Experts

Mixture-of-Experts (MoE) models rely on balanced expert utilization to fully realize their scalability. However, existing load-balancing methods are largely heuristic and operate on noisy mini-batch assignment statistics, introducing bias relative to population-level objectives. We propose $φ$-balancing, a principled framework that directly targets population-level expert balance by minimizing a strictly convex, symmetric, and differentiable potential of the expected routing distribution. Using convex duality, we derive an equivalent min-max formulation and obtain a simple online algorithm via mirror descent, yielding an efficient EMA-based routing adjustment with negligible overhead. Across large-scale pretraining and downstream fine-tuning, $φ$-balancing consistently outperforms prior Switch-style and loss-free baselines, demonstrating more stable and effective expert utilization.

Recent advances in neural operators have made partial differential equation (PDE) surrogate modeling increasingly scalable and transferable through large-scale pretraining and in-context adaptation. However, after a shared operator is fine-tuned to multiple regimes within a continuous physical family, it remains unclear whether the resulting weight-space updates merely form isolated regime experts or reveal reusable physical structure. Starting from a shared family anchor, we fine-tune low- and high-regime endpoint experts and show that their updates can be separated into a family-shared adaptation and a direction aligned with the underlying physical parameter. This separation reinterprets endpoint experts as finite-difference probes of a local physical direction in weight space, explaining why static averaging can interpolate between regimes but attenuates endpoint-specific physics. Building on this perspective, we propose Calibration-Conditioned Merge (CCM), a post-hoc coordinate readout method for composing neural PDE experts along this physical direction. Given physical metadata, a calibrated coordinate mapping, or a short observed rollout prefix, CCM infers the target composition coordinate and deploys a single merged checkpoint for the remaining rollout. We evaluate CCM on the reaction--diffusion system, viscosity-parameterized two-dimensional Navier--Stokes equations, and radial dam-break dynamics. Across these benchmarks, CCM achieves its strongest gains in extrapolative regimes, reducing out-of-distribution rollout error relative to the family anchor by 54.2%, 42.8%, and 13.8%, respectively. Further experiments across FNO scales, a DPOT-style backbone, and ablations confirm that endpoint fine-tuning is not arbitrary checkpoint drift, but reveals a calibratable physical direction for training-free transfer across PDE regimes.

Vector quantization is a fundamental tool for compressing high-dimensional embeddings, yet existing multi-codebook methods rely on static codebooks that limit expressiveness under heterogeneous data geometry. While recent dynamic quantizers like QINCo adapt codebooks to individual inputs and improve expressiveness, their strict sequential dependencies create decoding bottlenecks. We propose Residual Quantization via Mixture of Experts (RQ-MoE), a framework combining a two-level MoE with dual-stream quantization to enable input-dependent codebook adaptation for efficient vector quantization. RQ-MoE enables dynamic codebook construction and decouples instruction from quantization, facilitating parallel decoding. Theoretically, we show that standard Residual Quantization and QINCo can be recovered as constrained special cases of RQ-MoE, and derive a guideline for setting expert dimensionality in RQ-MoE. Extensive experiments show that RQ-MoE achieves state-of-the-art or on-par performance in reconstruction and retrieval, while providing 6x-14x faster decoding than prior vector quantization methods. The implementation is available at https://github.com/KDEGroup/RQ-MoE.

This research is primarily concerned with the critical problem of synthesizing a structured Retrieval-Augmented Generation (RAG) system for advanced AI applications in the domain of swimming. As the integration of Artificial Intelligence in sports science matures, its applications in swimming have become increasingly diverse, spanning from real-time technical coaching and talent scouting to comprehensive performance profiling and the dynamic personalization of training periodization. Within this landscape, RAG-based systems represent a pivotal advancement in Large Language Model (LLM) enhanced swimming analysis, as they allow for the grounding of generative outputs in authoritative domain knowledge, thereby ensuring the credibility of AI-generated advice, contextually and technically. Despite this potential, building robust RAG systems using only real-world aquatic data presents significant challenges, including ethical constraints regarding athlete biometrics, and the high cost of manual expert labeling. To address these barriers, we propose a novel generative framework that leverages a multimodal knowledge base gathered across four dimensions: physiological data, physiological literature, kinematic sensor data, and unstructured domain expertise. Our proposed framework utilizes a multi-agent LLM architecture to synthesize a high-fidelity dataset of 1,864 validated "Question-Context-Answer" triplets-drawn from 1,914 drafts evaluated against 12 physiological soundness rules. By providing a structured, synthetic ground truth, this work establishes a foundational benchmark for trustworthy AI in aquatics. The outcomes of this research promise to enhance the reliability of automated coaching and open a plethora of future directions in "Meta-Agent" development and athletic profiling, ultimately bridging the gap between raw data engineering and practical sports science application.

Sparse Mixture-of-Experts (SMoE) models enable scaling language models efficiently, but training them remains challenging, as routing can collapse onto few experts and auxiliary load-balancing losses can reduce specialization. Motivated by these hurdles, we study how routing decisions in SMoEs are formed mechanistically. First, we reveal a geometric coupling between routers and their corresponding experts. For a given token, the router weights for the selected expert and the expert weights processing it receive gradients along the same input direction, differing only in scalar coefficients. Thus, matched router--expert directions accumulate the same routed token history. This theoretical coupling also appears empirically in routing dynamics. In a $1$B SMoE trained from scratch, higher router scores predict stronger expert neuron activations, showing that routing decisions are mirrored inside the selected expert. Next, we analyze the effects of auxiliary load balancing on the router--expert geometric coupling, showing that such losses break this structure by spreading input-directed gradients across router weights, making distinct router directions nearly three times more similar to each other. Last, we demonstrate the centrality of geometric coupling for effective routing with a parameter-free online K-Means router, in which each expert maintains a running average of the hidden states routed to it and tokens are assigned based on cosine similarity. Compared with auxiliary-loss and loss-free balancing, this router achieves the lowest load imbalance with only a modest perplexity increase, indicating that geometric coupling captures a substantial part of what the router learns. Overall, our results explain how routers form assignment geometry that supports an effective division of labor.

Learning-to-Defer (L2D) methods route each query either to a predictive model or to external experts. While existing work studies this problem in batch settings, real-world deployments require handling streaming data, changing expert availability, and shifting expert distribution. We introduce the first online L2D algorithm for multiclass classification with bandit feedback and a dynamically varying pool of experts. Our method achieves regret guarantees of $O((n+n_e)T^{2/3})$ in general and $O((n+n_e)\sqrt{T})$ under a low-noise condition, where $T$ is the time horizon, $n$ is the number of labels, and $n_e$ is the number of distinct experts observed across rounds. The analysis builds on novel $\mathcal{H}$-consistency bounds for the online framework, combined with first-order methods for online convex optimization. Experiments on synthetic and real-world datasets demonstrate that our approach effectively extends standard Learning-to-Defer to settings with varying expert availability and reliability.

World models aim to predict plausible futures consistent with past observations, a capability central to planning and decision-making in reinforcement learning. Yet, existing architectures face a fundamental memory trade-off: transformers preserve local detail but are bottlenecked by quadratic attention, while recurrent and state-space models scale more efficiently but compress history at the cost of fidelity. To overcome this trade-off, we suggest decoupling future-past consistency from any single architecture and instead leveraging a set of specialized experts. We introduce a diffusion-based framework that integrates heterogeneous memory models through a contrastive product-of-experts formulation. Our approach instantiates three complementary roles: a short-term memory expert that captures fine local dynamics, a long-term memory expert that stores episodic history in external diffusion weights via lightweight test-time finetuning, and a spatial long-term memory expert that enforces geometric and spatial coherence. This compositional design avoids mode collapse and scales to long contexts without incurring a quadratic cost. Across simulated and real-world benchmarks, our method improves temporal consistency, recall of past observations, and navigation performance, establishing a novel paradigm for building and operating memory-augmented diffusion world models.

Mixture-of-Experts (MoE) architectures have become standard in large language models, yet many of their core design choices - expert count, granularity, shared experts, load balancing, token dropping - have only been studied one or two at a time over narrow configuration ranges. It remains an open question whether these choices can be optimized independently, without considering interactions. We present the first systematic study of over 2,000 pretraining runs spanning models up to 6.6B total parameters, in which we exhaustively vary total experts, expert dimension, heterogeneous expert sizing within a single layer, shared expert size and load-balancing mechanisms. We find that at every active-parameter scale that we study, performance consistently improves with total MoE parameters even at extreme active expert parameter ratios like 128.Further, the optimal expert size is nearly invariant to total parameter count and depends only on active parameter count. Third, we see that other choices like shared experts, heterogeneous experts and load-balancing settings have small effects relative to expert count and granularity, although dropless routing yields a consistent gain. Overall, our results suggest a simpler recipe: focus on expert count and granularity, other choices have minimal effect on final quality.

Low-Rank Adaptation (LoRA) dominates parameter-efficient fine-tuning of large language models, yet most variants target dense architectures. Mixture-of-Experts (MoE) models scale parameters at near-constant per-token compute, and their sparse activation patterns create untapped opportunities for more efficient adaptation. We propose Hot-Experts Layer-level Low-Rank Adaptation (HELLoRA), which attaches LoRA modules only to the most frequently activated experts at each layer. This simple mechanism reduces trainable parameters and adapter-induced FLOPs while improving downstream performance, an effect we attribute to a form of structured regularization that preserves pretrained expert specialization. To stress-test HELLoRA under extreme parameter budgets, we further compose it with LoRI to form HELLoRI, which freezes the up-projection and sparsifies the down-projection. Across three MoE backbones, namely OlMoE-1B-7B, Mixtral-8x7B, and DeepSeekMoE, and three task families covering mathematical reasoning, code generation, and safety alignment, HELLoRA consistently outperforms strong PEFT baselines. Relative to vanilla LoRA on OlMoE, HELLoRA uses 15.7% of the trainable parameters, reduces adapter FLOPs by 38.7%, achieves 1.9x the training throughput, and improves accuracy by 9.2%. On DeepSeekMoE, HELLoRA outperforms LoRA while using only 23.2% of its trainable parameters. These results demonstrate that activation-aware adapter placement is an effective and practical route to scaling PEFT for MoE language models.

Many continuous-control problems ship with a competent but suboptimal controller (a tuned PID, a hand-designed gait). A growing family of methods uses such controllers as queryable experts during RL, but each method has been proposed in isolation, on a different benchmark, without imperfect-expert testing. We harmonize the comparison on a shared SAC backbone, common HPO and evaluation protocols, 100/50 seeds per (env, method), and a degradation sweep over expert undertuning, action bias, and observation noise. The comparison surfaces three failure modes single-paper evaluations miss: (F1) a critic blind spot under argmax-plus-bootstrap that drags IBRL below no-expert SAC on experts close to the no-expert-RL ceiling (RL-near-ceiling, distinct from the absolute physical ceiling); (F2) residual saturation on far-from-optimal experts; and (F3) warm-start buffer poisoning that collapses training-time-handoff methods under deployment-time expert undertuning. No single method dominates: each wins on one task-structure regime and fails predictably elsewhere; on RL-near-ceiling experts (FourTank, GlassFurnace) no query-time method clears the expert within our 1M-step budget, leaving open whether this is a fundamental wall or a budget effect. We convert the spread into a testable decision rule keyed on three pre-training observables (expert quality, task termination, perturbation type). The benchmark, taxonomy, and decision rule are the primary contribution; we additionally describe EDGE, a softmax-over-ensemble-LCB design point used to demonstrate that both axes the taxonomy points to (gate form, scoring rule) are individually exploitable.

Many biological systems evolve through continuous local dynamics while switching between latent regimes defined by learning, stimulus context, internal state, or developmental stage. These processes are often observed only as unpaired longitudinal snapshots: the same cells, neurons, or animals are not tracked as matched trajectories, even though population states are sampled across successive stages. This creates two coupled challenges. First, trajectories must respect curved low-dimensional manifolds embedded in high-dimensional biological measurements. Second, the model must identify when the transport mechanism itself changes. We introduce FLUX (FLow matching for Unpaired longitudinal data with miXture-of-experts), a geometry-aware longitudinal flow-matching framework for joint transport modeling and unsupervised regime discovery. FLUX learns a data-dependent metric from pooled labeled and unlabeled observations, uses that metric to construct geometry-aware conditional paths between adjacent marginals, and decomposes the resulting velocity field into sparse expert vector fields selected by a Straight-Through Gumbel-Softmax router. Across manifold controls, a regime-switching Lorenz system, widefield cortical calcium imaging during associative learning, and embryoid body single-cell differentiation, FLUX reconstructs longitudinal transport while recovering interpretable regime structure. Ablations show that mixture-of-experts routing alone is insufficient: FLUX without geometric learning can fit local transport but fails or weakens regime discovery when regimes are encoded in local dynamics. These results suggest that geometry-aware velocity decomposition provides a general strategy for discovering latent biological state transitions from unpaired longitudinal snapshots.

Mixture-of-Experts (MoE) language models route each token to a small subset of experts, but whether the routes selected by a trained top-$k$ router are good ones is rarely evaluated directly. Holding the model fixed, we compare each standard route against sampled equal-compute alternatives for the same token and score each by the next-token probability it assigns to the realized token in a verified reasoning trajectory. The result is sharply token-conditional: the standard router is well-aligned with route utility on confident tokens but uninformative on the fragile tokens that drive hard reasoning, where lower-loss equal-compute routes consistently exist inside the frozen model but are not selected. The same pattern holds across Qwen3-30B-A3B, GPT-OSS-20B, DeepSeek-V2-Lite, and OLMoE-1B-7B, and follows structurally from how standard top-$k$ training evaluates routing decisions: the language modeling loss scores only the executed route, and load balancing depends only on aggregate routing statistics. A minimal router-only update to the final-layer router, leaving every expert and every other router frozen, is sufficient to shift pass@K on AIME 2024+2025 and HMMT 2025 for both Qwen3-30B-A3B and GPT-OSS-20B, suggesting that at least part of the failure reflects router-reachable misallocation rather than expert capacity alone.

Transformer architecture search (TAS) discovers optimal vision transformer (ViT) architectures automatically, reducing human effort to manually design ViTs. However, existing TAS methods suffer from the feature collapse problem, where subnets within a supernet fail to learn subnet-specific features, mainly due to the shared weights in a supernet, limiting the performance of individual subnets. To address this, we propose TAS-LoRA, a novel method that introduces parameter-efficient low-rank adaptation (LoRA) to enable subnet-specific feature learning, while maintaining computational efficiency. TAS-LoRA incorporates a Mixture-of-LoRAExperts (MoLE) strategy, where a lightweight router dynamically assigns LoRA experts based on subnet architectures, and introduces a group-wise router initialization technique to encourage diverse feature learning across experts early in training. Extensive experiments on ImageNet and several transfer learning benchmarks, including CIFAR-10/100, Flowers, CARS, and INAT-19, demonstrate that TAS-LoRA mitigates feature collapse effectively, improving performance over state-of-the-art TAS methods significantly.

Modern Mixture-of-Experts (MoE) architectures allocate expert capacity through a rigid per-layer rule: each transformer layer owns a separate expert set. This convention couples depth scaling with linear expert-parameter growth and assumes that every layer needs isolated expert capacity. However, recent analyses and our routing probe challenge this allocation rule: replacing a deeper layer's learned top-k router with uniform random routing drops downstream accuracy by only 1.0-1.6 points across multiple production MoE models. Motivated by this redundancy, we propose UniPool, an MoE architecture that treats expert capacity as a global architectural budget by replacing per-layer expert ownership with a single shared pool accessed by independent per-layer routers. To enable stable and balanced training under sharing, we introduce a pool-level auxiliary loss that balances expert utilization across the entire pool, and adopt NormRouter to provide sparse and scale-stable routing into the shared expert pool. Across five LLaMA-architecture model scales (182M, 469M, 650M, 830M, and 978M parameters) trained on 30B tokens from the Pile, UniPool consistently improves validation loss and perplexity over the matched vanilla MoE baselines. Across these scales, UniPool reduces validation loss by up to 0.0386 relative to vanilla MoE. Beyond raw loss improvement, our results identify pool size as an explicit depth-scaling hyperparameter: reduced-pool UniPool variants using only 41.6%-66.7% of the vanilla expert-parameter budget match or outperform layer-wise MoE at the tested scales. This shows that, under a shared-pool design, expert parameters need not grow linearly with depth; they can grow sublinearly while remaining more efficient and effective than vanilla MoE. Further analysis shows that UniPool's benefits compose with finer-grained expert decomposition.

Large language models are typically deployed as monolithic systems, requiring the full model even when applications need only a narrow subset of capabilities, e.g., code, math, or domain-specific knowledge. Mixture-of-Experts (MoEs) seemingly offer a potential alternative by activating only a subset of experts per input, but in practice, restricting inference to a subset of experts for a given domain leads to severe performance degradation. This limits their practicality in memory-constrained settings, especially as models grow larger and sparser. We introduce EMO, an MoE designed for modularity-the independent use and composition of expert subsets-without requiring human-defined priors. Our key idea is to encourage tokens from similar domains to rely on similar experts. Since tokens within a document often share a domain, EMO restricts them to select experts from a shared pool, while allowing different documents to use different pools. This simple constraint enables coherent expert groupings to emerge during pretraining using document boundaries alone. We pretrain a 1B-active, 14B-total EMO on 1T tokens. As a full model, it matches standard MoE performance. Crucially, it enables selective expert use: retaining only 25% (12.5%) of experts incurs just a 1% (3%) absolute drop, whereas standard MoEs break under the same setting. We further find that expert subsets in EMO specialize at semantic levels (e.g., domains such as math or code), in contrast to the low-level syntactic specialization observed in standard MoEs. Altogether, our results demonstrate a path toward modular, memory-efficient deployment of large, sparse models and open new opportunities for composable architectures.

Channel state information (CSI)-based human activity recognition (HAR) is vulnerable to performance degradation under domain shifts across varying physical environments. Continual learning (CL) offers a principled way to learn new domains sequentially while preserving past knowledge, but existing CL solutions for CSI-based HAR scale poorly with accumulating domains, rely on a large replay buffer, or incur linearly growing inference cost. In this letter, we propose Scene-Adaptive Mixture of Experts with Clustered Specialists (SAMoE-C), which formulates cross-domain CSI-based HAR as a mixture-of-experts system that enables scene-specific adaptation, via an attention-based semantic router that activates only selected experts for each input. Moreover, we develop a novel training protocol, which requires only a tiny replay buffer for stabilizing domain discrimination of the router. Experimental results on a four-scene CSI dataset demonstrate that SAMoE-C approaches the state-of-the-art accuracy, while maintaining a significantly lower inference cost. By jointly combining modular experts, selective activation with router and a lightweight training protocol, SAMoE-C enables scalable cross-domain CSI-based HAR deployment with low training overhead and high computational efficiency in real-world settings.

Vision-language-action (VLA) models inherit rich visual-semantic priors from pre-trained vision-language backbones, but adapting them to robotic control remains challenging. Full fine-tuning (FFT) is prone to overfitting on downstream robotic data and catastrophic forgetting of pretrained vision-language capabilities. Parameter-efficient fine-tuning (PEFT) better preserves pre-trained knowledge, yet existing PEFT methods still struggle to adapt effectively to robot control tasks. To address this gap, we propose VLA-GSE, a parameter-efficient VLA fine-tuning framework that improves control adaptation while retaining PEFT's knowledge preservation advantage. Specifically, VLA-GSE (Generalized and Specialized Experts) is initialized by spectrally decomposing the frozen backbone, assigning leading singular components to generalized experts (shared experts) and disjoint residual components to specialized experts (routed experts). This decomposition improves adaptation capacity under a fixed trainable-parameter budget. Under a comparable parameter budget, VLA-GSE updates only 2.51% of the full model parameters and consistently outperforms strong FFT and PEFT baselines. It achieves 81.2% average zero-shot success on LIBERO-Plus, preserves pre-trained VLM capability comparably to LoRA on multimodal understanding benchmarks, and improves real-world manipulation success under multiple distribution shifts. Code is available at: https://github.com/YuhuaJiang2002/VLA-GSE

Recent studies have explored large language models for time-series anomaly detection, yet existing approaches often rely on a single general-purpose model to directly infer anomaly indices or intervals, limiting controllability, interpretability, and reliability for complex anomaly patterns. We propose SAGE (Specialized Analyzer Group for Expert-like Detection), a multi-agent framework for structured anomaly diagnosis in univariate time series. It decomposes anomaly analysis into four specialized Analyzers for point, structural, seasonal, and pattern anomalies. Each Analyzer applies family-specific numerical tools and diagnostic visualizations to generate evidence, while an evidence-grounded Detector consolidates the evidence into confidence-scored anomaly records with intervals and candidate types. A Supervisor then converts these structured records into analyst-facing diagnostic reports. SAGE further constructs synthetic in-context examples from normal-reference training segments, without using real anomalous segments or anomaly-type labels as in-context examples. Across three benchmarks, SAGE achieves the best average performance among strong ML/DL and language-model-based baselines. Ablation studies and human evaluation further show that the proposed framework improves detection reliability and the practical usefulness of diagnostic outputs.

Resource-efficient machine learning increasingly uses sparse Mixture-of-Experts (MoE) architectures, where the gate acts as both a learning component and a routing interface controlling computation, communication, and accuracy. Motivated by finite-rate interpretations of MoE gating, we treat the gate as a stochastic channel and use $I(X;T)$ to quantify the routing information available to the selected expert. To make the associated information quantities tractable beyond synthetic examples, we develop a finite-bank MNIST construction using pretrained CNN experts and a discrete, data-dependent selection rule. Since the selected model belongs to a finite candidate set, the algorithmic mutual information $I(S;W)$ admits a closed-form discrete-entropy estimator from the empirical posterior $q(W|S)$. Sweeping a data-dependence parameter $α$, we observe that $\widehat I(S;W)$ monotonically tracks the generalization gap, while the Xu-Raginsky bound exhibits the expected looseness. We also compare with a uniform union-bound baseline and introduce an empirical estimator of $I(X;T)$ together with a Blahut-Arimoto procedure for tracing an accuracy-rate curve over the expert bank. The proposed framework provides a practical tool for analyzing resource-aware MoE inference systems and for interpreting $I(X;T)$ and $D(R_g)$ as design proxies for efficient expert routing.

Mechanical equipment forms the critical backbone of modern industrial production, yet domain shift severely limits the generalization of deep learning based fault diagnosis models across different equipment and operating conditions.Inspired by the success of foundation models in achieving zero-shotgeneralization, we propose YOTOnet (You Only Train Once), a novel architecture specifically designed for cross-domain fault diagnosis in mechanical equipment.YOTOnet comprises three core components: (1) a physics-aware Invariant Feature Distiller that extracts domain-agnostic representations using multi-scale dilated convolutions and FFT-based time-frequency fusion,(2) Domain-Conditioned Sparse Experts (DC-MoE) that adaptively route inputs to specialized processors via learned gating without external meta-data, and (3) a dual-head classification system with auxiliary supervision.Extensive validation on five public bearing datasets (CWRU, MFPT, XJTU,OTTAWA, HUST) through 30 cross-dataset protocols demonstrates the superiority of YOTOnet compared with other state-of-the-art methods. Critically, we observe a clear scaling effect-average test F1 improves from 0.5339(1 training dataset) to 0.705 (4 datasets), with a clear gain when moving from 3 to 4 datasets. These findings provide empirical evidence that foundation model principles can enable robust, train-once deployment for industrial fault diagnosis.

LoRA fine-tuning of diffusion transformers (DiT) on multi-style data suffers from \emph{style bleed}: a single low-rank residual cannot represent several distinct artist fingerprints, and the optimizer converges to their average. Mixture-of-experts LoRA in the HydraLoRA style replaces the up-projection with $E$ heads under a router, but when every expert is zero-initialized the router receives identical gradient from each head and remains at the uniform prior. The experts then evolve permutation-symmetrically, and the network trains as a single rank-$r$ LoRA at $E{\times}$ the cost. We present \textbf{Ortho-Hydra}, a re-parameterisation that combines an OFT-style Cayley-orthogonal shared basis with per-expert \emph{disjoint output subspaces} carved from the top-$(Er)$ left singular vectors of the pretrained weight. Disjointness makes the router's per-expert score non-degenerate at step~$0$, so specialization receives gradient signal before any expert has trained. We test the predicted deadlock on a DiT pipeline by comparing two HydraLoRA baselines, a zero-initialized shared-basis variant and the original $σ{=}0.1$ Gaussian-jitter mitigation, against Ortho-Hydra under a matched optimiser, dataset, and step budget. Neither baseline leaves the uniform prior within the first $1\text{k}$ steps; Ortho-Hydra begins de-uniformising within the first few hundred. End-task generation quality on multi-style data is out of scope; we report the construction, the cold-start mechanism, and the routing dynamics it changes. Code: https://github.com/sorryhyun/anima_lora.

The rapidly expanding artificial intelligence (AI) industry has produced diverse yet powerful prediction tools, each with its own network architecture, training strategy, data-processing pipeline, and domain-specific strengths. These tools create new opportunities for semi-supervised inference, in which labeled data are limited and expensive to obtain, whereas unlabeled data are abundant and widely available. Given a collection of predictors, we treat them as a mixture of experts (MOE) and introduce an MOE-powered semi-supervised inference framework built upon prediction-powered inference (PPI). Motivated by the variance reduction principle underlying PPI, the proposed framework seeks the mixture of experts that achieves the smallest possible variance. Compared with standard PPI, the MOE-powered inference framework adapts to the unknown performance of individual predictors, benefits from their collective predictive power, and enjoys a best-expert guarantee. The framework is flexible and applies to mean estimation, linear regression, quantile estimation, and general M-estimation. We develop non-asymptotic theory for the MOE-powered inference framework and establish upper bounds on the coverage error of the resulting confidence intervals. Numerical experiments demonstrate the practical effectiveness of MOE-powered inference and corroborate our theoretical findings.

Hybrid-thinking language models expose explicit think and no-think modes, but current designs do not separate them cleanly. Even in no-think mode, models often emit long and self-reflective responses, causing reasoning leakage. Existing work reduces this issue through better data curation and multi-stage training, yet leakage remains because both modes are still encoded in the same feed-forward parameters. We propose Path-Lock Expert (PLE), an architecture-level solution that replaces the single MLP in each decoder layer with two semantically locked experts, one for think and one for no-think, while keeping attention, embeddings, normalization, and the language-model head shared. A deterministic control-token router selects exactly one expert path for the entire sequence, so inference preserves the dense model's per-token computation pattern and each expert receives mode-pure updates during supervised fine-tuning. Across math and science reasoning benchmarks, PLE maintains strong think performance while producing a substantially stronger no-think mode that is more accurate, more concise, and far less prone to reasoning leakage. On Qwen3-4B, for example, PLE reduces no-think reflective tokens on AIME24 from 2.54 to 0.39 and improves no-think accuracy from 20.67% to 40.00%, all while preserving think-mode performance. These results suggest that controllable hybrid thinking is fundamentally an architectural problem, and separating mode-specific feed-forward pathways is a simple and effective solution.

Municipal solid waste incineration (MSWI) converts urban waste to energy but simultaneously emits carbon dioxide, carbon monoxide and multiple regulated air pollutants whose formation is tightly coupled within a single combustion system. Controlling these emissions across a network of diverse facilities poses a fundamentally different challenge from optimising a single plant: data-driven models trained at one site capture local statistical patterns that rarely survive transfer to another, because they lack the physical constraints and regime-level structure needed to generalise. Here we show that shared emission-control relationships can be identified across heterogeneous MSWI plants when physical conservation laws, operating-regime heterogeneity and carbon-pollutant coupling are treated jointly. We develop a carbon-pollutant mixture-of-experts (CPMoE) model that routes process observations through regime-specific expert networks under conservation-based regularisation, and combine it with physics-informed transfer learning to adapt a reference model to new facilities. Across 13 plants, CPMoE predicts six major pollutants and a composite system-level risk index with source-domain R2 of 0.668-0.904 and 0.666-0.970, respectively; after transfer to 12 target plants these values remain 0.661-0.842 and 0.610-0.841. Expert-utilisation patterns show that adaptation proceeds through structured regime re-weighting rather than re-learning from scratch. Embedding the transferred model in an offline digital twin and screening candidate operating adjustments against historical process records yields consistent risk-index reductions of 3.6-6.3% with simultaneous pollutant co-reductions in 94-100% of evaluated samples. These findings suggest a practical route toward transferable, system-level decision support for carbon-pollutant co-control in heterogeneous waste-to-energy networks.

LoRA-MoE has emerged as an effective paradigm for parameter-efficient fine-tuning, combining the low training cost of LoRA with the increased adaptation capacity of Mixture-of-Experts (MoE). However, existing LoRA-MoE frameworks typically adopt a fixed and uniform expert configuration across heterogeneous Transformer modules (\eg, attention query/key projections and MLP gating networks), ignoring their distinct functional roles and capacity requirements. This design leads to localized over-provisioning, redundant trainable parameters, and unnecessary optimizer-state overhead. Moreover, prior methods enforce load balancing among experts throughout training. Although beneficial in the early stage, this constraint becomes restrictive once routing patterns stabilize, limiting expert specialization on downstream tasks. In this paper, we propose DMEP, a novel LoRA-MoE fine-tuning framework based on Dynamic Module-wise Expert Pruning. DMEP tracks expert utilization during training and physically removes low-utility experts on a per-module basis, yielding a more compact expert structure tailored to different modules. The pruned model then continues training without the load-balancing constraint, freeing the remaining experts to focus entirely on the downstream task and develop specialized expertise. By jointly adapting module-wise expert capacity and eliminating unnecessary balancing, DMEP improves both parameter efficiency and training efficiency. Extensive experiments on multiple reasoning benchmarks show that DMEP reduces trainable parameters by 35%--43% and improves training throughput by about 10%, while maintaining or surpassing the downstream reasoning accuracy of uniform LoRA-MoE baselines.

First-principles atomistic simulations are essential for understanding complex material phenomena but are fundamentally limited by their computational cost. While Machine Learning Interatomic Potentials (MLIPs) have drastically improved cost for a given accuracy, their inference cost remains a bottleneck for massive systems or long timescales. To address this, we introduce a multifidelity "Mixture-of-Experts" framework based on the E(3)-equivariant Allegro architecture. Our method spatially partitions the simulation domain into a chemically complex region (e.g., reactive interfaces) and a simple region (e.g., bulk lattice), assigning models of varying capacity to each. Among the challenges in such static domain decomposition, the mechanical mismatch between models at the interface is particularly critical, as it can generate artificial stress fields and instability. We address this challenge with a co-training strategy in which the loss function includes agreement constraints -- penalties on per-atom energy and force discrepancies between models evaluated on shared bulk environments -- forcing the independent models to learn a consistent physical description of the bulk material. We validate this approach on a realistic Pt+CO catalytic system, demonstrating that the co-trained models maintain exact energy conservation, align their bulk mechanical response (e.g., equation of state and bulk modulus), and achieve predictive accuracy comparable to a full high-fidelity simulation at more than twice the computational speed.

Despite MoE models leading many benchmarks, supervised fine-tuning (SFT) for the MoE architectures remains difficult because its router layers are fragile. Methods such as DenseMixer and ESFT mitigate router collapse with dense mixing or auxiliary load-balancing losses, but these introduce noisy gradients that often degrade performance. In preliminary experiments, we systematically pruned experts and observed that while certain super experts are activated far more frequently, discarding less used experts still leads to notable performance degradation. This suggests that even rarely activated experts encode non-trivial knowledge useful for downstream tasks. Motivated by this, we propose an auxiliary-loss-free MoE SFT framework that combines bias-driven sparsification with always-active gated condenser experts. Rather than enforcing balanced activation across all experts, our method encourages task-relevant experts to remain active while pushing long-tailed experts toward inactivity. The condenser experts provide a persistent, learnable pathway that alleviates gradient starvation and facilitates consolidation of information that would otherwise remain fragmented across sparsely activated experts. Analysis further suggest that this design better preserves long-tailed expert information under sparse routing. Experiments on large-scale MoE models demonstrate that our approach outperforms state-of-the-art SFT baselines such as DenseMixer and ESFT, achieving average gain of 2.5%+ on both mathematical reasoning and commonsenseQA benchmarks.

Robotic parkour provides a compelling benchmark for advancing locomotion over highly challenging terrain, including large discontinuities such as elevated steps. Recent approaches have demonstrated impressive capabilities, including dynamic climbing and jumping, but typically rely on sequential multilayer perceptron (MLP) architectures with densely activated layers. In contrast, sparsely gated mixture-of-experts (MoE) architectures have emerged in the large language model domain as an effective paradigm for improving scalability and performance by activating only a subset of parameters at inference time. In this work, we investigate the application of sparsely gated MoE architectures to vision-based robotic parkour. We compare control policies based on standard MLPs and MoE architectures under a controlled setting where the number of active parameters at inference time is matched. Experimental results on a real Unitree Go2 quadruped robot demonstrate clear performance gains, with the MoE policy achieving double the number of successful trials in traversing large obstacles compared to a standard MLP baseline. We further show that achieving comparable performance with a standard MLP requires scaling its parameter count to match that of the total MoE model, resulting in a 14.3% increase in computation time. These results highlight that sparsely gated MoE architectures provide a favorable trade-off between performance and computational efficiency, enabling improved scaling of control policies for vision-based robotic parkour. An anonymized link to the codebase is https://osf.io/v2kqj/files/github?view_only=7977dee10c0a44769184498eaba72e44.

Mixture-of-Experts (MoE) has become the dominant architecture for scaling large language models: frontier models routinely decouple total parameters from per-token computation through sparse expert routing. Scaling laws show that under fixed active computation, model quality scales predictably with total parameters, and MoEs realize this by increasing expert count. However, training large MoEs is expensive, as memory requirements and inter-device communication both scale with total parameter count. We propose expert upcycling, a method for progressively expanding MoE capacity by increasing the number of experts during continued pre-training (CPT). Given a trained E-expert model, the upcycling operator constructs an mE-expert model through expert duplication and router extension while holding top-K routing fixed, preserving per-token inference cost. Duplication provides a warm initialization: the expanded model inherits the source checkpoint's learned representations, starting from a substantially lower loss than random initialization. Subsequent CPT then breaks the symmetry among duplicated experts to drive specialization. We formalize the upcycling operator and develop a theoretical framework decomposing the quality gap into a capacity term and an initialization term. We further introduce utility-based expert selection, which uses gradient-based importance scores to guide non-uniform duplication, more than tripling gap closure when CPT is limited. In our 7B-13B total parameter experiments, the upcycled model matches the fixed-size baseline on validation loss while saving 32% of GPU hours. Comprehensive ablations across model scales, activation ratios, MoE architectures, and training budgets yield a practical recipe for deploying expert upcycling, establishing it as a principled, compute-efficient alternative to training large MoE models from scratch.

Neural NLI models overfit dataset artifacts instead of truly reasoning. A hypothesis-only model gets 57.7% in SNLI, showing strong spurious correlations, and 38.6% of the baseline errors are the result of these artifacts. We propose Product-of-Experts (PoE) training, which downweights examples where biased models are overconfident. PoE nearly preserves accuracy (89.10% vs. 89.30%) while cutting bias reliance by 4.71% (bias agreement 49.85% to 45%). An ablation finds lambda = 1.5 that best balances debiasing and accuracy. Behavioral tests still reveal issues with negation and numerical reasoning.

Automating medical reports for retinal images requires a sophisticated blend of visual pattern recognition and deep clinical knowledge. Current Large Vision-Language Models (LVLMs) often struggle in specialized medical fields where data is scarce, leading to models that overfit and miss subtle but critical pathologies. To address this, we introduce DREAM (Dynamic Retinal Enhancement with Adaptive Multi-modal Fusion), a novel framework for high-fidelity medical report generation that excels even with limited data. DREAM employs a unique two-stage fusion mechanism that intelligently integrates visual data with clinical keywords curated by ophthalmologists. First, the Abstractor module maps image and keyword features into a shared space, enhancing visual data with pathology-relevant insights. Next, the Adaptor performs adaptive multi-modal fusion, dynamically weighting the importance of each modality using learnable parameters to create a unified representation. To ensure the model's outputs are semantically grounded in clinical reality, a Contrastive Alignment module aligns these fused representations with ground-truth medical reports during training. By combining medical expertise with an efficient fusion strategy, DREAM sets a new state-of-the-art on the DeepEyeNet benchmark, achieving a BLEU-4 score of 0.241, and further demonstrates strong generalization to the ROCO dataset.

Visual Question Answering (VQA) requires models to identify the correct answer options based on both visual and textual evidence. Recent Mixture-of-Experts (MoE) methods improve option reasoning by grouping similar concepts or routing based on examples. However, unstable routing can lead to inconsistent expert selection in the same question type, while overly stable routing may reduce flexibility. To address this, we propose Concept-Guided Routing framework (CoGR-MoE), which incorporates semantics of the answer options to guide expert selection in the training phase. Next, option features are used to reweight the selected experts, producing discriminative representations for each candidate option. These option-level representations are further used for option comparison and optimized via contrastive learning. The experimental results indicate that CoGR-MoE delivers strong performance across multiple VQA tasks, demonstrating the effectiveness of our approach.

Recent deployments of large language models (LLMs) as autonomous trading agents raise questions about whether financial decision-making competence generalizes beyond specific market patterns and how it should be trained and evaluated in noisy markets lacking ground truth. We propose a structured framework for training and evaluating such models. Central to our approach is a curated, multiple-choice question (MCQ) dataset derived from classic textbooks and historical markets, verified by an AI committee, enriched with structured reasoning traces, and augmented to reduce shortcut learning. To evaluate whether performance on isolated MCQs generalizes to real-world trading, we introduce a two-stage protocol combining test-set evaluation with an MCQ-based chronological trading simulation. Extensive evaluations across market regimes provide statistically robust evidence that open models trained with our framework exhibit competitive, risk-aware behavior over time, outperform open-source baselines, and approach frontier-model performance at smaller scale. We release the dataset and evaluation framework to support further research.

Vision Language Models (VLMs) have been applied to several specific domains and have shown strong problem-solving capabilities. However, astronomical imaging, a quite complex problem involving multidisciplinary knowledge and several subtasks, has not been adequately studied. Due to the complexity of the astronomical imaging process, both world-class astronomical organizations, such as NASA, and expert enthusiasts devote a great deal of time and effort. This is because the processes in astronomical imaging have complex underlying correlations that significantly influence one another, making the quality diagnosis and error localization of astronomical images challenging. To address this problem, we propose AstroVLM, a collaborative multi-agent system for diagnosing the quality of astronomical images. Experiment results show that AstroVLM outperforms all baselines on real-world astronomical imaging quality diagnosis tasks, providing a reference for language models to handle complicated multi-process tasks.

Human data annotation, especially when involving experts, is often treated as an objective reference. However, many annotation tasks are inherently subjective, and annotators' judgments may evolve over time. This study investigates changes in the quality of annotators' work from a competence perspective during a process of social influence recognition. The study involved 25 annotators from five different groups, including both experts and non-experts, who annotated a dataset of 1,021 dialogues with 20 social influence techniques, along with intentions, reactions, and consequences. An initial subset of 150 texts was annotated twice - before and after the main annotation process - to enable comparison. To measure competence shifts, we combined qualitative and quantitative analyses of the annotated data, semi-structured interviews with annotators, self-assessment surveys, and Large Language Model training and evaluation on the comparison dataset. The results indicate a significant increase in annotators' self-perceived competence and confidence. Moreover, observed changes in data quality suggest that the annotation process may enhance annotator competence and that this effect is more pronounced in expert groups. The observed shifts in annotator competence have a visible impact on the performance of LLMs trained on their annotated data.

Multimodal recommenders combine collaborative behavior with visual and textual item evidence, whose usefulness varies across user-item interactions. Independently trained source-specific diagnostic probes partition held-out interactions into behavior-, appearance-, semantics-, and mixed-evidence regimes across five benchmarks, within which capacity-matched fixed fusion rules exhibit systematic regime-dependent performance crossovers. This diagnostic observation motivates MAGNET, a multimodal graph recommender with two core mechanisms. First, a calibrated expert bank organizes trainable experts by anchor source (behavior, appearance, or semantics) and fusion family (dominant, balanced, or complementary), while an interaction-conditioned router selects a sparse composition using all three evidence sources. Second, an entropy-triggered, coverage-aware progressive schedule decouples population-level routing-mass coverage from per-instance decisiveness, transitioning from broad routing exploration to confident specialization once sufficient coverage is sustained, while preserving coverage thereafter. MAGNET separately encodes the observed interaction graph and a filtered content-induced structural view, applying cross-view alignment after independent propagation. We evaluate MAGNET on four Amazon domains and the non-Amazon MicroLens-100K benchmark, including tail-item and low-history warm-start user evaluation alongside matched expert-design controls. Under the shared-feature, fixed-split protocol, MAGNET-DV exceeds the strongest protocol-compatible non-MAGNET baseline in every reported main-table cell, supported by seed-wise difference tests. Both the fixed and KL-anchored structured variants achieve higher five-seed mean NDCG@20 than matched homogeneous and free-mixture alternatives across all five datasets. MAGNET supports route-level diagnostics through explicit expert semantics.

In current Mixture of Experts (MoE) architectures, linear memory scaling is present, the memory grows as the number of experts increases. $N$ independent expert weight matrices require $\mathcal{O}(N \cdot d^2)$ memory which exceeds the memory budget of edge devices. Current compression methods like quantization, pruning, and low-rank factorization reduce constant factors, but the scaling bottleneck is still unresolved. We introduce ButterflyMoE, a method which treats experts not as independent matrices but as geometric reorientations of a shared quantized substrate. Diversity amongst the experts arises from viewing different angles of the shared capacity and not from redundant storage. Learned rotations are applied to a shared ternary prototype. With this, each expert yields $\mathcal{O}(d^2 + N \cdot d \log d)$ memory-reducing per-expert cost from $\mathcal{O}(d^2)$ to $\mathcal{O}(d \log d)$. The key insight is that training these rotations with quantization reduces activation outliers and stabilizes extreme low-bit training where other static methods collapse. Across language modeling benchmarks, ButterflyMoE achieves 80$\times$ memory reduction at 8 experts with a highly favorable memory-accuracy tradeoff.At this 80x compression ButterflyMoE outperforms an equal memory dense baseline, showing that orbital parameterization extracts fundamentally more utility per byte. When scaled up to 256 experts, ButterflyMoE asymptotically compresses the memory by 150 $\times$. ButterflyMoE reduces the constant factor of linear scaling with compression ratio growing with the expert count.

Enabling lifelong learning in LLMs demands resolving the stability-plasticity dilemma (i.e., models must incorporate new knowledge without overwriting prior representations) while maintaining scalability under bounded parameter growth. Existing PEFT methods fail to satisfy this triad; shared-parameter approaches suffer from catastrophic interference, while task-isolated expansions preclude knowledge transfer and scale linearly. We propose Mixtures of SubExperts (MoSEs), a modular and sparse framework that factorizes model capacity into reusable, compositional primitives. MoSEs augment transformer layers with lightweight SubExperts and a learned sub-routing function that dynamically selects and composes a sparse subset of modules conditioned on task inputs. This induces a structured decomposition of the parameter space where knowledge is localized yet accessible, mitigating interference while preserving reuse. Specifically, MoSEs balance the dilemma via three pillars: (i) stability by isolating knowledge within sparsely activated modules, (ii) plasticity through routing-driven recombination and selective expansion, and (iii) scalability via sublinear growth in effective capacity. Notably, the routing mechanism enables compositional generalization, allowing new tasks to be represented as combinations of previously acquired sub-functions. We empirically validate MoSEs on TRACE and SuperNI, showing reduced forgetting, improved forward transfer, and better parameter efficiency over strong PEFT baselines. MoSEs establish a new Pareto frontier, achieving state-of-the-art performance while maintaining strict parameter budgets. Our results suggest that modular sparsity and compositional routing are key inductive biases for building foundation models that continually learn without saturation.

Expert domain writing, such as scientific writing, typically demands extensive domain knowledge. Although large language models (LLMs) show promising potential in this task, evaluating the quality of automatically generated scientific writing is a crucial open issue, as it requires knowledge of domain-specific criteria and the ability to discern expert preferences. Conventional automatic evaluation metrics and LLM-as-a-judge systems, primarily designed for mainstream NLP tasks, are insufficient to grasp expert preferences and domain-specific quality standards. To address this gap and support realistic human-AI collaborative writing, we focus on related work generation, one of the most challenging scientific tasks, as an exemplar. We propose GREP, a multi-turn evaluation framework that integrates classical related work evaluation criteria with expert-specific preferences. GREP decomposes the evaluation into smaller fine-grained dimensions. This localized evaluation is further augmented with contrastive examples to provide detailed contextual guidance for the evaluation dimensions. Empirical investigation reveals that GREP is able to assess the quality of related work sections in a much more robust manner compared to standard LLM judges, reflects natural scenarios of scientific writing, and bears a strong correlation with the assessment of human experts. We also observe that generations from state-of-the-art LLMs struggle to satisfy validation constraints of a suitable related work section.

Graph-based multi-task learning at billion-scale presents a significant challenge, as different tasks correspond to distinct billion-scale graphs. Traditional multi-task learning methods often neglect these graph structures, relying solely on individual user and item embeddings. However, disregarding graph structures overlooks substantial potential for improving performance. In this paper, we introduce the Macro Graph of Experts (MGOE) framework, the first approach capable of leveraging macro graph embeddings to capture task-specific macro features while modeling the correlations between task-specific experts. Specifically, we propose the concept of a Macro Graph Bottom, which, for the first time, enables multi-task learning models to incorporate graph information effectively. We design the Macro Prediction Tower to dynamically integrate macro knowledge across tasks. MGOE has been deployed at scale, powering multi-task learning for a leading billion-scale recommender system, Alibaba. Extensive offline experiments conducted on three public benchmark datasets demonstrate its superiority over state-of-the-art multi-task learning methods, establishing MGOE as a breakthrough in multi-task graph-based recommendation. Furthermore, online A/B tests confirm the superiority of MGOE in billion-scale recommender systems.

Mental health is a growing global concern, prompting interest in AI-driven solutions to expand access to psychosocial support. \emph{Peer support}, grounded in lived experience, offers a valuable complement to professional care. However, variability in training, effectiveness, and definitions raises concerns about quality, consistency, and safety. Large Language Models (LLMs) present new opportunities to enhance peer support interactions, particularly in real-time, text-based interactions. We present and evaluate an AI-supported system with an LLM-simulated distressed client (\client{}), context-sensitive LLM-generated suggestions (\suggestions{}), and real-time emotion visualisations. 2 mixed-methods studies with 12 peer supporters and 6 mental health professionals (i.e., experts) examined the system's effectiveness and implications for practice. Both groups recognised its potential to enhance training and improve interaction quality. However, we found a key tension emerged: while peer supporters engaged meaningfully, experts consistently flagged critical issues in peer supporter responses, such as missed distress cues and premature advice-giving. This misalignment highlights potential limitations in current peer support training, especially in emotionally charged contexts where safety and fidelity to best practices are essential. Our findings underscore the need for standardised, psychologically grounded training, especially as peer support scales globally. They also demonstrate how LLM-supported systems can scaffold this development--if designed with care and guided by expert oversight. This work contributes to emerging conversations on responsible AI integration in mental health and the evolving role of LLMs in augmenting peer-delivered care.

Labeling corpora constitutes a bottleneck to create models for new tasks or domains. Large language models mitigate the issue with automatic corpus labeling methods, particularly for categorical annotations. Some NLP tasks such as emotion intensity prediction, however, require text regression, but there is no work on automating annotations for continuous label assignments. Regression is considered more challenging than classification: The fact that humans perform worse when tasked to choose values from a rating scale lead to comparative annotation methods, including best-worst scaling. This raises the question if large language model-based annotation methods show similar patterns, namely that they perform worse on rating scale annotation tasks than on comparative annotation tasks. To study this, we automate emotion intensity predictions and compare direct rating scale predictions, pairwise comparisons and best-worst scaling. We find that the latter shows the highest reliability. A transformer regressor fine-tuned on these data performs nearly on par with a model trained on the original manual annotations.

Institutions of higher learning, research institutes and other scientific research units have abundant scientific and technological resources of experts and scholars, and these talents with great scientific and technological innovation ability are an important force to promote industrial upgrading. The scientific and technological resources of experts and scholars are mainly composed of basic attributes and scientific research achievements. The basic attributes include information such as research interests, institutions, and educational work experience. However, due to information asymmetry and other reasons, the scientific and technological resources of experts and scholars cannot be connected with the society in a timely manner, and social needs cannot be accurately matched with experts and scholars. Therefore, it is very necessary to build an expert and scholar information database and provide relevant expert and scholar retrieval services. This paper sorts out the related research work in this field from four aspects: text relation extraction, text knowledge representation learning, text vector retrieval and visualization system.

Mixtures of experts (MoE) are conditional mixture models in which both the mixing proportions and the component densities depend on the predictors, and are widely used for regression, classification and model-based clustering of heterogeneous data. Fitting MoE by maximum likelihood becomes unstable, and sometimes infeasible, when the predictors are numerous or correlated. We propose a regularized maximum likelihood framework for simultaneous parameter estimation and feature selection in MoE whose experts belong to the generalized linear model family, covering Gaussian, Poisson and multinomial responses within a single formulation. Sparsity is induced in both the gating network and the experts through $\ell_1$ penalties, and the penalized log-likelihood is maximized by a proximal Newton-EM algorithm whose M-step reduces to weighted Lasso problems with closed-form coordinate-ascent updates. Unlike existing penalized MoE procedures, the algorithm requires neither a local quadratic approximation of the penalty nor any matrix inversion, it returns exactly sparse estimates without thresholding, and a proximal Newton-type variant guarantees a monotone increase of the penalized objective at every iteration. On simulated data and five real data sets, the method recovers the actual sparsity support and delivers prediction and clustering accuracy that is competitive with, and often better than, state-of-the-art regularized MoE. The source codes of our developed algorithms and their documentation are publicly available on Github at https://github.com/nv-thin/GLM-RMoE.

For effective deployment in real-world environments, humanoid robots must autonomously navigate a diverse range of complex terrains with abrupt transitions. While the Vanilla mixture of experts (MoE) framework is theoretically capable of modeling diverse terrain features, in practice, the gating network exhibits nearly uniform expert activations across different terrains, weakening the expert specialization and limiting the model's expressive power. To address this limitation, we introduce CMoE, a novel single-stage reinforcement learning framework that integrates contrastive learning to refine expert activation distributions. By imposing contrastive constraints, CMoE maximizes the consistency of expert activations within the same terrain while minimizing their similarity across different terrains, thereby encouraging experts to specialize in distinct terrain types. We validated our approach on the Unitree G1 humanoid robot through a series of challenging experiments. Results demonstrate that CMoE enables the robot to traverse continuous steps up to 20 cm high and gaps up to 80 cm wide, while achieving robust and natural gait across diverse mixed terrains, surpassing the limits of existing methods. To support further research and foster community development, we release our code publicly.

Mixture-of-Experts (MoE) models activate few experts per token, yet their full expert sets can exceed GPU memory and require repeated weight transfers during decoding. We formulate expert-cache management as a model-side algorithmic problem and propose cache-aware post-training that jointly adapts the MoE backbone and lightweight auxiliary routers while preserving the native inference-time Top-K rule. The update-only Temporal Router learns same-layer retention across tokens without proactive loading. The full Spatio-Temporal Router adds a Spatio Router that uses the causal predecessor's hidden state to refine the temporal cache before target-layer access. We evaluate both modes on Qwen3 and GPT-OSS across GSM8K, MATH, and CommonsenseQA. Temporal Router consistently improves hit rate and reduces expert-weight traffic over matched LM-only baselines. On Qwen3, the full mode improves adjusted hit rate by 1.15--18.03 points and reduces traffic by 4.6--53.3% relative to the strongest evaluated prefetching baseline; GPT-OSS results are competitive but task-dependent. Auxiliary-only training preserves baseline accuracy but yields modest coverage gains; joint post-training achieves substantially higher coverage. Sensitivity analyses distinguish the effects of cache capacity, refinement budget, and cache-loss weight on coverage, traffic, and quality.

Mixture-of-Experts (MoE) has become a practical architecture for scaling LLM capacity while keeping per-token compute modest, but deploying MoE models on a single, memory-limited GPU remains difficult because expert weights dominate the HBM footprint. Existing expert offloading and prefetching systems reduce the resident set, yet they often pay expert-loading costs on the critical path when activation becomes dense. Post-training quantization (PTQ) lowers the footprint without transfers, but prevailing pipelines fix expert bit-widths offline and assume routing remains stable, even though MoE expert utilization is heavy-tailed and the hot set can shift across workloads. We present DynaExq, a runtime-aware mixed-precision serving system that treats single-GPU MoE inference under a hard HBM envelope as an online, budget-constrained precision allocation problem. The key insight is to keep the experts that dominate runtime traffic resident at higher precision, while maintaining a low-precision fallback for the remaining experts, so the system can reduce transfer volume and avoid the waiting latency that limits offloading and prefetching under dense activation. DynaExq estimates long-horizon expert hotness from router traces, selects a per-layer high-precision resident set via a budget-feasible top-$n$ rule, and applies promotions and demotions asynchronously through stable expert handles so the forward pass always executes on a fully materialized expert version. Across Qwen3-MoE-30B/80B and six benchmarks, DynaExq improves accuracy over static PTQ on Qwen3-80B (73.09% to 77.57%) under comparable device-memory budgets and achieves up to 2.73x higher throughput than offloading/prefetch baselines at batch size 32.