Attention Mechanisms

Follow this topic Ask about this topic

Momentum

42 papers in the last four weeks, up 100% on the four weeks before. 0.4% of all new papers.

Jul 13Week of Sep 28

Latest papers 378

Sep 23, 2026cs.CV

TEEP-RCNN: Texture-Enhanced Edge-aware Perception for Steel Surface Defect Detection via Improved Convolutional Block Attention in Faster R-CNN

Steel surface defect detection is critical for automated industrial quality control but remains challenging due to subtle inter-class texture differences and pronounced class imbalance. We introduce TEEP-RCNN (Texture-Enhanced Edge-aware Perception Region-based CNN), a two-stage detector built on Faster R-CNN with a Feature Pyramid Network backbone and an improved Convolutional Block Attention Module (CBAM). Our CBAM adds dropout regularization in the channel attention MLP and batch normalization on the spatial attention branch, reducing co-adaptation and stabilizing gating logits. Training uses a differential learning rate protocol with cosine annealing warm-up, separating update rates for the pre-trained ResNet-101 backbone and the detection head. At inference, predictions are refined via Test-Time Augmentation fused with Weighted Box Fusion (WBF), improving localization stability on elongated and boundary-adjacent defects. On the NEU-DET benchmark across six defect categories, TEEP-RCNN achieves 73.3% mAP@50 and 37.9% mAP@50-95 in only 10 training epochs on a single GPU, competitive with YOLOv11m (76.2% mAP@50, 100 epochs) while outperforming it on the rolled-in-scale category under the COCO metric. Per-class analysis shows the spatial attention branch is most effective on elongated texture defects such as patches and scratches, while crazing remains an open challenge across both paradigms due to its distributed non-local texture structure.

Kirtan Rajesh

Independent Researcher

Object Detection Convolutional Neural Networks Attention Mechanisms Anomaly Detection

Sep 23, 2026cs.LG

A Systematic Benchmark of Explainable Methods for Temporal Attribution in Sequential Recommendation Systems

Sequential RecSys are central to modern personalization, exploiting user's historical interaction sequences to drive next-step decisions. Deep learning models, particularly CNN and Transformer-based architectures, have proven highly effective at capturing temporal dependencies in these histories. For transparency and trust, understanding which past interactions drive a given recommendation is increasingly important --- both for developers auditing model behavior and for users seeking a rationale. However, the non-linearities that give these models their predictive power also render them black boxes, making it difficult to attribute decisions to specific interactions. While gradient-based, perturbation-based, and attention-based explainability methods exist, a systematic benchmark of their faithfulness for sequential recommendation is missing. We address this gap by introducing a dual-model masking metric in which one model supplies per-timestep attribution scores and a separately trained, masking-robust probe measures the resulting change in predicted probability. Using this metric, we benchmark ten XAI methods across CNN, Transformer, SASRec, and BERT4Rec backbones on KuaiRand and MovieLens, complemented by analyses of temporal attribution patterns, item popularity confounding, and robustness to input corruption. Our key findings are: (1) gradient-based methods, particularly GradientSHAP and Integrated Gradients, yield the most faithful and robust attributions; (2) raw attention weights are unreliable, but gradient-weighted attention restores faithfulness on shorter sequences, with degradation on longer horizons as softmax attention probabilities converge toward uniform importance scores, diminishing the method's ability to identify informative interactions; and (3) temporal attribution patterns in faithful methods reflect genuine task structure rather than recency or popularity bias.

Akash Pandey, Kanisha Shah, Addrish Roy +5

Capital One, AI Foundations San Jose, CA, USA · Capital One, AI Foundations New York, NY, USA · Capital One, AI Foundations McLean, VA, USA +1

Gradient-Based Attribution Sequential Recommendation Temporal Credit Assignment Attention Mechanisms Explainability Evaluation Explainable Recommendation

Sep 22, 2026cs.NE

Rethinking Pairwise Token Interaction in Spiking Transformers

Spiking Transformers inherit token interaction mechanisms from conventional Transformers, yet their sparse binary representations fundamentally alter how token-to-token communication is established. In particular, spike-based query-key matching produces highly sparse and input-dependent interaction patterns, coupling information propagation to the instantaneous availability of matching spike events. This motivates a different interaction paradigm in which long-range communication does not rely solely on pairwise spike coincidence. We therefore propose Gated Spike Axial Propagation (GSAP), a spike-native token interaction mechanism that decouples information propagation from context selection. Instead of directly determining communication through query-key matching, GSAP first propagates spike-based context along the horizontal and vertical axes, allowing information to reach distant tokens through structured sequential propagation. A receiver-conditioned gate then determines how much of the propagated context is incorporated at each token, while a lightweight local pathway preserves fine-grained neighborhood information. In this way, GSAP reformulates token interaction as a propagate-then-select process, enabling structured long-range communication while retaining the sparse event-driven nature of spiking representations. Code is available at https://github.com/Fancyssc/GSAP.

Sicheng Shen, Dongcheng Zhao, Zhiyuan Li +5

Institute of Automatoin, CAS · Center for Excellence in Brain Science and Intelligence Technology, State Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology,Institute of Neuroscience, CAS · Zhongguancun Academy

Spiking Transformers Transformer Attention Spiking Neural Networks Attention Mechanisms

Sep 21, 2026cs.AI

Attention as a Routing Graph: Live Circuit Extraction from a Single Forward Pass

Finding circuits in language models usually means running many careful interventions. We try something simpler: treat attention as a routing map from one forward pass, keep a small set of routes that point toward the answer, and ask whether those routes actually matter. They often do. On induction and IOI (tasks where the "right" circuit is already known), ablating our extracted edges hurts the model much more than ablating a random set of the same size. We evaluate n=100 prompts per cell on GPT-2 Small, GPT-2 Medium, and Pythia-410M, with paired gap tests and bootstrap confidence intervals. The extract step costs one forward; a head-by-head patch sweep costs about two orders of magnitude more. We are not claiming a complete circuit atlas. We are claiming a cheap sketch that carries real causal signal on known tasks, with clear failure modes when it does not. Code and evaluation artifacts are at https://github.com/Aquinf03/live-circuit-routing.

Ash Manvi, Samreena Tajreen

Aquin Labs

Circuit Discovery Attention Mechanisms Circuits in Language Models Causal Interventions in Language Models

Sep 21, 2026cs.RO

TACIT: Tactile Contact Supervision for Spatial Attention in Dexterous Manipulation

Visuomotor policies trained from a few demonstrations may reproduce demonstrated trajectories without reliably following changes in object position. Existing approaches with explicit attention typically obtain spatial priors from human annotation or visual models. We introduce TACIT (tactile contact informs attention), which uses measured tactile contacts from teleoperated demonstrations to supervise spatial attention without additional point annotation. Gaussian targets over preceding camera point clouds supervise an attention head whose pooled output conditions a visuotactile diffusion policy. Targets are used only during training; tactile observations remain inputs at inference. In the primary real-robot benchmark, with ten demonstrations per task and five demonstrated placement regions, TACIT achieves 66.7% success on ball placement and 73.3% on peg insertion, compared with 10.0% and 20.0% for input-matched 3D visuotactile fusion and 20.0% and 43.3% for vision-only DP3. TACIT enters the 150 mm palm-to-object approach region within 12 seconds in all 30 trials per task; all remaining failures occur after arrival. Across three training seeds on real ball and simulated peg, TACIT outperforms input-matched fusion and an architecture-matched control without explicit attention supervision, supporting the contribution of supervision beyond branch capacity. Pre-contact and contact-time supervision show no consistent ordering. These results demonstrate that measured tactile contact provides effective spatial supervision for approach behavior from few demonstrations within the evaluated workspace.

Yanhou Lai, Fucai Zhu, Ruiqiang Wang +1

Graduate School of Information Sciences, Tohoku University, Sendai 980-8579, Japan

Contact-Rich Robotic Manipulation Robot Imitation Learning Attention Mechanisms Tactile Sensing for Robotic Manipulation

Sep 17, 2026cs.CL

On-Demand Attention: Language Models Know When to Recall

Reasoning and agentic workloads increasingly demand efficient long-context inference. Yet full-attention decoding reads the growing history at every step, regardless of its benefit to the next prediction. We show that a pretrained model's decoding states already contain information predictive of this benefit, before the global read. Building on this finding, we introduce On-Demand Attention (ODA), a local-first decoding method that uses a lightweight recall head to selectively invoke global attention as its predicted benefit changes during generation. ODA trains only the recall head, leaving pretrained weights unchanged and the complete historical KV cache available for future recall. We further implement GPU-side conditional execution in vLLM, translating reduced global reads into practical decoding speedups over full attention at long context lengths. Experiments across Qwen and Gemma models, including hybrid-attention backbones, show that selective recall recovers most of the performance lost under local attention while substantially reducing global reads. These findings support long-context inference in which pretrained models guide their own access to the information they retain.

Haibo Feng, Ruiqi Liang, Hanyang Peng +1

Southern University of Science and Technology · Peking University · Peng Cheng Laboratory

Language Model Decoding Long-Context Language Model Inference LLM Inference Acceleration Attention Mechanisms

Sep 17, 2026cs.CL

Relational Attention for Data-Efficient Language Modeling

We present Relational BabyLM, a system submission to the BabyLM 2026 challenge that combines two cognitively motivated inductive biases in a single decoder-only Transformer. Architecturally, we replace standard self-attention with a Dual Attention Transformer (DAT), which separates the routing of object-level ("sensory") lexical features from structural/relational information (Altabaa and Lafferty, 2025; Altabaa et al., 2024; Webb et al., 2024; Kerg et al., 2022; Webb et al., 2021). Relational attention (RA) disentangled from self-attention greatly increases data efficiency and out-of-training-sample generalization on purely relational tasks, but language modeling requires object-level and relational information to be integrated as well as disentangled, and RA-based LMs have remained largely unexplored. BabyLM's data-constrained training and comprehensive evaluation is an ideal testing ground for whether that data efficiency transfers. As a training intervention, we add a Next-Latent Prediction (NextLat; Teoh et al. 2026) objective that encourages hidden states to compress history incrementally into a dense belief state. Architecture is the dominant factor for structural linguistic generalization; the objective is secondary but still significant. DAT's three relational attention types (full RA vs. the simpler RCA and DisRCA variants) are largely interchangeable at 10M words; full RA pulls ahead at 100M. We also introduce a novel symbol-retrieval mechanism (RoPE-based, as opposed to learned, relative symbols) that matches learned symbol libraries while adding no parameters. On the strict (100M-word) track, our best model ranks 6th of 55 overall and 3rd of 55 on the leaderboard's NLP-task subset at the time of writing; our two strongest models outperform the GPT-2 baseline on most benchmarks, with one attaining the highest EWoK score among strict-track entries.

Adrian Brasoveanu, Ece Takmaz, Jakub Dotlačil

Language Model Pretraining Decoder-Only Language Models Language Modeling Attention Mechanisms

Sep 17, 2026cs.LG

SETTer: Sparse-Encoder Transformer for Long-term Multivariate Time Series Forecasting

Long-term multivariate time series plays a significant role in many application areas such as power systems, trading, etc. However, their accurate prediction is quite difficult for conventional forecasting methods as they often exhibit high dimensionality and complex relationships. Recent works show that transformer-based approaches are quite effective for long-term forecasting thanks to their attention mechanism. However, in the presence of complex high-dimensional inputs, they show evidence of oversmoothing, limited capacity, and opacity. To this end, this paper introduces SETTer, a transformer-based model that addresses these challenges by incorporating novel techniques for decoupled self-attention and hybrid masking. The proposed techniques enable SETTer to effectively capture the dominant short- and long-term patterns across the temporal and channel dimensions. In addition, we enrich the model layers with simple explainable structures that indicate the discriminative pattern of SETTer. We show that with a single-layer transformer architecture, SETTer can effectively model long-term dependencies in the presence of varying data complexities. Extensive experiments on real-word benchmark datasets for long-term multivariate time series forecasting demonstrate that SETTer outperforms state-of-the-art models in 88% of the scenarios.

Abraham Ezema, Chijioke Eze, Ferdinanda Ponci +1

Institute for Automation of Complex Power Systems, RWTH Aachen University, 52064 Aachen, Germany · RWTH Aachen University, 52064 Aachen, Germany · institute for Monitoring and Distributed Control for Power Systems Teaching and Research Area, RWTH Aachen University, 52064 Aachen, Germany +1

Multivariate Time Series Forecasting Long-Term Time Series Forecasting Explainable Time Series Forecasting Time Series Forecasting Attention Mechanisms Time Series Transformers

Sep 17, 2026cs.LG

Beyond Flattened Tokens: Structure-Preserving EEG Decoding with Reusable TriDim Blocks

Effective EEG decoding requires representations that preserve organization among channels, local waveform dynamics, and long-range temporal context. Existing EEG architectures often capture these structures using separate specialized modules or collapse them into a single token sequence, making it difficult to maintain their distinct roles and coordinate their interactions throughout the backbone. We propose TriDim, a reusable block that preserves the representation shape and keeps three EEG axes explicit: channel, sample position within each patch, and patch position across the recording. These axes correspond to spatial, short-term temporal, and long-term temporal information, respectively. Each TriDim block applies feed-forward transformations along individual axes and cross-axis attention to coordinate information exchange among them. By stacking TriDim blocks with a multi-level tri-axis readout, we construct TriDimEEG, a standalone EEG decoder. Under strict cross-subject evaluation on eight datasets spanning clinical diagnosis, sleep staging, motor imagery, and emotion recognition, TriDimEEG achieves the best overall performance among fifteen evaluated models, with a 4.3% relative improvement in average accuracy over the second-best model. Replacing Transformer blocks in three EEG foundation models with TriDim blocks yields an average relative improvement of 7.4% in downstream accuracy while reducing parameter counts by 17.0% to 47.3%. These results establish TriDim as an effective and reusable building block and TriDimEEG as a strong standalone EEG decoder. Code and parameters of TriDimEEG are available at https://github.com/ncclab-sustech/TriDim_model.

Shiyue Su, Song Wang, Zekai Zhan +7

Department of Biomedical Engineering, Southern University of Science and Technology, Shenzhen, 518055, China. · Omni-Intelligence, Shenzhen, China. · Department of Computer Science, The Chinese University of Hong Kong, Shenzhen, 518172, China. +2

Cross-Subject EEG Decoding EEG Decoding Attention Mechanisms EEG Foundation Models

Sep 16, 2026cs.LG

Directions That Don't Drift: Stiefel Manifold Routing for Transformer Attention

The query and key projections

\WQ,\WK

in attention are almost always trained by Euclidean optimizers with no geometric constraint. We constrain them to the Stiefel manifold and optimize with a Riemannian Adam carrying one scalar second moment per frame---the form of \citet{becigneul2019}, here extended to the compact, non-Hadamard

\St(d,r)

with a tangent projector, step-norm cap, and polar retraction. Four propositions prove steepest descent in the embedded metric, gradient-scale independence, well-conditioning, and exact

\mathrm{O}(d)

-equivariance. A fifth records that weight decay has \emph{identically zero} Riemannian gradient on

\St(d,r)

(

W{=}WI_r

lies in the normal space), so decay cannot act on the constrained frames. On a CIFAR-10 patch benchmark at

n{=}10\mathrm{k}

this rule gains

\mathbf{+6.79}

,pp over AdamW across 12 paired starts (

t{=}38.33

,

12/12

); earlier fixed-step Riemannian SGD gains

+1.97

,pp, of which

+1.69

,pp comes from frozen orthonormal initialization alone. The corrected Adam's lead grows with data:

+1.9

,pp at

n{=}1\mathrm{k}

to

+6.7

,pp at

n{=}50\mathrm{k}

. A 12-seed ablation credits all gain to the scale-free step (

+4.63

,pp,

12/12

), nothing to the projector or equivariance; a targeted

\varepsilon

-sweep causally confirms the mechanism (

-2.6

,pp at

\varepsilon{=}0.1

,

p{<}0.001

). Two five-seed grokking studies confirm the constrained arm does not grok better than the baseline (

p{=}0.019

, A2 wins): the weight-decay exemption has no grokking consequence. A single-seed pilot exploiting this localization achieves the first stable grokking under slingshot conditions---Stiefel + targeted circuit regularization keeps routing-frame isometry error

10^6\times

lower than the unconstrained ablation through every collapse.

Rubén Darío Guerrero

NeuroTechNet S.A.S., 1108831, Bogotá, Colombia

Stiefel Manifold Optimization Riemannian Optimization Transformer Attention Attention Mechanisms

Sep 15, 2026cs.CV

Event-based Selective Attention for Multi-resolution Fast Region of Interest (ROI) Detection

Neuromorphic vision systems operate under strict constraints on bandwidth, memory, and energy, particularly at the edge, motivating early mechanisms for data reduction and selective processing. In this work, we investigate a multi-scale training-free, saliency-based, bottom-up visual attention model that operates directly on low-resolution event-based input and selects Regions of Interest (ROI) from the visual scene. The model is evaluated across multiple downscaling factors applied to the incoming event stream, with input resolutions reduced by up to 256x relative to full resolution. Performance is assessed on the Prophesee Automotive dataset, the largest publicly available event-based dataset, demonstrating robust ROI selection across different scales on a real-world use-case. The proposed approach is capable of detecting ROIs belonging to multiple object classes, including various vehicle types, pedestrians, traffic lights, and traffic signs, with accuracy up to 70.8%, while operating at millisecond temporal resolution, 16x finer than the temporal resolution provided by the dataset ground truth. These results highlight the potential of combining early event downscaling with saliency-based attention as an effective front-end for efficient edge neuromorphic vision systems.

Luca Peres, Giulia D'Angelo, Chiara Bartolozzi +1

Department of Cybernetics Faculty of Electrical Engineering Czech Technical University in Prague Prague, Czech Republic · Istituto Italiano di Tecnologia Genoa, Italy

Event-Based Vision Visual Attention Edge Inference Attention Mechanisms

Sep 9, 2026cs.LG

Training Trajectories Determine Circuit Removability in Annealable Soft-Prior Transformers

Soft positional priors can help small Transformers learn retrieval circuits, but it is unclear whether the resulting circuits remain functional once the prior is removed. We test this with an annealable soft-prior Transformer whose attention biases can be learned, faded, or zeroed during training and evaluation. On associative recall, unforced models perform well with the prior active (

0.772 \pm 0.020

) but collapse at zero gate (

0.095 \pm 0.009

). Smooth fade-to-zero training preserves high zero-gate accuracy (

0.734 \pm 0.028

), whereas forced-zero training, hard switching, and post hoc continuation fail to recover the same effect. The pattern also appears on Markov induction. Linear regression ICL provides a boundary case because zero-gate training can learn that task directly. Mechanistic traces show that circuit consolidation occurs after the gate reaches zero, even though the responsible heads vary across seeds. These results suggest that circuit removability in small discrete retrieval tasks depends on the training trajectory, not just the final architecture.

Zonglin Yang, Ziming Zhao, Wei Tang +5

Criminal Science and Technology, Guangdong Police College, China · Cybersecurity and Law Enforcement, Guangdong Police College, China · Traffic Management Engineering, Guangdong Police College, China

Transformer Interpretability Attention Mechanisms Neural Network Training Dynamics

Sep 8, 2026cs.CL

It's Not RoPE that Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention

Large Language Models (LLMs) often exhibit "Attention Sink" (AS) and the accompanying "Massive Activations" (MAs) at the initial position of a sequence. These phenomena frequently co-occur, and MAs can pose challenges for low-bit quantization. In this study, we analyze the factors underlying AS and MAs that emerge at the initial position regardless of the token occupying it. Our experiments suggest that self-concentration of attention, resulting from the causal mask, and the subsequent Value-non-mixing in attention outputs contribute to AS and MAs. These findings provide new empirical evidence on the internal dynamics of LLMs, offering insights that may inform future quantization strategies and advance our understanding of the internal mechanisms of attention layers.

Raito Kiya, Satoki Ohashi, Kosuke Sato +6

1Tohoku University · 2SOKENDAI · 3NINJAL +3

Self-Attention Attention Sinks Attention Mechanisms

Sep 8, 2026cs.LG

Adaptive Anisotropic Attention for Axis-Structured Signals

Dense self-attention treats all token pairs as equally plausible before learning, an interaction-isotropic prior that can be mismatched to structured signals. For structured, low signal-to-noise ratio (SNR) signals such as EEG, dependencies are organized along the electrode and time axes, and this uniform prior exposes each token to many irrelevant interactions. We introduce Adaptive Anisotropic Attention (AAA), which splits attention into two paths: a temporal path, where each token attends to the tokens of its own electrode across time, and a spatial path, where it attends to the tokens of the other electrodes at the same time step. A small gate predicts, for every token, a convex combination of the two path outputs: two non-negative weights that sum to one. On six EEG downstream tasks, the resulting model, AXON (AXis-factorized Operator Network), improves mean balanced accuracy over a dense baseline under both linear probing and full fine-tuning. We show that both paths (temporal and spatial) are necessary and that the weighted sum beats a hard choice of one path; most of the benefit comes from the gate learning a different temporal/spatial balance at each layer of the network. Controlled audio spectrogram experiments show that axis factorization transfers beyond EEG. These results suggest that aligning attention with the natural axes of structured signals provides a useful inductive bias.

Mahir Jain, Parshva Runwal, Aditya Ray Mishra +4

Mannas AI · IIT Mandi

EEG Decoding Attention Mechanisms Temporal Attention

Sep 8, 2026cs.LG

Why shared attention vectors fail: a case for outcome-indexed tuning

Dimensional attention in learning is often implemented as a globally shared attention vector, where each stimulus dimension corresponds to a single scalar. These scalars are learned by models through gradient-descent on error, where predictive features acquire more salience. We show that under multi-outcome learning, where models predict more than one outcome, this shared vector becomes unstable; it collapses to its bounds and prevents the models from learning meaningful attentional tunings for learning and generalization. We address this by introducing an outcome-indexed attentional matrix that converts globally shared attentional tuning into an outcome-indexed representation. We present an analysis of the unstable shared vectors and derive the conditions under which it holds. Empirically, three synthetic experiments benchmark the proposed attention matrices and show that they converge to meaningful representations, something shared attention vectors fail to do. These results suggest that outcome-indexed attentional matrices are a general fix for gradient-based attentional processes, which improves models of learning under multi-outcome conditions.

Lenard Dome

Department of Psychiatry and Psychotherapy, Faculty of Medicine University of Tübingen, Tübingen, Germany · German Center for Mental Health (DZPG), Tübingen, Germany

Attention Mechanisms

Sep 8, 2026cs.CL

Do New Attention Mechanisms Actually Fix Attention Sinks at Million-Token Context?

Long context language models now advertise windows of one million tokens, but two habits limit how much of that window is used. Attention heads with nothing useful to read still spend their budget on the first token, which is called the attention sink, and where a fact sits in the context changes whether the model finds it. Gated attention cut first token attention from 46.7 percent to 4.8 percent at NeurIPS 2025, and Kimi K3 pairs that idea with Kimi Delta Attention and Attention Residuals behind a one million token window, eight times past the range where these diagnostics have been reported. This paper asks whether the fix survives that jump. We build SinkProbe, a suite that measures sink mass, massive activation, position resolved recall and the recency gap, and apply it to four small models that differ only in how they mix tokens and depth. Three results follow. The training objective produces the sink, not the architecture. Gating did not reproduce its published effect at our scale. Sink mass, activations and position bias moved independently. Code, data and the measurement protocol are released at https://github.com/sararizwan7/Attention-Mechanisms-in-1M-Context-Window

Sara Rizwan, Samaanah Abdus Salam

Shadan Women’s College of Engineering and Technology Hyderabad, Telangana, India

Long-Context Language Model Evaluation Long-Context Language Modeling Gated Attention Transformer Attention Attention Sinks Attention Mechanisms

Sep 8, 2026cs.LG

Nyström Attention Matches Full Attention for Cross-Sectional Stock Prediction

MASTER's inter-stock multi-head attention -- the module responsible for modeling cross-sectional stock relationships -- accounts for 42.5% of model parameters and 25% of predictive value. We systematically decompose this module and uncover a surprising structure: the learned attention is near-uniform (perplexity 278/300), yet forcing exact uniformity eliminates all cross-sectional discrimination. Spectral analysis resolves this paradox: the deviation from uniformity is low-rank (effective rank ~65, top-10 modes capture 96.5% of energy), explaining why sparse approximations consistently fail while Nystrom low-rank attention (m=32 landmarks) matches full O(N^2) attention at O(mN) cost -- certified equivalent via TOST at both N=300 (5 seeds, Rank IC p=0.003) and N=800 (10 seeds, Rank IC p=0.034). Additional findings include: (i) attention anti-correlates with return similarity (Spearman rho = -0.614; on the industry-labeled subset, -0.645 unconditionally and -0.627 after controlling for industry, beta, and volatility), suggesting complementarity-seeking rather than correlation mining; (ii) all graph-based alternatives degrade performance, with hard masking worse than complete module removal; and (iii) at N ~ 3,500 with adapted architectures, no cross-stock module (GCN, Nystrom, or MASTER-style pipeline) significantly outperforms a per-stock LSTM baseline (n=4 seeds), indicating that the benefits observed at smaller scales do not trivially transfer. These results establish that the inter-stock attention's value resides in a compressible, dynamic, near-global redistribution that rewards low-rank approximation but resists sparsification.

Kunhan Guo

The University of Hong Kong

Low-Rank Attention Financial Forecasting Time Series Forecasting Efficient Attention Attention Mechanisms

Sep 7, 2026cs.CL

DeepTable: Structural Attention Biases and Tree Path Encoding for Hierarchical Table Understanding

Large language models (LLMs) have demonstrated strong performance in table understanding. However, they typically process table content and headers as linearized token sequences. This representation weakens the two-dimensional and hierarchical structural relationships encoded by multi-level row and column headers. Existing parameter-efficient fine-tuning methods incorporate basic row and column information but do not explicitly capture the rich structural dependencies induced by hierarchical table headers. We propose DeepTable, a structure-aware approach for table understanding with LLMs. DeepTable comprises two complementary components. Structural Attention Bias (SAB) introduces learnable biases into the attention logits to explicitly represent whether pairs of table tokens share the same row or column. Tree Path Encoding (TPE) represents each table token using the ancestor paths of its row and column headers, preserving its position within the multi-level table structure. We integrate DeepTable with TableLoRA (He et al., 2025) to inject structural information into parameter-efficient adaptation. Across three LLM backbones, DeepTable consistently improves the corresponding TableLoRA baselines on three table question answering benchmarks, achieving average gains of 7.42 points on HiTab, 3.23 points on WikiTQ, and 2.01 BLEU points on FeTaQA. These results demonstrate the effectiveness of the proposed structural biases across different LLM backbones.

Jyun-Ying Yen, Cheng-Kuan Lin, Yu-Chee Tseng

Department of Computer Science National Yang Ming Chiao Tung University, Taiwan

Table QA Table Structure Recognition Attention Mechanisms

Sep 7, 2026cs.AI

RedKnot-MLA: Multi-Head Offline-Online Reuse for DeepSeek-V4 Long-Context Serving

Multi-head latent attention (MLA) exposes many logical query heads through one packed latent KV stream. This representation is memory efficient, but it removes the physical per-head cache boundary assumed by conventional head-wise reuse. We present our system, a DeepSeek-V4 realization of RedKnot's head-aware reuse principle. Each immutable document is processed offline at canonical position zero; certified Local-head contributions are retained as MLA-Off. At serving time, query-side RoPE relocation restores the document's request position, a small Global-head set and protected Local token rows are recomputed as MLA-Online, and the two paths are merged before a single shared output projection. The packed MLA latent is never split. DeepSeek-V4-Flash uses 37 reusable layers and a 56/8 Local/Global partition, giving a 75.29% analytic logical head-row ceiling; the Pro-0813 profile uses 55 layers and 112/16 heads, giving 78.89%. Frozen Flash operating points show hot-artifact TTFT speedups of 2.02-3.84x. At 256K, the archived three-dataset study reports an aggregate F1 change of +3.24 percentage points, an EM change of +4.16 points, and a 78.7-79.5% analytic major-operator arithmetic saving, while one dataset decreases by 2.81 F1 points. A separate author-reported 256K hot-artifact QPS measurement is approximately 2.0x; because its raw concurrency trace is not included in this bundle, we mark it as preliminary rather than archived evidence. We describe the factorization, position repair, token-row closure, sparse-MoE support, TP8 integration, and the measurement boundaries needed to interpret these results.

Yang Liu, Zhaokai Luo, Huayi Jin +10

Xiaohongshu Inc., China · Huawei Cloud · Shanghai Jiao Tong University +2

KV Caching LLM Inference Acceleration KV-Cache Reuse Attention Mechanisms Multi-Head Latent Attention

Sep 6, 2026cs.CV

RoLA: Rotary-Positioned Low-Rank Linear Attention for Efficient Diffusion Transformers

Diffusion Transformers (DiTs) achieve strong video generation quality, but their dense spatiotemporal self-attention scales quadratically with sequence length and quickly becomes the dominant inference bottleneck. Sparse low-rank hybrids alleviate this cost by combining a local sparse branch with a global compressed branch. In video DiTs equipped with 3D Rotary Position Embeddings (RoPE), the global branch faces a structural compatibility issue: when RoPE is applied before a nonlinear feature map, the rotation and nonlinearity generally do not commute, making it difficult to keep a query-independent linear summary while preserving relative rotary geometry. Existing work often sidesteps this issue by replacing genuine cross-token global aggregation with coordinate-conditioned surrogates or learnable absolute positional modules. These compromises can be effective, but they approximate relative decay from absolute coordinates and introduce extra positional parameters. We propose \textbf{RoLA}, a rotary-positioned low-rank linear-attention branch that keeps genuine cross-token aggregation while remaining compatible with a reusable linear summary. The design applies RoPE \emph{outside} the nonlinear low-rank feature map and reuses a truncated subset of the pre-trained rotary schedule matched to the low-rank bottleneck. This yields a linear-time low-rank global branch with relative positional behavior by design and no additional positional parameters; the full sparse--low-rank module still includes the fixed-sparsity sparse branch. Experiments on open-source video DiTs show that the resulting method remains competitive in generation quality at 90% sparsity while achieving 2.63

\times

end-to-end inference speedup on Wan2.1-14B (720p, 81 frames, measured on an NVIDIA H100 GPU).

Zekun Zhang, Yixiang Cai, Yuxi Liu +9

Peking University, Melon Group · Alibaba Group · University of Electronic Science and Technology of China +1

Rotary Positional Embeddings Low-Rank Attention Video Diffusion Models Diffusion Transformer Linear Attention Attention Mechanisms+1

Sep 3, 2026cs.LG

Hardware-Aware FP4 FlashAttention-4

Blackwell's 4-bit floating-point (FP4) tensor cores do not automatically make attention faster because softmax conversion and on-chip dependencies dominate once its matrix products shrink. We address this with \emph{Direct-P} for noncausal inference and a causal path that passes the forward quantization directly into backward. Direct-P maps scores directly to FP4 probabilities and reaches up to 2.13

\times

the bfloat16 (BF16) forward throughput on an NVIDIA GB200. The causal path reconstructs probabilities from saved quantized queries and keys and uses 8-bit floating-point (FP8) gradient operands, accelerating a complete single-GPU 8-billion-parameter update by up to 1.14

\times

. Matched distributed training retains FP8 probabilities and values; every tested MXFP4 probability/value training trajectory diverges.

Robert Hu

GRAPHCORE RESEARCH

Mixed-Precision Quantization GPU Kernel Optimization Transformer Attention Efficient Attention Attention Quantization Attention Mechanisms

Sep 3, 2026cs.CC

The Head Complexity of Boolean Functions in Single-Layer Attention

What can a single layer of self-attention compute? We study head complexity: the minimum number of attention heads required to compute a function in a one-layer attention-only model. We establish an exact hierarchy under this measure:

k

heads compute

k

-bit parity but cannot compute

(k+1)

-bit parity. The lower bound is unconditional in the two resources a transformer might otherwise exploit; it holds at unbounded embedding dimension and unbounded numerical precision. The proof rests on an alternating-sum obstruction: after clearing the softmax denominators, every monomial in the resulting decision polynomial omits at least one of the

k+1

input bits, forcing its correlation with parity to vanish. The same obstruction yields lower bounds for related tasks, including the well-studied multi-hop induction-head task. We also establish compactness bounds for embedding dimension and numerical precision. Specifically, a compactness theorem shows that any function computable at all can be computed with embedding dimension and precision bounded by the discrete data of the task, namely, head count, alphabet size, and length. Thus, potentially unbounded dimension or precision provably cannot substitute for heads. Finally, we derive nearly matching universal bounds for general binary functions:

2^n

heads suffice to compute every

n

-bit binary function, with one head per monomial in its multilinear expansion, while a counting argument shows almost all such functions require

Ω(2^n/n^2)

heads. This lower bound matches the upper bound to within a

\operatorname{poly}(n)

factor, even when dimension and precision are unbounded. Together, these results characterize head requirements for Boolean computation in this model.

Rajmohan Rajaraman, Ravi Sundaram, Amanuel Tesfaye

Northeastern University.

Self-Attention Attention Mechanisms Multi-Head Attention

Sep 3, 2026cs.LG

VestigeKV: The NoPE-MLA KV Cache Carries Its Own Sparse-Attention Signal in a Vestigial Branch

A long-lived KV cache must be compressed before the queries that will read it exist. Selection by observed attention collapses there: on a NoPE-MLA model, H2O and SnapKV retrieve 0.00 and 0.33 of needles at 8x compression, because a token's importance has not yet been observed. VestigeKV instead derives a sparse attention pattern from a signal the cache already carries, occupying the sparse-attention literature's one unoccupied quadrant: training-free and query-independent. In NoPE-MLA the 64-dimensional decoupled branch is a vestige of RoPE that training repurposes into a salience channel; reading 11% of each row, it partitions the cache into an attended tier and a GPU-resident archive that no row ever leaves, reachable each step by a certified, query-adaptive trigger. Nothing is trained and cache rows are never quantized, so every quality effect attributes to selection and scheduling. On Kimi Linear 48B, retrieval holds at 1.00 under 8x and 0.96 under 32x from 8k to 65k context, with zero gap to full-row selection, and the recall tier holds 128x at 1.00 (8k). Both tiers stay on the GPU, so the win is speed, not memory: the per-step scan reads ~26% of the bytes dense attention would, and on a two-node sglang deployment the crossover sits at ~40k context, reaching 1.18x at 256k and 1.39x at 496k. The mechanism is exclusive to NoPE: the identical operator on a RoPE MLA collapses to 0.08, query-independent salience exists only without rotation, and query-universal exact merging is provably impossible under RoPE. All thresholds were frozen before their data; 20 archived verdicts and 8 closed routes accompany the paper.

WenJie Fan

Yotta Labs

KV Caching KV-Cache Management LLM Inference Acceleration Sparse Attention KV-Cache Compression Attention Mechanisms+1

Sep 3, 2026cs.LG

High-Dimensional Learning Dynamics of Attention-Indexed Models

Attention mechanisms are central to modern foundation models, yet their training dynamics remain poorly understood, especially when the attention matrices have extensive rank. In this work, we study attention-indexed models, a broad framework that can represent multi-layer and multi-head attention architectures. First, we show that, in a suitable high-dimensional limit, the population-loss landscape is characterized by a finite set of trace order parameters. In contrast, online stochastic gradient descent (SGD) is governed by an infinite hierarchy of matrix moments, which we show can be exponentially well-approximated by a finite truncated system. Second, this framework reveals that attention parameterization itself can act as an architectural implicit bias. Direct optimization of an attention matrix

S\in\mathbb{R}^{d\times d}

can remain trapped in an uninformative state. Tied attention (

S=WW^\top

) induces an automatic symmetry-breaking mechanism and yields weak recovery in

Θ(d^2\log d)

samples. For untied attention,

S=UV^\top

, we uncover a fast-slow mechanism: the pre-activation mean first evolves on a fast timescale, while the overlaps evolve on a slower one. Weak recovery on the

Θ(d^2\log d)

scale occurs when the state selected by the fast dynamics breaks the initial symmetry.

Yizhou Xu, Margarita Sagitova, Lenka Zdeborová +1

1Information, Learning and Physics Laboratory, École Polytechnique Fédérale de Lausanne (EPFL) · 2Statistical Physics of Computation Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)

Implicit Bias Attention Mechanisms Neural Network Training Dynamics Statistical Learning Theory

Sep 1, 2026cs.LG

Scaled Idempotence in Transformer Attention: Paired OV Geometry and Shared-Value Algebras

We identify a recurrent algebraic regularity in Transformer attention: a sparse subset of effective OV operators

T=OV^\top

nearly closes under composition,

T^2\approxαT

. Across six pretrained endpoints spanning 2.8B--235B parameters, 3.98--8.00% of heads reach squared closure alignment

\mathcal{P}\geq0.9

, while no matched within-layer O/V mismatch does. An exact principal-coordinate factorization,

T=Q_OKQ_V^\top

and

T^2=Q_O(KDK)Q_V^\top

, separates within-support transport from read--write return geometry. Across all 7,304 heads in nine MHA/GQA models, scrambling only the orientation of

K

while preserving singular values, norms, factor spans, and principal angles reduces median closure from 0.336 to

1.04\times10^{-4}

; trained orientation wins for 98.64% of heads and in every layer. Constructive searches show that high closure is feasible in every surveyed layer, but usually not attained. Retrospective trajectories in three independently trained lineages further separate broadly available capacity from the orientations attained by final strong heads. Under exact value sharing, headwise closure extends to a right-action algebra,

T_iT_j=α_jT_i

. Seven-model experiments verify the approximate law and reveal distinct oblique projections with a shared value-defined kernel. These results characterize scaled idempotence as a sparse trained orientation within broadly available geometric capacity and show how value sharing extends a headwise relation into a local operator algebra.

Jiming Feng, Junliang Li

College of Computer Science, Beijing University of Technology Beijing, China · School of Mathematics, Statistics and Mechanics, Beijing University of Technology Beijing, China

Attention Head Analysis Transformer Attention Attention Mechanisms Multi-Head Attention

Aug 31, 2026cs.CL

Toppling the Hierarchy in Byte-level Language Modeling

This work examines recent byte-level models and their failure to perfectly manipulate characters. State-of-the-art byte-level models use a hierarchical structure, starting at the byte level, downsampling to the word level, and then upsampling back to bytes. While this improves training and inference efficiency, we find that the hierarchical design itself limits character-level understanding, with pure byte-level models consistently outperforming hierarchical variants on character manipulation tasks. Ablating transformer layers into attention and feed-forward components further reveals that byte-level attention is the primary mechanism driving this behavior. Together, our results provide an explanation for the character-level failures of hierarchical byte models and establish a clear trade-off between computational efficiency and fine-grained character understanding.

Lukas Edman, Alexander Fraser

School of Computation, Information and Technology, TU Munich · Munich Center for Machine Learning · Munich Data Science Institute

Efficient Language Model Training Byte-Level Language Model Language Modeling Attention Mechanisms

Aug 31, 2026cs.CV

SELECT: SELEctive Context Transfer for Class-Incremental Semantic Segmentation

Class-Incremental Semantic Segmentation (CISS) is fundamentally challenged by catastrophic forgetting and background shift, where learning new concepts degrades performance on previously seen classes. While existing methods attempt to balance stability (retaining old knowledge) and plasticity (learning new knowledge), they often fail to leverage prior knowledge effectively. These approaches typically rely on indiscriminate knowledge transfer or ambiguous initializations, which can dilute crucial semantic information. To overcome this limitation, we propose SELECT, a novel approach for Selective Context Transfer, which instead grounds each new class in a small set of semantically similar past classes. Its core is a Context Transfer Attention mechanism that aggregates the learned tokens from similar classes into a structured initialization for the new class. To ensure this transfer does not corrupt the borrowed representations, we add a controlled noise perturbation and a margin-based context-transfer loss that enforces separation between the new class token and its source tokens. Extensive experiments on Pascal VOC and ADE20K show that SELECT consistently outperforms prior work, achieving mIoU of 2.2% on VOC and 2.8% on ADE, providing an effective handle on the stability-plasticity dilemma. Code is available at https://github.com/avigupta2798/SELECT.

Avi Gupta, Saurabh Yadav, Koteswar Rao Jerripothula +1

IIIT Delhi · Microsoft, India · IIT Kanpur +1

Class-Incremental Learning Attention Mechanisms Semantic Segmentation Catastrophic Forgetting

Aug 31, 2026cs.LG

TPR-Attention for Combinatorial Generalization

Systematic generalization remains a significant challenge in deep learning. In particular, combinatorial generalization - generalizing to new configurations of known factors of variation - is effortless for humans but difficult for standard neural architectures that rely on statistical correlations rather than explicit structural representations. We introduce a new architectural component that embeds structured inductive bias into deep learning: an attention mechanism operating over tensor-product representations (TPRs). Through controlled experiments on compositional tasks, we show that this TPR-attention mechanism outperforms existing architectural components in combinatorial generalization. These results highlight the value of integrating explicit compositional structure into neural attention and point toward a promising path for models capable of systematic generalization.

Melisa Civelekoğlu, Isabeau Prémont-Schwarz

Universit´e Laval · Institut intelligence et donn´ees · Mila – Qu´ebec Artificial Intelligence Institute

Attention Mechanisms Compositional Representation Learning Compositional Generalization

Aug 30, 2026cs.LG

A Lightweight Phenology-Aware YOLOv5 Framework for Tomato Growth Stage Detection in Resource-Constrained Bhutanese Greenhouse Environments

Accurate detection of tomato growth stages is essential for stage-specific greenhouse management and precision agriculture. In Bhutan, greenhouse cultivation is affected by altitude variability, large diurnal temperature fluctuations, diffuse illumination, limited automation, and a scarcity of locally annotated datasets, limiting the applicability of conventional deep learning models. This work proposes Pheno-Lite + Efficient Channel Attention (ECA), a lightweight, phenology-aware object detection architecture derived from Ultralytics YOLOv5 for tomato growth stage recognition. A balanced dataset of 2,464 annotated images was constructed from locally collected greenhouse images in Bhutan and publicly available tomato images, with augmentation designed to simulate local greenhouse conditions. The dataset includes vegetative (820), flowering (824), fruiting (820), and background (26) samples. The proposed architecture introduces two customized backbone modules: C3 PhenoLite, which enhances spatial and texture feature extraction using depthwise residual refinement, and C3 ECA, which strengthens inter-channel feature interactions through efficient channel attention. The proposed model achieves 90.6% precision, 88.8% recall, and 92.6% mAP@50, with 4.0 million parameters and 10.9 GFLOPs at 640 x 640 resolution. These results demonstrate its potential for real-time and climate-resilient greenhouse deployment in Bhutan.

Sherab Gocha, Sou Nobukawa

Graduate School of Information and Computer Science · Chiba Institute of Technology · Chiba, Japan +1

Plant Phenotyping Object Detection Real-Time Object Detection Attention Mechanisms

Aug 28, 2026cs.LG

Intrinsic Interaction Geometry Controls the Low-Rank Complexity of Softmax Attention

How much matrix rank is required to preserve every bounded value output of normalized softmax attention? We study the unrestricted maximum-row-

\ell_1

approximation rank

r_\varepsilon(A)

, exactly the least rank achieving uniform error over all bounded vector-valued values. Row softmax exposes the intrinsic interaction

C=P_m(\log A)P_N

, whereas invertible

Q/K

gauges leave

A

fixed while changing the Euclidean geometry of a chosen query/key factorization. We replace that coordinate-dependent description by a projective residual

q(C-T)

and an attained factor-radius size

κ(T)

. For every rank-

r

retained interaction with

τ(T)<\varepsilon

, we prove

r_\varepsilon(A)\le \min\left\{ N,\; C_r\left( 1+\frac{κ(T)} {(\varepsilon-τ(T))^2} \right)^{r/2} \right\},

with the same unknown dimension constant as the underlying weighted Gibbs-row cover. The profile is gauge invariant, termwise no worse than native retained-subspace bounds at the same declared dimension, and has a worst-case sharp

r/2

size exponent at fixed

r

and

\varepsilon

. We then measure

r_\varepsilon(A)

directly on learned attention using 9,978 certified brackets across BERT, GPT-2, Qwen2.5, and two ViT checkpoints; where certificates do not close, the optimum remains interval-valued. A pre-specified 2,302-cell held-out study further shows that the historical native-coordinate geometry block contains coarse, mostly head-level information but no detectable incremental information beyond a strong calibrated baseline. The new intrinsic descriptor is not evaluated in that study. Together, the theory and measurements distinguish an operator-intrinsic complexity control from a stronger empirical explanation that the learned-head evidence does not support.

Yuhe Sui, Jianing Zhang, Yingzhi Tang

Nanyang Technological University, Singapore · Carnegie Mellon University

Softmax Attention Low-Rank Attention Low-Rank Approximation Attention Mechanisms