled ablation, reveal that the linear carry is not the depth-memory mechanism: gradient descent routes memory through the block's more expressive nonlinear recurrence and leaves the stability-constrained carry at rest -- the carry's role is stabilization, not memory. We characterize the boundary of this claim: on tasks with axis-aligned per-channel structure, gradient descent does recruit the carry. All results are derived analytically and verified in a from-scratch, CPU-scale implementation; verification at larger scale is needed.

Recursive models show promise on reasoning and language tasks, yet their test-time scaling lacks a principled criterion for selecting trajectories or determining recurrent depth. We introduce \textbf{Energy-guided Recursive Model (ERM)}, which uses Hopfield-type memories of valid local and global structures to assign intrinsic energies to candidate trajectories. These energies guide candidate selection and suggest an effective range of recurrent depths, implying that deeper recurrence does not necessarily improve reasoning accuracy. They also enable sampling methods such as parallel tempering to improve exploration. For reasoning tasks, ERM achieves optimal solutions on Sudoku ($98.97\%$), Pencil Puzzle Bench (PPBench, $88.04\%$) and Maze ($99.30\%$), reaching the best accuracy in recursive modeling. On language modeling, ERM reduces RedPajama-V2 perplexity by $1.74\%$ with marginal inference overhead. The results support energy guidance as a practical framework for improving test-time scaling in recursive models.

Recent work on looped language models suggests that many reasoning problems benefit from greater computational depth rather than from additional independent parameters. Existing studies, however, focus almost exclusively on Transformer backbones, leaving open whether this principle also applies to state-space language models. We investigate Looped Mamba and Looped Hybrid Mamba-Transformer architectures, which repeatedly apply a shared Mamba (or hybrid) block to introduce explicit finite-depth recurrent computation. On two controlled reasoning tasks-Mano (modular-arithmetic manipulation) and p-hop induction-Looped Mamba consistently outperforms parameter-matched non-looped baselines and, in several settings, matches or exceeds non-looped models of equal effective depth. We then extend the study to language model pre-training under matched iso-parameter and iso-FLOPs protocols, which jointly disentangle the effects of parameter sharing and effective depth: looped models remain competitive on downstream benchmarks with substantially fewer distinct parameters, although deeper non-looped models retain an advantage in validation perplexity under strict iso-FLOPs comparisons. Finally, we adapt Ouro's two-stage exit gate to Looped Mamba for threshold-controlled selection among recurrent-step outputs. Executing such exits on a state-space backbone, however, leaves the recurrent state without its deeper updates, and validation perplexity then degrades severely. We therefore introduce a cache-hole adaptation that aligns continued training with skipped-state inference. At the scales studied, the adapted model keeps perplexity close to full computation and matches or exceeds full-compute exit-state selection on downstream benchmarks while executing roughly half of the recurrent steps, which translates into measured inference speedups once the prefill is compute-bound.

This paper proposes a new notion of robust invertibility for nonlinear dynamical systems, and introduces constructive parameterizations of recurrent neural network which are robustly invertible by design. We define robust invertibility as the existence of a causal inverse system such that both the forward and inverse systems are contracting and have bounded incremental input-output gains (the system is bi-Lipschitz), implying that both forward prediction and input reconstruction are robust to signal perturbations and initial-state mismatch. We construct robustly invertible recurrent models via series composition of static orthogonal layers and dynamic layers satisfying a strong input-output monotonicity property, and provide a differentiable neural network parameterizations in the form of the bi-Lipschitz recurrent equilibrium network (BiLipREN). Additionally, composition with dynamic orthogonal layers yields a nonlinear minimum-phase/all-pass (a.k.a. inner--outer) factorization. We illustrate the utility of the framework through a series of application examples in data-driven internal model control, dynamic surrogate loss learning, and signal-space normalizing flows, illustrating its utility for robust control, trajectory optimization, and generative modeling of complex trajectory distributions.

Neural networks can learn algorithmic input-output mappings, but trusting a learned executor requires more than a correct final answer because the state transitions that produce it are usually hidden. To make those transitions visible, we introduce a trace-supervised symbolic neural CPU, a factorized learned execution architecture that combines recurrent control, an explicit operation router over a fixed differentiable arithmetic-logic unit bank, destination-masked register writeback, complete trajectory supervision and matched fixed-point replay. The model exposes the selected operation, source and destination registers, register trajectory, memory signals and writeback semantics at every step. On the principal 16-wide benchmark, the non-quantized executor reproduces reference execution exactly, while the eight-bit quantization-simulated executor preserves the symbolic operation path through programs of 1,000 instructions. When the same execution is evaluated against a matched fixed-point replay, the residual numerical drift disappears, showing that it comes from a mismatch between continuous and low-precision reference semantics rather than from execution failure. We compare recurrent, Transformer, temporal-convolution, temporal graph-inspired and state-space controllers, and the ablations show that operation-gate supervision is necessary for an inspectable execution path. Hidden-opcode memory-pressure tasks expose the remaining limits in delayed state use and temporal binding. We also extend the interface with ValueMemory, hybrid adaptive leaky integrate-and-fire controllers, candidate-constrained symbolic control trained through behaviour cloning and actor-critic reinforcement learning, and an RV32I base-integer semantic bridge. Together, these results establish a trace-verifiable framework for interpretable, low-precision and controllable neural execution.

Fixed-state sequence models compress an unbounded past into a bounded state, which caps their associative recall at roughly the state dimension; attention escapes the cap by keeping a key-value entry for every token, at quadratic compute and a cache that grows with the sequence. We study the middle ground: a sparse cache that allocates a slot only when an input is novel, so its size tracks the number of distinct items rather than the number of tokens. The allocation rule is the DP-means clustering rule, the small-variance limit of a Dirichlet-process mixture, used not as latent-variable inference but as the key-value memory operator for a deep recurrent backbone. We develop it in two forms, a static cache with a fixed concentration and a surprise-adaptive variant whose concentration follows the recent novelty rate. On a controlled associative-recall benchmark with redundancy we show that the cache matches full-attention recall while storing only the distinct items, that it dominates a fixed-budget eviction cache on the recall-versus-size frontier, and that on a state-space backbone it answers both a recall query and a long-range aggregate at the lowest memory of any model tested. The allocation is learnable end to end: a two-parameter novelty-threshold gate trained on the task loss alone recovers the rule exactly, whereas an over-parameterized gate fails, so the operative ingredient is the inductive bias rather than capacity. The evidence is a family of controlled mechanism studies at modest scale, with the distinct-items property confirmed on four real streams (recommendation, systems logs, clinical events, and insurance claims); a real-backbone, real-corpus language validation is pursued in a companion study.

Medieval document transcribers have very different practices; on top of that, heterogeneous digitization policies have resulted in corpora where the character-set must be viewed as fluid. In this paper we address the problem of changing between character-sets in a flexible manner. We focus on one-to-one character mappings and train characterlevel one-to-one RNNs to undo them with self-supervision; recovering half the CER even with 20 text lines. We analyse the use of these one-to-one networks for HTR post-correction and we see that they obtain significant improvements while totally ignoring ins-dels. We then use the exact same networks with character-level alignment groundtruth compiled from parallel corpora in a training and inference mode we call Banded RNNs. We use such networks to successfully expand abbreviations in medieval charter transcriptions. Finally we introduce an elaborate heuristic which takes the characters of two arbitrary character-sets and defines a metric encapsulating what we consider to be semantic similarity of characters. We call the construction of such mappings letter lemmatization and present a rich Python library that efficiently performs all presented methods.

Modern sequence models are increasingly deployed as agents that interleave token generation with calls to external tools. We give an exact, architecture-level account of when such tool access increases computational expressivity. We model any fixed finite-precision recurrent sequence model, including finite-precision state-space models (SSMs) with $B$ bits of internal state, as a deterministic finite-state controller interacting with an oracle through a finite command/observation interface. Our results form a sharp dichotomy. First, tools that are themselves finite-state add essentially nothing: a product-state simulation internalizes any finite-state bounded-interface oracle with finite memory set $M$ at a cost of only $\log_2 |M| + O(1)$ additional bits, so the augmented system remains finite-state. Second, a single minimal infinite-state tool, namely a tape supporting only local $\mathtt{read}$, $\mathtt{write}$, and $\mathtt{move}$ commands, makes the system Turing complete: for every single-tape Turing machine with state set $Q$ and tape alphabet $Γ$, a controller with $O(\log |Q| + \log |Γ|)$ bits of internal memory simulates it, and we exhibit a concrete exponential separation: $\mathrm{EQ}_n$ requires $2^n$ states without tools but a single constant-size controller with the tape tool. Third, we show that this construction is realized exactly by a natural one-layer finite-precision selective affine SSM controller with binary one-hot hidden states, $\{0,1\}$ transition matrices, and zero biases. Selectivity is essential to the construction. In the supplementary material, we make all constants explicit, prove a logarithmic oracle-assisted universal simulation, where $O(\log B)$ recurrent bits suffice to simulate any $B$-state Turing machine, and prove a matching impossibility result.

Contemporary language models are dominated by the transformer architecture, which leverages self-attention mechanisms to enable more efficient, parallelized training across a wide set of documents and corpora. This has allowed transformers to effectively model data across a wide range of modalities and contexts. However, transformers, along with their conventional counterparts such as recurrent neural networks (RNNs) and convolutional neural networks (CNNs), often struggle to maintain efficiency when processing long contexts. We introduce ResonatorLM, a new mechanism that replaces attention with a physics-derived alternative. ResonatorLM treats token sequences as a single, driven one-dimensional latent field and replaces attention dot products with causal functions of damped resonators. We implement ResonatorLM on a traditional network architecture and test it on standard long-context modeling tasks. We find that in a small, 6M matched setting, training and prefill speedups increase with sequence length, decode speed reaches 6.47x compared to that of a standard, optimized transformer at 32K tokens, and accuracy reaches 61.31 percent (compared to 55.32 percent) on WikiText.

Streaming 3D reconstruction relies on a compact recurrent scene state to process long image streams in linear time and bounded memory. However, repeated updates can gradually corrupt this state, causing reliable historical information to be overwritten by noisy or ambiguous observations. We introduce ReCal3R, a reliability-calibrated learning rate method for recurrent 3D reconstruction. Instead of directly applying a candidate learning rate, our method estimates state token reliability from the maintained scene state and uses it to calibrate a candidate learning rate derived from token alignment, state reconstruction residual, and recent update pressure. The resulting token-wise learning rate interpolates between a conservative base rate and the candidate rate, suppressing aggressive updates on unreliable tokens while preserving adaptation to informative frames. Applied to CUT3R as a training-free calibration rule, ReCal3R reaches strong performance on long sequences in pose, depth, and reconstruction quality, including a 3.7$\times$ reduction in ATE, with comparable runtime and memory. Code is available at: https://github.com/Powertony102/ReCal3R.

HiPPO gives recurrent states memory semantics as coefficients of online polynomial projections, but in fixed channel coordinates. Modern selective SSMs, by contrast, rely on token-dependent control and channel interaction. We introduce SHiPPO (Sylvester HiPPO), a transported projection-memory prior that lifts HiPPO coefficient memories into a moving channel frame. For any fixed or realized right-transport path, SHiPPO transports the approximation family and channel metric together; conditional on that path, the state is ordinary HiPPO in a tied moving frame and follows Sylvester coefficient dynamics, preserving the left online-memory operator while adding right-action transport. For selective-SSM execution, we derive a restricted group-local realization with controller-compatible right actions, exponential-adjusted updates, exact block-affine scan, and recurrent decoding. We also give a simultaneous-reducibility criterion identifying when right transports collapse to static mixing plus independent scalar or blockwise banks. Controlled diagnostics show that larger current-token write rank improves ordinary prediction error but cannot recover order-sensitive changes to already-written memory; transported-memory variants recover this signal, which disappears when the transport pathway is removed. A finite-field associative-recall diagnostic with interleaved bindings, operations, and queries provides complementary autoregressive evidence while leaving the preferred right-action realization open. Taken together, these results support SHiPPO as a mechanistically grounded transported-memory prior, with evidence focused on memory mechanisms rather than broad sequence-modeling dominance.

Orthogonalizing the mLSTM memory matrix at read time with five differentiable Newton-Schulz iterations improves noisy associative recall. We replicate this effect and investigate its mechanism. Training on MAD noisy recall exhibits a long chance-level plateau followed by a sharp increase in accuracy. The orthogonalized read improves conditioning during this plateau and can be removed after escape. Ablations support three findings. First, the benefit requires a self-consistent read and gradient: an exact recursive least-squares read (the Mesa layer) yields a similar benefit, while straight-through variants, delta-rule writes, frozen random keys, and Frobenius normalization show no improvement over baseline. Second, across a learning-rate x task-difficulty grid, orthogonalization multiplies escape hazard roughly six-fold, with no detectable dependence on difficulty, and widens the range of learning rates that produce successful runs. Third, adding orthogonalization at inference leaves chance-level failures unresolved, while removing it gradually after escape yields standard mLSTMs at near-perfect accuracy. Schedule changes alone recover much of the reported gain. A batch-size x learning-rate analysis separates the effects of per-step learning rate and gradient noise on escape hazard (elasticities +3.0 and -1.65, respectively), linking the original vocab-96 result to its large-batch training regime. Direct decoding of the memory state recovers roughly half of the associations in behaviorally failed models, indicating a readout-learning limitation despite substantial stored information. These results show that fixed-budget recall benchmarks are sensitive to trainability and provide a tractable setting for investigating abrupt behavioral transitions through measurements of internal representations.

Quantum Fast-Weight Programmers (QFWPs) store temporal information in dynamically programmed variational-circuit parameters rather than in nonlinear recurrent hidden states, offering a practical route to quantum sequence modeling. Self-Modulating QFWP improves this framework by using input-dependent gates for both new fast-weight updates and the accumulated fast-weight state, but its unbounded old-state multiplier can diverge in long-sequence regimes. We propose a bounded old-state modulation rule that applies a sign-preserving tanh gate only to the recurrent memory branch while leaving the additive update and new-update modulation unchanged. We evaluate standard QFWP, full Self-Modulating QFWP, Only-New, and Only-Old variants on two CUDA-Q quantum-dynamics forecasting tasks and on Milan SMS telecommunication activity prediction. The quantum-dynamics results show that old-state modulation is the most consistent source of improvement over Standard QFWP, and that bounding the old-state gate removes long-sequence divergence while improving aggregate robustness. On Milan SMS forecasting, the original unbounded Self-Modulating QFWP converges across the tested grid and shows its clearest gains at longer input windows, with behavior close to the Only-Old ablation. These findings identify accumulated-memory modulation as the key mechanism of Self-Modulating QFWP and bounded old-state gating as a targeted stabilization strategy.

Linear-attention and state-space language models compress the prefix into a fixed-size recurrent state, yielding O(1) memory at the cost of a lossy exact memory: when many key--value associations compete, earlier facts are overwritten and needle recall degrades. Inspired by Complementary Learning Systems, we give linear attention a hippocampal complement. HOLA (Hippocampal Linear Attention) keeps the usual delta-rule state as a compressive memory and adds a bounded exact KV cache, forming a semiparametric test-time memory: the state models linearly compressible structure, while the cache stores associations that should not be forced through that state. The cache writes without a learned eviction module, keeping tokens with large beta * ||e||, the prediction residual actually committed to the state; a decoupled RMSNorm-gamma cache read then turns these exact KV pairs into sharp retrieval rather than soft averaging. At 340M parameters trained on 15B SlimPajama tokens, HOLA lowers Wikitext perplexity from 27.32 to 22.92 (-16.1%), below a full-attention Transformer++ (26.88), and improves LAMBADA perplexity from 30.95 to 30.26. It also achieves the best linear in-context retrieval and remains much more robust than GDN or a matched HOLA+recency cache on RULER needle-in-a-haystack recall out to 32k tokens (16x its training length).

Recurrent memory agents extend LLMs to arbitrarily long contexts by iteratively consolidating input into a fixed-size memory window. Despite their scalability, these agents exhibit a well-documented reliability problem: end-to-end performance degrades systematically as context length grows. We diagnose this failure by decomposing performance into two factors--memory capture and memory retention--and quantitatively confirm that retention is the dominant bottleneck. Retention collapses because existing designs maintain memory as a monolithic text block, forcing every update to risk overwriting previously retained content. Motivated by this diagnosis, we propose Multi-Head Recurrent Memory (MHM), a general, training-free framework that partitions memory into independent heads governed by a stage-wise select-then-update strategy. At each step, exactly one head is selected for update while the remaining heads are structurally shielded from overwriting, shifting the burden of retention from model behavior to architectural design. As a lightweight instantiation, we introduce Least-Recently-Updated MHM (MHM-LRU), which guarantees uniform head utilization with zero additional token overhead. Extensive experiments on long-context benchmarks show that MHM-LRU substantially improves both retention and end-to-end accuracy across the 100K--1M token range, where baselines degrade sharply. On RULER-HQA at 896K tokens, MHM-LRU improves the memory retention rate from less than 30% to 73.96%. These gains generalize across model families, scales, and task types, positioning architectural optimization as a practical and cost-efficient path toward reliable long-context recurrent memory.

We introduce TiRex-2, a recurrent xLSTM-based time series foundation model that generalizes the univariate TiRex to multivariate forecasting with both past and future covariates. Real-world forecasting is inherently sequential: observations arrive continuously, variables evolve jointly, and a subset of covariates is known ahead of time. Existing Transformer-based time series foundation models capture cross-variate dependencies but incur quadratic complexity in context length and require full-history recomputation as new observations arrive. TiRex-2 addresses these limitations through a memory-centric recurrent design that operates at constant per-patch cost under streaming. The model combines a bidirectional time mixer with an asymmetric grouped-attention variate mixer, enabling the integration of future-known covariates while preserving strict causality over target variables. To our knowledge, this is the first time series foundation model that achieves this combination of properties. To support scalable multivariate pretraining, we propose a synthetic coupling pipeline that composes diverse multivariate samples on the fly from large univariate corpora. Empirically, TiRex-2 achieves state-of-the-art zero-shot performance on GIFT-Eval and fev-bench, remains stable when streamed to arbitrary context lengths, and maintains constant inference cost per patch. The model uses 38.4M active parameters in univariate mode, with an additional 44.1M parameters activated for multivariate forecasting.

Long-horizon multivariate time series forecasting (LTSF) remains challenging due to non-stationarity, regime shifts, and error accumulation. The Variability-Aware Recursive Neural Network (VARNN) is designed to track such variability by maintaining a residual-memory state driven by one-step prediction errors. However, its original formulation is limited to one-step sequence regression and does not directly support multi-step forecasting. In this work, we extend VARNN to long-horizon forecasting and introduce StateFlow, a recurrent forecasting framework that uses VARNN as a dual-state recurrent backbone to capture two complementary signals from the lookback sequence: a hidden-state trajectory representing primary temporal dynamics, including trend, seasonality, level changes, and recurring patterns, and a residual-memory trajectory representing structured local prediction deviations, driven from a nonlinear recurrent transformation of errors between one-step base predictions and observed values. A chunk-based decoder separately summarizes these trajectories and maps them to the future horizon for direct multi-step forecasting. We further employ a two-stage optimization strategy that first trains the VARNN encoder through a one-step base prediction objective to optimize the internal representations over the lookback sequence, and then trains a horizon-specific decoder for direct multi-step forecasting. Experiments on standard LTSF benchmarks show that StateFlow achieves competitive performance against strong linear, recurrent, convolutional, and Transformer-based baselines while preserving linear recurrent encoding and a compact model design.

Alternating recurrent events -- event-times of a specific nature that trigger a secondary refractory period -- occur in a wide-range of fields, including behavioral science, criminal justice, and biostatistics. Analysis of these events requires careful attention to the statistical nuance, including correlated observations and repeated outcomes subject to potential censoring. We develop an online dynamic prediction framework appropriate for predicting subsequent alternating recurrent events, by developing neural network theory for a statistical audiences and applying inverse probability weighted pseudo-observations. The proposed model is applied to dynamically predict alternating recurrent event-free time, showing good performance in simulation, and outstanding capability in application to predicting periods of low mood for first-year medical residents. We close with a discussion.

Long-range learning is hard for recurrent networks trained with stochastic gradient descent, because the influence of a past input fades with the lag $\ell$, and if it fades too fast the dependence cannot be learned from finite data. This fade is captured by an envelope $f(\ell)$. An exponential fade makes the data needed to learn a lag-$\ell$ dependence grow exponentially, putting long horizons out of reach; a power-law fade keeps the cost polynomial. We show that the asymptotic decay class of $f(\ell)$ is not fixed by the architecture. Instead, it emerges from the coupling between the state dynamics and parameter dynamics, settling into either a collapsed regime (fast, exponential forgetting) or an extended, anti-collapsed regime (slow, power-law forgetting). The intuition is a competition within these coupled dynamics. Training drives the network's effective time scales toward short ones, while rare, heavy-tailed fluctuations of the learning dynamics push a few of them to very long values. The extended regime survives only when these heavy-tailed pushes are strong enough to balance the pull. We make this mathematically precise with a coarse-grained stochastic process and prove exactly when the extended regime exists. A single exponent, the spectral exponent~$β$, then governs both the spread of time scales and how slowly the network forgets. Realizing the regime in practice needs one more ingredient: the joint action of the architecture and the optimizer must be able to hold such a broad spread. A network whose capacity to generate broad time-scale spectra is severely constrained still collapses, even when supplied with strong heavy-tailed forcing. Heavy-tailed fluctuations thus act not as noise to be suppressed, but as the mechanism that sustains long-range learning.

Adaptive Financial Transformer (AFT) is proposed for stock return prediction under non-stationary financial markets. The model incorporates a Market Regime Encoder, an Adaptive Gate Network, and an Adaptive Financial Context module to dynamically bias self-attention based on semantic relationships between financial indicators. Unlike conventional Transformer architectures that treat all input features uniformly, the proposed approach groups 95 engineered financial features into 11 semantic categories and adapts attention according to latent market regimes. The study also identifies and corrects sequence alignment and backtesting issues that can inflate reported trading performance, and introduces a financially-aware composite objective that jointly optimizes prediction error, directional accuracy, and non-overlapping Sharpe ratio. Extensive experiments compare the proposed architecture against classical machine learning models, recurrent neural networks, and Transformer baselines using chronological evaluation, five random seeds, ablation studies, hyperparameter optimization, explainability analysis, and multi-stock validation. Results demonstrate competitive predictive performance while reducing model complexity by 15.2% and improving parameter efficiency through feature selection, providing an interpretable Transformer architecture for financial time-series forecasting.

We introduce the context-ready transformer, a new recurrent neural network architecture built from a D-layer transformer block that pre-contextualizes each token before it enters the block. During left-to-right generation, a correction network combines the previous position's block output -- a cached summary of past context -- with the current token embedding, so the tokenenters the block already contextualized rather than as a raw embedding. At sequential inference, the correction chain makes the architecture a recurrent neural network. For training, we unroll the correction process K times over the full sequence, processing all positions in parallel at each step. A pretrained transformer can also be converted to a context-ready model by adding a zero-initialized correction FFN and fine-tuning. We evaluate across widths, depths, block sizes, and two datasets, with all comparisons against standard transformers, variants, and ablations. A D=5 model beats a 12-layer transformer while generating 1.7x faster on an A100. With K=10, a single-layermodel (D=1) beats a 6-layer transformer with a 2.6x inference speedup, and sequential inference matches parallel K=10 to within 0.01 PPL. The architecture benefits most from wide representations and long contexts. On a pointer-chasing task, D=1 trained with BPTT solves all 10 composition levels, while standard transformers exhibit staircase-like depth dependence.

Recurrent delta-rule models keep a fixed-size state matrix S (d_v x d_k) that compresses all past context. The state of the art (GDN-2) gates this update with element-wise matrix erase/write masks. This is powerful but has two defects. First, both gates are computed from the incoming token alone, making the model memory-blind: it decides what to erase without seeing what it has stored. Second, value-axis coupling in the erase gate blocks the WY-form triangular chunk solver that drives efficient training -- the intra-chunk system splits into d_v independent solves, collapsing throughput to serial-recurrence cost. We introduce CARVE (Content-Aware Recurrent with Value Efficiency), which fixes both and, via a single-launch "megakernel" scheduling of the same WY-form math, trains faster than the matrix-gated baseline it replaces. The key idea is architectural: restricting all gating to the key axis makes the intra-chunk coupling independent of the value index, restoring one unmodified WY-form solve. Within this constraint, CARVE conditions both gates on a content signal read once per chunk from the chunk-boundary state and folded algebraically into each gate's low-rank projection (by associativity, U(Sq)=(US)q), giving memory-aware gating at negligible extra traffic. At init the content projections are zero, so CARVE is bit-identical to the baseline; we prove the one-chunk staleness perturbs gates by only O(1/sqrt(L)), matching a measured 0.18% deviation flat up to L=128. At 1.3B parameters / 100B FineWeb-Edu tokens on H100 (three seeds), CARVE improves every axis: WikiText perplexity 15.72 vs 15.90 (hybrid 15.41 vs 15.62), +0.63 pp average common-sense accuracy, and state-of-the-art RULER and real-world recall -- while training +1.4% faster at matched depth and +19.3% at iso-quality depth, at +13% peak memory. Backed by six formal guarantees.

Implicit neural representation (INR) has emerged as a powerful prior for multi-dimensional data (e.g., multispectral images and videos). However, most INR methods employing periodic activation functions (e.g., Sine) predominantly rely on function composition. This mechanism introduces optimization instability as network depth increases, thereby limiting their performance. Meanwhile, these methods fail to incorporate proper physical priors to effectively alleviate spectrum bias. To address these issues, inspired by the commonalities between deep periodic networks and generalized Fourier series, we propose a novel Calibrated Harmonic Overlaid Implicit Neural Representation (CHOIR). Specifically, we utilize Coordinated Harmonic Superposition (CHS) to replace the conventional function composition used in most INRs, thereby ensuring optimization stability when scaling network depth. Furthermore, we introduce a Perceptual Spectrum Calibration (PSC) to mitigate spectrum bias. This calibration embeds the ubiquitous power-law spectrum prior of natural images and adjusts the globally fixed spectrum towards a physically plausible log-uniform distribution. Extensive experiments on various multidimensional data recovery problems demonstrate that our method achieves superior performance over state-of-the-art approaches. Code is available at https://github.com/chorl0229/CHOIR.

Delta-rule linear attention improves recurrent memory updates by correcting what is already stored at the current write address before writing new content. However, the active correction is still anchored to that same write address. As a result, stale information stored at a different address cannot be actively removed before new content is written elsewhere. We propose Erase-then-Delta Attention (EDA), a memory update rule that decouples where to erase from where to write. The key insight is that recurrent memory models should not only correct the current write, but also selectively suppress outdated memory at an independently chosen address. Concretely, our method first applies a targeted erase step along a learned erase direction, and then performs the standard delta-style corrective write along the current write direction. This preserves the corrective behavior of delta-rule updates while expanding their memory-management capacity. Language-model pretraining experiments across dense 2.5B and MoE 25B-A2.8B model families show that EDA performs best in both settings. The gain persists after 80B-token long-context midtraining of the MoE models, where EDA also performs best in long-context evaluations from 4k to 128k contexts. A compact update analysis and memory-state probes suggest why: EDA keeps the delta-rule corrective write intact while allocating an additional cleanup path most strongly when passive decay is weak. These results suggest that recurrent memory models should decide not only what to write, but also what stale information to erase and where.

While parameter-efficient fine-tuning (PEFT) typically targets attention projectors, its efficacy for tasks requiring sequential state accumulation remains under-explored. We examine if PEFT for such tasks can benefit from state space model (SSMs) adapters, and if MLP blocks are better injection sites. We introduce Hankel Reduced order Model (HRM) adapter, an SSM-based residual module initialized via Balanced Truncation of empirical Hankel Grammians. By leveraging the time-invariance of the system matrix $\bar{A}$, HRM enables an exact FFT-based parallel scan, achieving computational parity with LoRA across all context lengths. In iso-parametric evaluations on Mistral-7B (8.4M trainable parameters), HRM outperforms LoRA variants on LongBench tasks, including QuALITY (+34.8% relative accuracy) and QMSum (+71.6% relative ROUGE-1). HRM further demonstrates consistent superiority across 18 configurations of synthetic state-tracking (DFA, Parity) and character-level language modeling (enwik8). Gate analysis reveals that HRM adapters effectively learn to modulate recurrence, providing a robust architectural alternative to low-rank adaptation for long-context sequence modeling.

In the first part of this work, we construct the first type of two-dimensional (2D) hyperbolic neural quantum state (NQS) in the form of the Lorentz 2DRNN (Recurrent Neural Network) and benchmark its performance against the Euclidean 2DRNN in the paradigmatic $N\times N$ 2D Transverse Field Ising Model (2DTFIM) setting with different lattice sizes up to $N=12$ and at different transverse magnetic field strengths. We find that hyperbolic Lorentz 2DRNN NQS definitively outperform Euclidean 2DRNN NQS when the system is at the phase transition point when the physics can be described by a conformal field theory (CFT), which is known to be dual to an Anti-de-Sitter (AdS) space whose spatial geometry is hyperbolic. In the second part of this work, we benchmark the performances of the recently introduced one-dimensional (1D) hyperbolic NQS including Poincaré RNN/GRU and Lorentz RNN/GRU against their Euclidean NQS versions in $N\times N$ 2DTFIM, which has to be converted to a one-dimensional setting to allow for the use of 1D NQS. The findings in this case extend our previous results that 1D hyperbolic NQS definitively outperform 1D Euclidean NQS, thanks to the combined effects of the hierarchical structure comprising the first and $N^{th}$ neighbor interactions present in the 1D system arising from the 2D lattice and the CFT physics at the critical point. While more studies with larger system sizes are required, our work serves as a proof-of-concept for the utility, effectiveness as well as the superior performances of one- and two-dimensional hyperbolic NQS ansatzes compared to the existing Euclidean NQS in many-body quantum physics systems, especially when these systems exhibit structural hierarchy or when they are at criticality, or a combination of both.

Understanding how the architecture of neural networks shapes the computations they carry is a central challenge in neuroscience and machine learning. While specific circuit architectures have been linked to particular network computations and theoretical bounds on expressivity of broad classes of networks have been found, we are still missing general principles connecting the structure of finite networks to their computational capabilities. Here, we characterize the computational abilities of recurrent neural networks as a function of their connectivity by training a large collection of different networks to compute a large set of Boolean functions. For small networks, we constructed the complete ``catalogs'' of network-function performance, which revealed that computational capacity varies widely across architectures and that most networks show poor performance, and most functions are hard to compute. However, we show that having local 2- and 3-cycles in a network strongly enhances its computational ability, and networks with such cycles are often the minimal architectures that can solve particular functions. We further show that a small set of structural statistics accurately predict networks' performance. Extending our analysis to large networks showed that typical networks fail even to approximate a randomly selected function. Surprisingly, adding a small number of sparsely connected biologically-inspired interneurons to the network dramatically increases computational capacity. As in small networks, adding short cycles improved networks' capacity, outperforming acyclic or reachability-matched controls. Thus, our results identify local cycles as design principles linking neural connectivity to computational power, and offer a general framework to explore structure-function relations in computing networks.

Recent advances in quantum computing and machine learning have motivated the development of quantum models for sequential data processing. In this paper, we propose a Recursive Quantum Long Short-Term Memory model, or Recursive QLSTM, which extends QLSTM through metacore-based recursive constructions. We numerically test the model under different input sequence lengths, metacore designs, and recursive rules, and identify the best-performing architecture among these variants. For this selected model, we further provide theoretical arguments explaining why its recursive structure improves temporal information propagation and enhances learning performance. Our results suggest that Recursive QLSTM offers a flexible and effective framework for quantum recurrent learning over input time series of various lengths.

Biological systems maintain function in fluctuating environments by transforming past stimulation into internal dynamical states that support future-oriented responses. Reservoir computing provides a computational analogue, but standard formulations often treat the recurrent substrate as a fixed random network and train only the readout. Here we ask how the substrate itself changes when reservoir architecture is placed under evolutionary selection for prediction. Using the Kuramoto--Sivashinsky equation as a testbed for spatiotemporal chaos, we evolved reservoirs over five construction hyperparameters: size, connectivity degree, spectral radius, input scaling, and readout regularization. Evolution reduced prediction error at the population level, extended the low-error forecast horizon, and organized the design space along a diminishing-return size--efficiency frontier. Structural analyses showed that evolved reservoirs remained within a conserved stochastic-block-model-like spectral envelope while refining low-eigenvalue modes, locking modularity to an intermediate band, and pruning connection cost within that band. Pareto analysis showed that elite reservoirs occupied a horizontal floor in the cost--modularity plane, indicating that accuracy and efficiency were achieved jointly rather than through a simple trade-off. These findings show that evolutionary optimization does not merely improve prediction, but exposes interpretable structural constraints on the recurrent substrate: it stabilizes a task-suitable dynamical class and refines the architectural degrees of freedom most relevant for prediction. Evolutionary reservoir computing therefore provides a bio-inspired framework for studying how predictive demands shape adaptive dynamical networks.

While deep learning models achieve state-of-the-art performance in complex tasks, they remain brittle when faced with new environments or sensory deprivation. In contrast, biological systems exhibit remarkable tolerance to these challenges. We address this vulnerability by developing a recurrent neural network (RNN) whose architecture is directly derived from the synaptic-resolution brain connectome of the fruit fly Drosophila melanogaster. We demonstrate the feasibility of training the fly connectome neural network (FLYNN) to perform vision-based navigation in MuJoCo, achieving performance comparable to modern hand-crafted networks of similar parameter counts. Crucially, FLYNN exhibits superior resistance to out-of-distribution (OOD) data and tolerance to sensory loss without further training. It remained functional even under total vision loss while hand-crafted networks largely failed, even when specifically trained with camera dropout. Principal Component Analysis (PCA) of the internal state of FLYNN suggests that it exhibits a particularly high degree of representational modularity, which might be related to its robustness. Our work provides a new direction for designing resilient artificial agents following the topology of biological brains.

The intricate structures of biological neural networks largely emerge during development, guided by a comparatively compressed blueprint encoded in the genome. The connectivity that emerges from this decoding process is rich in structure, and already equips the organism with functional modules upon birth. This initial structure serves as a scaffold that can be gradually refined and fine-tuned through lifelong experience, via a variety of plasticity mechanisms. Drawing inspiration from this interaction between evolutionary and developmental modes of learning, we use hypernetworks to learn a compressed generative process that generates the connectivity of a modular reservoir. We show that this marriage between curriculum-based meta-learning and modular reservoir computing can generate sparse recurrent networks that solve difficult temporal tasks with minimal training and without concessions to robustness.

Transformers are AI's workhorse with strong performance in modeling sequential data, but their computational cost becomes prohibitive when processing long sequences. We target long-horizon streaming vision and robotics applications like map-free pose estimation, where it is particularly impractical to store and maintain a history of observations. Recurrent Transformers address this limitation by maintaining fixed-size memory but their performance lags behind that of transformers operating over the full observation history. We argue that this gap does not stem from architectural limitations, but from differences in how these models learn to compress past information. Without access to an observation history, recurrent models must explicitly decide what to retain in memory at each step, a significantly harder learning problem. In this work, we propose a distillation approach that transfers the compression strategy of a classical full-history transformer to a recurrent variant. We enable this by designing a teacher model that explicitly compresses its observation history into a fixed-size bottleneck representation. By directly supervising the student's memory with this bottleneck representation, we align the two compression mechanisms. We show that this approach allows to train a recurrent latent robotic memory with linear-time complexity while substantially narrowing the performance gap to full-history transformers.

Mainstream LLM serving systems reuse prefix work mainly through paged or radix key-value (KV) caches. This is highly effective for high-throughput, high-concurrency serving, but it manages only one positional fragment of execution state: the KV cache. We study the opposite regime: low-latency, small-batch, on-device physical-AI serving, where interactive LLM agents, speech systems, and robot policies repeatedly branch, reset, interrupt, and re-enter under tight responsiveness budgets. We introduce execution-state capsules, a graph-bound checkpoint and restore mechanism for the complete restorable state at a committed boundary. FlashRT is a white-box, backend-facing kernel runtime whose evaluated NVIDIA CUDA backend runs captured graph plans over contiguous static buffers with no block-table indirection. Because the live state is a closed set of named buffers, a capsule can snapshot, restore, fork, or roll back the whole execution boundary, including KV, recurrent state, convolution state, MTP state, and metadata. This moves reuse from token-addressed KV fragments to graph-bound execution-state boundaries. On an RTX 5090, capsule restore is byte-exact at the stored-state level and token-identical under greedy decode. A KV-only ablation diverges, showing that recurrent state is load-bearing. GPU-resident snapshot and restore are sub-millisecond, and TTFT speedup over cold prefill grows from 3.9x at 2k tokens to 27x at 16k tokens. On Jetson AGX Thor and DGX Spark, the same correctness and structural properties hold. Capsules are not a replacement for high-throughput KV-cache serving; they define a complementary latency-first serving point for explicit execution-state reuse.

Classical approximation theorems ask for a new neural network whenever the target accuracy is improved. This paper studies the opposite possibility: can the network be chosen once and for all, and can accuracy be bought only by letting it run longer? We prove that this is possible for every continuous function on [-1,1]. More precisely, each such function is uniformly approximated by the time evolution of a single ReLU recurrent neural network with fixed weights and fixed hidden dimension. The mechanism behind the construction is a new intermediate model, the Turing machine with neural units (TMNU). This model retains the algorithmic freedom needed to implement polynomial approximation schemes, while remaining rigid enough to be simulated by RNNs with explicit bounds on hidden dimension and weight magnitude. The resulting convergence rates reflect the underlying polynomial approximation rates. We complement the construction with minimax lower bounds showing that runtime is not merely a proof artifact, but an unavoidable resource in this fixed-network approximation paradigm.

Convolutional networks, recurrent networks, and transformers each encode different inductive biases -- locality, sequential memory, and content-dependent pairwise interaction -- and have remained mathematically distinct since their inception. We show that this fragmentation reflects not a fundamental diversity in how signals should be processed, but rather incomplete views of a single underlying mathematical object: a learnable integral transform. We introduce the Integral Transform Network (ITNet), a unified architecture built around a learnable kernel that depends jointly on positions and features. This kernel is implemented as a small neural network, specifically an MLP, that models pairwise interactions, enabling the model to adapt its behavior from data. We show that convolution, self-attention (including multi-head), and autoregressive recurrence (including LSTM, GRU, S4, and Mamba) arise as special cases under appropriate parameterizations, and that ITNet is a universal approximator of continuous operators. To make this practical, we develop tiled kernel fusion, importance-weighted Monte Carlo integration, and learned low-rank factorization, enabling efficient and scalable computation. A single ITNet architecture with a shared operator and lightweight modality-specific encoders matches or exceeds specialized baselines on ImageNet-1K , GLUE, ModelNet40, VQA,v2 and NLVR2. The results demonstrate that a single learned interaction mechanism can recover the behavior of all three architectural families from data.

Recurrent networks can contain substantial functional redundancy in weight space: changing a recurrent matrix may leave the input-output rollout nearly unchanged on a task distribution, while similar-scale changes can destroy the same behavior. We study this redundancy in one-layer tanh RNNs using ordered real Schur coordinates. The Schur form separates spectral blocks from directed nonnormal couplings, giving a diagnostic basis for structured ablations that keep the input and readout maps fixed. In a fixed-length copy task, selected nonnormal Schur couplings can be removed with little loss in some trained solutions, whereas other couplings are necessary for accurate autonomous replay. Across flip-flop, sine generation, and context-dependent integration, the loss-preserving ablation profile varies across tasks and trained solutions. These results identify candidate approximate functional invariances, not universal symmetries of recurrent weight space. Schur-coordinate ablations provide a practical diagnostic for which structured perturbations preserve a trained recurrent solution and which ones disrupt its computation.

SWave is a complex-valued recurrent language model (169.26M parameters, D=384, L=16, T=2048) trained on FineWeb-Edu using 2xH100 NVL. It was designed around three founding premises: that representing language as complex waves rather than real-valued numbers enables richer information encoding; that a Cayley-parameterised unitary transition provides a mathematical guarantee against state decay or explosion; and that a hidden state which rotates rather than shrinks preserves signal integrity over arbitrarily long contexts. The core of SWave evolved substantially across three development phases. The Resonance Head was found to structurally admit imaginary-channel collapse as a global loss minimum (a failure mode we term cos-domination collapse) and was superseded by an untied head with independent real and imaginary embedding tables from the Phase-Associative Memory (PAM) architecture. This resolved the degenerate minimum and enabled stable 200,000-step training (best-step PPL 22.0 at step 89,861). ComplexNorm and the Wave Propagation Scan proved load-bearing throughout all three phases and were retained to the final architecture. ProtectGatedScan was reframed as a structural prior rather than a learned behaviour. The four multi-scale retention concepts showed no measurable improvement under controlled evaluation and were found non-load-bearing. The ComplexGatedUnit was superseded by a real-valued squared-ReLU channel mixer with fewer parameters. The auxiliary training objectives showed no benefit once structural constraints were resolved. The investigation yields a formal characterisation of cos-domination collapse, a parallel scan with a log-space backward pass for numerical stability, six transferable engineering principles for complex-valued recurrent training, and a plan-to-code traceability methodology for catching structural divergences that conventional test suites miss.

Compositional learning systems must balance plasticity, the ability to acquire new knowledge, with stability, the preservation of previously learned components, especially when tasks share structure and risk interference. We study how modular architecture, task similarity, and representational dimensionality jointly shape compositional continual learning in a sequential A-B-A paradigm, comparing a task-partitioned recurrent network to a single-network baseline while inducing high- and low-dimensional regimes via weight-scale manipulations. In a high-dimensional "lazy" regime, both architectures achieve similar performance and internal geometry, suggesting that explicit modular structure has little impact when representations are weakly constrained. In a lower-dimensional "rich" regime, modularity becomes decisive: the modular network develops graded task-specific subspaces that overlap for similar tasks, partially align for moderately dissimilar tasks, and separate for dissimilar tasks, yielding a more compositional and interpretable organization than the single network. These findings identify the representational regime induced by initialization scale, which co-varies with representational dimensionality, as a key factor governing when compositional, modular structure is functionally beneficial in continual learning, and support viewing safety and robustness as problems of adaptive allocation of representational subspaces rather than fixed separation versus sharing.

Disruption recovery in industrial assembly lines requires timely decisions under machine faults, worker absence, and emergency orders. Existing methods either rely on rigid handcrafted recovery logic or learn adaptive policies that do not readily exploit heterogeneous external recovery knowledge at decision time to reduce abnormal recovery time (ART) and preserve on-time delivery (OTD). To address this gap, we propose a phase-aware guidance injection framework that augments a trained recurrent MAPPO (RMAPPO) scheduling policy through logit-level action bias during evaluation. The framework provides a unified decision-time interface for rule-based, replay-based, and online LLM-based guidance, while activating intervention only during abnormal and recovery phases. Experiments on a custom AssemblyLineEnv show that high-quality rule guidance yields the strongest gains, replay-based guidance degrades smoothly under imperfect availability, and online LLM guidance still provides useful intermediate improvements. These results show that decision-time guidance injection can exploit heterogeneous recovery hints without redesigning the actor.

How the wiring and functional organization of cortex shape recurrent computation remains a central question in both neuroscience and machine learning. Here, we leverage data released through the Machine Intelligence from Cortical Networks (MICrONS) program--a functional connectomics resource spanning multiple areas of mouse visual cortex, in which dense calcium imaging is co-registered with high-resolution electron microscopy reconstruction from the same animal--to build biologically grounded recurrent neural networks. Using neuronal spatial coordinates, anatomical connectivity, and function-derived relationships from nearly 12,000 coregistered excitatory neurons, we initialize recurrent weights and impose communication-aware spatial constraints during learning. Across three cognitive decision-making tasks, networks constrained by cortical structure and function consistently outperform baseline and partially constrained models. Functional weight initialization provides the largest gain, while real spatial embedding yields robust additional improvements across conditions. These biologically grounded networks also develop low-entropy, modular, and small-world organization, and retain strong performance even when recurrence is restricted to positive weights. Together, our results show that the machinery of cortex--its geometry, wiring, and functional structure--can be harnessed as a powerful inductive basis for building recurrent networks that learn more effectively while converging toward key organizational principles of biological computation.

This study investigates the application of machine learning models to predict exam outcomes using physiological data collected during examination sessions. Physiological stress indicators, including electrodermal activity, heart rate, and skin temperature, were analyzed to uncover their association with academic performance. A variety of machine learning approaches were employed, ranging from standard models like logistic regression, random forest, and support vector machines to more advanced architectures, including transformers, long short-term memory (LSTM), and gated recurrent unit (GRU) models. This diversity aimed to capture the complex interactions within the data effectively. A key focus was assessing the adaptability of transformers in processing numerical data and evaluating their performance in this novel context. Standard performance metrics, such as accuracy, precision, recall, and F1-score, were used to compare model efficacy. The experimental results demonstrate that while deep learning models generally excel at capturing complex relationships in physiological data, simpler models like random forests can sometimes achieve superior performance while offering computational efficiency and interpretability. Furthermore, transformers demonstrated notable versatility, showcasing performances comparable to those of the LSTM and GRU models. This research underscores the importance of experimenting with a broad class of models that align with the objectives of the problem at hand, balancing precision, efficiency, and interpretability. By elucidating the relationships between physiological signals and academic performance, this study contributes to understanding stressors affecting students' mental health. It further promotes leveraging physiological data to enhance student well-being and academic outcomes.

Linear attention replaces softmax attention's growing KV cache with a fixed recurrent state, but this compression limits exact state tracking and long-context memory. We introduce \emph{Semidirect Fourier Delta Attention} (SFDA), a phase-controlled generalization of Kimi Delta Attention that replaces real diagonal decay with block-rotational Fourier control: $$S_t=(I-β_t k_tk_t^*)Λ_tS_{t-1}+β_tk_tv_t^*, \qquad Λ_t=\diag(α_t\odot e^{iθ_t}).$$ Our main result is a constructive chunk-WY factorization for products $A_t=Λ_t-u_tr_t^*$, giving $$A_t\cdots A_1=Γ_t-Y_tM_tW_t^*$$ with rank growth bounded inside fixed chunks. This yields an exact affine chunk transfer, formal stability and complexity bounds, and a compact characterization of phase-plus-low-rank memory. We verify the algebra numerically and show in toy state-tracking experiments that SFDA learns cyclic memory where the phase-disabled KDA baseline remains near chance. Fused kernels and large-scale language-model comparisons are left to future work.

Spiking Neural Networks (SNNs) are well-regarded for their biological plausibility and energy efficiency in processing sequential data. However, dominant SNN architectures typically rely on first-order Ordinary Differential Equations (ODEs) to govern neuronal state transitions. This first-order assumption imposes a "memoryless" bottleneck, limiting the model's capacity to capture the complex, long-range dependencies inherent in long-sequence tasks. In this work, we propose LongSpike, a novel SNN framework that integrates fractional-order State-Space Modeling, or f-SSM, from control theory into the spiking domain. By extending traditional integer-order SSMs to the fractional-calculus regime, LongSpike enables the hierarchical integration of neuronal dynamics with long-memory kernels. To mitigate the computational overhead and parallelization challenges typically associated with fractional operators, we leverage a state-space formulation that supports efficient, parallel training. Empirical evaluations on challenging benchmarks, including Long Range Arena (LRA), large-scale WikiText-103, and Speech Commands, demonstrate that LongSpike outperforms state-of-the-art SNNs in accuracy while preserving sparse synaptic computation. The code is available at https://github.com/xinruihe389-commits/LongSpike.

Vision-language-action (VLA) models predict chunks of future actions from the current observation, an assumption that fails under partial observability, where decisions depend on information no longer visible. Existing memory-augmented VLAs simultaneously introduce recurrence, retrieval, compression modules, auxiliary objectives, hierarchical memory, or task-specific architectural changes, so the contribution of recurrence itself remains entangled with surrounding machinery. We present a controlled isolation study of recurrence in a strong pretrained VLA backbone. Our formulation augments the transformer with a small set of learnable memory tokens carried across timesteps and updated through self-attention, trained end to end with truncated backpropagation through time, with no auxiliary losses and no architectural changes. We instantiate this as $μ$VLA, a family of OpenVLA-OFT variants parameterized by memory width m, TBPTT length K, and the memory update rule (cross-step gradients or a detached EMA), so that recurrence is the only varying factor. On MIKASA-Robo, $μ$VLA improves average success rate on five training tasks from 0.42 to 0.84 at the strongest setting and reaches 0.23 on held-out tasks with the same memory structure versus 0.07 for the memoryless baseline. On tasks requiring different memory structure, performance remains near baseline. On LIBERO, the strongest recurrent variant achieves 96.2% average success, indicating no regression under full observability. We interpret these results as a calibration of the capability envelope of minimal in-backbone recurrence, identifying the regime in which it is sufficient and the regime where additional memory structure is required. Demos and videos can be found in https://avanturist322.github.io/mu-vla/.

Certified local robustness verification for recurrent neural networks (RNNs) is challenging because approximation errors introduced by nonlinear relaxations can propagate through recurrent connections and accumulate over time. As a result, scalable linear bound propagation methods often become overly conservative and fail to certify inputs that are in fact robust, especially when many pre-activation intervals cross zero. We propose an abstraction-refinement framework for RNN verification that partitions such intervals to remove the dominant relaxation error: on each refined branch, ReLU becomes exact, and smooth activations such as tanh and sigmoid admit substantially tighter linear envelopes. To control the combinatorial cost of splitting in long sequences, we introduce a SHAP-guided timestep selection strategy that ranks hidden states by their contribution to the verification objective and refines only the most critical timesteps in temporal order. Experiments on CIFAR10 and MNIST stroke benchmarks demonstrate consistent improvements in verification success and robustness-margin tightness over abstraction-only baselines, while exposing clear runtime trade-offs between ReLU and tanh models.