# Research Papers — 2026-01 · bundle page 30 of 189

> Full-text markdown of papers #1451–1500 of 9403 published in 2026-01, newest first. AI-readable mirrors of the full 9403-paper 2026-01 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-01/page/30
[← Previous bundle](https://aipod.dev/papers/archive/2026-01/page/29.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-01/page/31.md)

### NEXUS: Bit-Exact ANN-to-SNN Equivalence via Neuromorphic Gate Circuits with Surrogate-Free Training

Authors: Zhengzheng Tang
Published: 2026-01-29
Categories: cs.NE, cs.AI
arXiv: https://arxiv.org/abs/2601.21279v2

Spiking Neural Networks (SNNs) promise energy-efficient computing through event-driven sparsity, yet all existing approaches sacrifice accuracy by approximating continuous values with discrete spikes. We propose NEXUS, a framework that achieves bit-exact ANN-to-SNN equivalence -- not approximate, but mathematically identical outputs. Our key insight is constructing all arithmetic operations, both linear and nonlinear, from pure IF neuron logic gates that implement IEEE-754 compliant floating-point arithmetic. Through spatial bit encoding (zero encoding error by construction), hierarchical neuromorphic gate circuits (from basic logic gates to complete transformer layers), and surrogate-free STE training (exact identity mapping rather than heuristic approximation), NEXUS produces outputs identical to standard ANNs up to machine precision. Experiments on models up to LLaMA-2 70B demonstrate identical task accuracy (0.00% degradation) with mean ULP error of only 6.19, while achieving 27-168,000$\times$ energy reduction on neuromorphic hardware. Crucially, spatial bit encoding's single-timestep design renders the framework inherently immune to membrane potential leakage (100% accuracy across all decay factors $β\in(0.1,1.0)$), while tolerating synaptic noise up to $σ=0.2$ with >98% gate-level accuracy.

---

### GeoRC: A Benchmark for Geolocation Reasoning Chains

Authors: Mohit Talreja, Joshua Diao, Jim Thannikary James, Radu Casapu, Tejas Santanam, Ethan Mendes, Alan Ritter, Wei Xu, James Hays
Published: 2026-01-29
Categories: cs.CV, cs.AI, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2601.21278v2

Vision Language Models (VLMs) are good at recognizing the global location of a photograph -- their geolocation prediction accuracy rivals the best human experts. But many VLMs are startlingly bad at \textit{explaining} which image evidence led to their prediction, even when their location prediction is correct. In this paper, we introduce GeoRC, the first benchmark for geolocation reasoning chains sourced directly from Champion-tier GeoGuessr experts, including the reigning world champion. This benchmark consists of 800 ``ground truth'' reasoning chains across 500 query scenes from GeoGuessr maps, with expert chains addressing hundreds of different discriminative attributes, such as soil properties, architecture, and license plate shapes. We evaluate LLM-as-a-judge and VLM-as-a-judge strategies for scoring VLM-generated reasoning chains against our expert reasoning chains and find that Qwen 3 LLM-as-a-judge correlates best with human-expert scoring. Our benchmark reveals that while large, closed-source VLMs such as Gemini and GPT 5 rival human experts at predicting locations, they still lag behind human experts when it comes to producing auditable reasoning chains. Small open-weight VLMs such as Llama and Qwen catastrophically fail on our benchmark -- they perform only slightly better than a baseline in which an LLM hallucinates a reasoning chain with oracle knowledge of the photo location but \textit{no visual information at all}. We believe the gap between human experts and VLMs on this task points to VLM limitations at extracting fine-grained visual attributes from high resolution images. We open source our benchmark for the community to use.

---

### More Code, Less Reuse: Investigating Code Quality and Reviewer Sentiment towards AI-generated Pull Requests

Authors: Haoming Huang, Pongchai Jaisri, Shota Shimizu, Lingfeng Chen, Sota Nakashima, Gema Rodríguez-Pérez
Published: 2026-01-29
Categories: cs.SE, cs.AI, cs.HC
arXiv: https://arxiv.org/abs/2601.21276v1

Large Language Model (LLM) Agents are advancing quickly, with the increasing leveraging of LLM Agents to assist in development tasks such as code generation. While LLM Agents accelerate code generation, studies indicate they may introduce adverse effects on development. However, existing metrics solely measure pass rates, failing to reflect impacts on long-term maintainability and readability, and failing to capture human intuitive evaluations of PR. To increase the comprehensiveness of this problem, we investigate and evaluate the characteristics of LLM to know the pull requests' characteristics beyond the pass rate. We observe the code quality and maintainability within PRs based on code metrics to evaluate objective characteristics and developers' reactions to the pull requests from both humans and LLM's generation. Evaluation results indicate that LLM Agents frequently disregard code reuse opportunities, resulting in higher levels of redundancy compared to human developers. In contrast to the quality issues, our emotions analysis reveals that reviewers tend to express more neutral or positive emotions towards AI-generated contributions than human ones. This disconnect suggests that the surface-level plausibility of AI code masks redundancy, leading to the silent accumulation of technical debt in real-world development environments. Our research provides insights for improving human-AI collaboration.

---

### Lightweight High-Fidelity Low-Bitrate Talking Face Compression for 3D Video Conference

Authors: Jianglong Li, Jun Xu, Bingcong Lu, Zhengxue Cheng, Hongwei Hu, Ronghua Wu, Li Song
Published: 2026-01-29
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.21269v1

The demand for immersive and interactive communication has driven advancements in 3D video conferencing, yet achieving high-fidelity 3D talking face representation at low bitrates remains a challenge. Traditional 2D video compression techniques fail to preserve fine-grained geometric and appearance details, while implicit neural rendering methods like NeRF suffer from prohibitive computational costs. To address these challenges, we propose a lightweight, high-fidelity, low-bitrate 3D talking face compression framework that integrates FLAME-based parametric modeling with 3DGS neural rendering. Our approach transmits only essential facial metadata in real time, enabling efficient reconstruction with a Gaussian-based head model. Additionally, we introduce a compact representation and compression scheme, including Gaussian attribute compression and MLP optimization, to enhance transmission efficiency. Experimental results demonstrate that our method achieves superior rate-distortion performance, delivering high-quality facial rendering at extremely low bitrates, making it well-suited for real-time 3D video conferencing applications.

---

### Reinforcement Learning from Meta-Evaluation: Aligning Language Models Without Ground-Truth Labels

Authors: Micah Rentschler, Jesse Roberts
Published: 2026-01-29
Categories: cs.NE
arXiv: https://arxiv.org/abs/2601.21268v1

Most reinforcement learning (RL) methods for training large language models (LLMs) require ground-truth labels or task-specific verifiers, limiting scalability when correctness is ambiguous or expensive to obtain. We introduce Reinforcement Learning from Meta-Evaluation (RLME), which optimizes a generator using reward derived from an evaluator's answers to natural-language meta-questions (e.g., "Is the answer correct?" or "Is the reasoning logically consistent?"). RLME treats the evaluator's probability of a positive judgment as a reward and updates the generator via group-relative policy optimization, enabling learning without labels. Across a suite of experiments, we show that RLME achieves accuracy and sample efficiency comparable to label-based training, enables controllable trade-offs among multiple objectives, steers models toward reliable reasoning patterns rather than post-hoc rationalization, and generalizes to open-domain settings where ground-truth labels are unavailable, broadening the domains in which LLMs may be trained with RL.

---

### Model-Free Neural Filtering: A Comparison with Classical Filters in Nonlinear Systems

Authors: Zhuochen Liu, Hans Walker, Rahul Jain
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.21266v3

Neural network models are increasingly used for state estimation in control and decision-making, yet it remains unclear to what extent they behave as principled filters in nonlinear dynamical systems. Unlike classical filters, which rely on explicit dynamics and noise models, neural estimators can be trained purely from data. We present a systematic comparison between model-free neural estimators and classical filtering methods across multiple nonlinear scenarios. On the neural side, we evaluate Transformer-based models, recurrent neural networks, and state-space models; on the classical side, we compare against particle filters and nonlinear Kalman filters. Results show that structured state-space models (SSMs), in particular Mamba and Mamba-2, are consistently strong among neural estimators. They approach strong classical filters in several nonlinear systems and outperform weaker classical baselines without access to system models, while the evaluated neural implementations achieve substantially higher inference throughput on the tested hardware. Accurate model-based filters can still dominate when their assumptions are well matched. We attribute the relative strength of SSMs to filtering-aligned inductive bias: recursive latent-state updates make them structurally closer to classical filters under fixed parameter budgets, finite data, and long-horizon evaluation.

---

### CausalEmbed: Auto-Regressive Multi-Vector Generation in Latent Space for Visual Document Embedding

Authors: Jiahao Huo, Yu Huang, Yibo Yan, Ye Pan, Kening Zheng, Wei-Chieh Huang, Yi Cao, Mingdong Ou, Philip S. Yu, Xuming Hu
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.21262v4

Although Multimodal Large Language Models (MLLMs) have shown remarkable potential in Visual Document Retrieval (VDR) through generating high-quality multi-vector embeddings, the substantial storage overhead caused by representing a page with thousands of visual tokens limits their practicality in real-world applications. To address this challenge, we propose an auto-regressive generation approach, CausalEmbed, for constructing multi-vector embeddings. By incorporating iterative margin loss during contrastive training, CausalEmbed encourages the embedding models to learn compact and well-structured representations. Our method enables efficient VDR tasks using only dozens of visual tokens, achieving a 30-155x reduction in token count while maintaining highly competitive performance across various backbones and benchmarks. Theoretical analysis and empirical results demonstrate the unique advantages of auto-regressive embedding generation in terms of training efficiency and scalability at test time. As a result, CausalEmbed introduces a flexible test-time scaling strategy for multi-vector VDR representations and sheds light on the generative paradigm within multimodal document retrieval. Our code is available at https://github.com/Z1zs/Causal-Embed.

---

### Music Plagiarism Detection: Problem Formulation and a Segment-based Solution

Authors: Seonghyeon Go, Yumin Kim
Published: 2026-01-29
Categories: cs.SD, cs.AI, cs.LG, eess.AS
arXiv: https://arxiv.org/abs/2601.21260v2

Recently, the problem of music plagiarism has emerged as an even more pressing social issue. As music information retrieval research advances, there is a growing effort to address issues related to music plagiarism. However, many studies, including our previous work, have conducted research without clearly defining what the music plagiarism detection task actually involves. This lack of a clear definition has slowed research progress and made it hard to apply results to real-world scenarios. To fix this situation, we defined how Music Plagiarism Detection is different from other MIR tasks and explained what problems need to be solved. We introduce the Similar Music Pair dataset to support this newly defined task. In addition, we propose a method based on segment transcription as one way to solve the task. Our demo and dataset are available at https://github.com/Mippia/ICASSP2026-MPD.

---

### MoCo: A One-Stop Shop for Model Collaboration Research

Authors: Shangbin Feng, Yuyang Bai, Ziyuan Yang, Yike Wang, Zhaoxuan Tan, Jiajie Yan, Zhenyu Lei, Wenxuan Ding, Weijia Shi, Haojin Wang, Zhenting Qi, Yuru Jiang et al.
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.21257v2

Advancing beyond single monolithic language models (LMs), recent research increasingly recognizes the importance of model collaboration, where multiple LMs collaborate, compose, and complement each other. Existing research on this topic has mostly been disparate and disconnected, from different research communities, and lacks rigorous comparison. To consolidate existing research and establish model collaboration as a school of thought, we present MoCo: a one-stop Python library of executing, benchmarking, and comparing model collaboration algorithms at scale. MoCo features 26 model collaboration methods, spanning diverse levels of cross-model information exchange such as routing, text, logit, and model parameters. MoCo integrates 25 evaluation datasets spanning reasoning, QA, code, safety, and more, while users could flexibly bring their own data. Extensive experiments with MoCo demonstrate that most collaboration strategies outperform models without collaboration in 61.0% of (model, data) settings on average, with the most effective methods outperforming by up to 25.8%. We further analyze the scaling of model collaboration strategies, the training/inference efficiency of diverse methods, highlight that the collaborative system solves problems where single LMs struggle, and discuss future work in model collaboration, all made possible by MoCo. We envision MoCo as a valuable toolkit to facilitate and turbocharge the quest for an open, modular, decentralized, and collaborative AI future.

---

### Hypersolid: Emergent Vision Representations via Short-Range Repulsion

Authors: Esteban Rodríguez-Betancourt, Edgar Casasola-Murillo
Published: 2026-01-29
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.21255v3

A central problem in self-supervised learning is preventing representation collapse. Most methods avoid it through global mechanisms, such as contrastive expansion, variance constraints, decorrelating dimensions, or enforcing certain output distributions. In this work, we study a different design: short-range repulsion. We introduce Hypersolid, a self-supervised objective that combines view alignment with local collision avoidance. Our method induces a latent geometry of compact, semantically aligned neighborhoods with low anisotropy. This geometry is especially effective for unsupervised clustering and fine-grained separation, although it comes at the cost of weaker transferability.

---

### Lossless Copyright Protection via Intrinsic Model Fingerprinting

Authors: Lingxiao Chen, Liqin Wang, Wei Lu, Xiangyang Luo
Published: 2026-01-29
Categories: cs.CR, cs.CV
arXiv: https://arxiv.org/abs/2601.21252v1

The exceptional performance of diffusion models establishes them as high-value intellectual property but exposes them to unauthorized replication. Existing protection methods either modify the model to embed watermarks, which impairs performance, or extract model fingerprints by manipulating the denoising process, rendering them incompatible with black-box APIs. In this paper, we propose TrajPrint, a completely lossless and training-free framework that verifies model copyright by extracting unique manifold fingerprints formed during deterministic generation. Specifically, we first utilize a watermarked image as an anchor and exactly trace the path back to its trajectory origin, effectively locking the model fingerprint mapped by this path. Subsequently, we implement a joint optimization strategy that employs dual-end anchoring to synthesize a specific fingerprint noise, which strictly adheres to the target manifold for robust watermark recovery. As input, it enables the protected target model to recover the watermarked image, while failing on non-target models. Finally, we achieved verification via atomic inference and statistical hypothesis testing. Extensive experiments demonstrate that TrajPrint achieves lossless verification in black-box API scenarios with superior robustness against model modifications.

---

### Position: Certifiable State Integrity Should Be Built from Local Validity, Not Global Scale

Authors: Enzo Nicolás Spotorno, Joao R. Campos, Antônio Augusto Medeiros Fröhlich
Published: 2026-01-29
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.21249v2

Breakthroughs in language and vision have motivated increasingly general foundation models for time series and physical dynamics, where evidence is promising but less mature. In safety-critical Cyber-Physical Systems (CPS), globally parameterized dynamics models must remain valid through regime shifts, degradation, and lifecycle updates while supporting traceable assurance evidence. We argue that shared global parameters couple adaptation, frequency-sensitive fault representation, and verification evidence in ways that obstruct retained regime knowledge and bounded analysis. For systems whose operating conditions can be covered by a finite set of locally valid models, assurance evidence should be organized around those bounded models and their explicit composition, not around a single globally adapted model; increasing scale cannot supply explicit validity boundaries. HYDRA is a reference architecture for this position: frozen local specialists, runtime arbitration, and explicit coverage-gap monitoring. It illustrates how modular auditability and formal analysis of bounded components support Certifiable State Integrity, while whole-system certification, switching guarantees, and empirical validation remain separate obligations.

---

### NFCDS: A Plug-and-Play Noise Frequency-Controlled Diffusion Sampling Strategy for Image Restoration

Authors: Zhen Wang, Hongyi Liu, Jianing Li, Zhihui Wei
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.21248v1

Diffusion sampling-based Plug-and-Play (PnP) methods produce images with high perceptual quality but often suffer from reduced data fidelity, primarily due to the noise introduced during reverse diffusion. To address this trade-off, we propose Noise Frequency-Controlled Diffusion Sampling (NFCDS), a spectral modulation mechanism for reverse diffusion noise. We show that the fidelity-perception conflict can be fundamentally understood through noise frequency: low-frequency components induce blur and degrade fidelity, while high-frequency components drive detail generation. Based on this insight, we design a Fourier-domain filter that progressively suppresses low-frequency noise and preserves high-frequency content. This controlled refinement injects a data-consistency prior directly into sampling, enabling fast convergence to results that are both high-fidelity and perceptually convincing--without additional training. As a PnP module, NFCDS seamlessly integrates into existing diffusion-based restoration frameworks and improves the fidelity-perception balance across diverse zero-shot tasks.

---

### Conditional Generative Framework with Peak-Aware Attention for Robust Chemical Detection under Interferences

Authors: Namkyung Yoon, Sanghong Kim, Hwangnam Kim
Published: 2026-01-29
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.21246v1

Gas chromatography-mass spectrometry (GC-MS) is a widely used analytical method for chemical substance detection, but measurement reliability tends to deteriorate in the presence of interfering substances. In particular, interfering substances cause nonspecific peaks, residence time shifts, and increased background noise, resulting in reduced sensitivity and false alarms. To overcome these challenges, in this paper, we propose an artificial intelligence discrimination framework based on a peak-aware conditional generative model to improve the reliability of GC-MS measurements under interference conditions. The framework is learned with a novel peak-aware mechanism that highlights the characteristic peaks of GC-MS data, allowing it to generate important spectral features more faithfully. In addition, chemical and solvent information is encoded in a latent vector embedded with it, allowing a conditional generative adversarial neural network (CGAN) to generate a synthetic GC-MS signal consistent with the experimental conditions. This generates an experimental dataset that assumes indirect substance situations in chemical substance data, where acquisition is limited without conducting real experiments. These data are used for the learning of AI-based GC-MS discrimination models to help in accurate chemical substance discrimination. We conduct various quantitative and qualitative evaluations of the generated simulated data to verify the validity of the proposed framework. We also verify how the generative model improves the performance of the AI discrimination framework. Representatively, the proposed method is shown to consistently achieve cosine similarity and Pearson correlation coefficient values above 0.9 while preserving peak number diversity and reducing false alarms in the discrimination model.

---

### Less Noise, More Voice: Reinforcement Learning for Reasoning via Instruction Purification

Authors: Yiju Guo, Tianyi Hu, Zexu Sun, Yankai Lin
Published: 2026-01-29
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2601.21244v3

Reinforcement Learning with Verifiable Rewards (RLVR) has advanced LLM reasoning, but remains constrained by inefficient exploration under limited rollout budgets, leading to low sampling success and unstable training in complex tasks. We find that many exploration failures arise not from problem difficulty, but from a small number of prompt tokens that introduce interference. Building on this insight, we propose the Less Noise Sampling Framework (LENS), which first prompts by identifying and removing interference tokens. then transfers successful rollouts from the purification process to supervise policy optimization on the original noisy prompts, enabling the model to learn to ignore interference in the real-world, noisy prompting settings. Experimental results show that LENS significantly outperforms GRPO, delivering higher performance and faster convergence, with a 3.88% average gain and over 1.6$\times$ speedup on math reasoning, and a 1.83% gain on scientific and general reasoning. Our work highlights the critical role of pruning interference tokens in improving rollout efficiency, offering a new perspective for RLVR research.

---

### Solving the Offline and Online Min-Max Problem of Non-smooth Submodular-Concave Functions: A Zeroth-Order Approach

Authors: Amir Ali Farzin, Yuen-Man Pun, Philipp Braun, Tyler Summers, Iman Shames
Published: 2026-01-29
Categories: math.OC, cs.LG, math.NA
arXiv: https://arxiv.org/abs/2601.21243v4

We consider max-min and min-max problems with objective functions that are possibly non-smooth, submodular with respect to the minimiser and concave with respect to the maximiser. We investigate the performance of a zeroth-order method applied to this problem. The method is based on the subgradient of the Lovász extension of the objective function with respect to the minimiser and based on Gaussian smoothing to estimate the smoothed function gradient with respect to the maximiser. In expectation sense, we prove the convergence of the algorithm to an $ε$-saddle point in the offline case. Moreover, we show that, in the expectation sense, in the online setting, the algorithm achieves $O(\sqrt{N(1+\bar{P}_N)})$ online duality gap, where $N$ is the number of iterations and $\bar{P}_N$ is the path length of the sequence of optimal decisions. The complexity analysis and hyperparameter selection are presented for all the cases. The theoretical results are illustrated via numerical examples.

---

### Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks

Authors: Luwei Sun, Dongrui Shen, Feng Chuanwen, Jianfe Li, Yulong Zhao, Han Feng
Published: 2026-01-29
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.21242v2

Motivated by challenges in conditional generative modeling, where the target conditional density takes the form of a ratio f1 over f2, this paper develops a theoretical framework for approximating such ratio-type functionals. Here, f1 and f2 are kernel-based marginal densities that capture structured interactions, a setting central to diffusion-based generative models. We provide a concise proof for approximating these ratio-type functionals using deep neural networks with the SignReLU activation function, leveraging the activation's piecewise structure. Under standard regularity assumptions, we establish L^p(Omega) approximation bounds and convergence rates. Specializing to Denoising Diffusion Probabilistic Models (DDPMs), we construct a SignReLU-based neural estimator for the reverse process and derive bounds on the excess Kullback-Leibler (KL) risk between the generated and true data distributions. Our analysis decomposes this excess risk into approximation and estimation error components. These results provide generalization guarantees for finite-sample training of diffusion-based generative models.

---

### TIDE: Tuning-Integrated Dynamic Evolution for LLM-Based Automated Heuristic Design

Authors: Chentong Chen, Mengyuan Zhong, Ye Fan, Jialong Shi, Jianyong Sun
Published: 2026-01-29
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.21239v2

Although Large Language Models have advanced Automated Heuristic Design, treating algorithm evolution as a monolithic text generation task overlooks the coupling between discrete algorithmic structures and continuous numerical parameters. Consequently, existing methods often discard promising algorithms due to uncalibrated constants and suffer from premature convergence resulting from simple similarity metrics. To address these limitations, we propose TIDE, a Tuning-Integrated Dynamic Evolution framework designed to decouple structural reasoning from parameter optimization. TIDE features a nested architecture where an outer parallel island model utilizes Tree Similarity Edit Distance to drive structural diversity, while an inner loop integrates LLM-based logic generation with a differential mutation operator for parameter tuning. Additionally, a UCB-based scheduler dynamically prioritizes high-yield prompt strategies to optimize resource allocation. Extensive experiments across nine combinatorial optimization problems demonstrate that TIDE discovers heuristics that significantly outperform state-of-the-art baselines in solution quality while achieving improved search efficiency and reduced computational costs.

---

### PTQ4ARVG: Post-Training Quantization for AutoRegressive Visual Generation Models

Authors: Xuewen Liu, Zhikai Li, Jing Zhang, Mengjuan Chen, Qingyi Gu
Published: 2026-01-29
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.21238v1

AutoRegressive Visual Generation (ARVG) models retain an architecture compatible with language models, while achieving performance comparable to diffusion-based models. Quantization is commonly employed in neural networks to reduce model size and computational latency. However, applying quantization to ARVG remains largely underexplored, and existing quantization methods fail to generalize effectively to ARVG models. In this paper, we explore this issue and identify three key challenges: (1) severe outliers at channel-wise level, (2) highly dynamic activations at token-wise level, and (3) mismatched distribution information at sample-wise level. To these ends, we propose PTQ4ARVG, a training-free post-training quantization (PTQ) framework consisting of: (1) Gain-Projected Scaling (GPS) mitigates the channel-wise outliers, which expands the quantization loss via a Taylor series to quantify the gain of scaling for activation-weight quantization, and derives the optimal scaling factor through differentiation.(2) Static Token-Wise Quantization (STWQ) leverages the inherent properties of ARVG, fixed token length and position-invariant distribution across samples, to address token-wise variance without incurring dynamic calibration overhead.(3) Distribution-Guided Calibration (DGC) selects samples that contribute most to distributional entropy, eliminating the sample-wise distribution mismatch. Extensive experiments show that PTQ4ARVG can effectively quantize the ARVG family models to 8-bit and 6-bit while maintaining competitive performance. Code is available at http://github.com/BienLuky/PTQ4ARVG .

---

### Characterizing the Effect of Noise in Language Generation in the Limit

Authors: Aaron Li, Ian Zhang
Published: 2026-01-29
Categories: cs.DS, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2601.21237v2

Kleinberg and Mullainathan recently proposed a formal framework for studying the phenomenon of language generation, called language generation in the limit. In this model, an adversary gives an enumeration of example strings from an unknown target language, and the algorithm is tasked with correctly generating unseen strings from the target language within finite time. Refined notions of non-uniform and uniform generation were later introduced by Li, Raman, and Tewari (2025), and a noisy model was introduced by Raman and Raman (2025), which allows the adversary to insert extraneous strings. A natural question in the noisy model is to quantify the effect of noise, by studying the impact of each additional extraneous string. We show two complementary results in this setting. We first show that for both uniform and non-uniform generation, a single noisy string strictly reduces the set of collections that can be generated, thus answering an open question in Raman and Raman (2025). Then, we show for both uniform and non-uniform generation that generation with a single noisy string is equivalent to generation with any finite amount of noise, sharply contrasting with the strict hierarchy for noisy generation in the limit shown by Bai, Panigrahi, and Zhang (2026). Finally, we leverage our previous results to provide the first known characterization for non-uniform noise-dependent generatability.

---

### SHARP: Social Harm Analysis via Risk Profiles for Measuring Inequities in Large Language Models

Authors: Alok Abhishek, Tushar Bandopadhyay, Lisa Erickson
Published: 2026-01-29
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.21235v1

Large language models (LLMs) are increasingly deployed in high-stakes domains, where rare but severe failures can result in irreversible harm. However, prevailing evaluation benchmarks often reduce complex social risk to mean-centered scalar scores, thereby obscuring distributional structure, cross-dimensional interactions, and worst-case behavior. This paper introduces Social Harm Analysis via Risk Profiles (SHARP), a framework for multidimensional, distribution-aware evaluation of social harm. SHARP models harm as a multivariate random variable and integrates explicit decomposition into bias, fairness, ethics, and epistemic reliability with a union-of-failures aggregation reparameterized as additive cumulative log-risk. The framework further employs risk-sensitive distributional statistics, with Conditional Value at Risk (CVaR95) as a primary metric, to characterize worst-case model behavior. Application of SHARP to eleven frontier LLMs, evaluated on a fixed corpus of n=901 socially sensitive prompts, reveals that models with similar average risk can exhibit more than twofold differences in tail exposure and volatility. Across models, dimension-wise marginal tail behavior varies systematically across harm dimensions, with bias exhibiting the strongest tail severities, epistemic and fairness risks occupying intermediate regimes, and ethical misalignment consistently lower; together, these patterns reveal heterogeneous, model-dependent failure structures that scalar benchmarks conflate. These findings indicate that responsible evaluation and governance of LLMs require moving beyond scalar averages toward multidimensional, tail-sensitive risk profiling.

---

### PHDME: Physics-Informed Diffusion Models without Explicit Governing Equations

Authors: Kaiyuan Tan, Kendra Givens, Peilun Li, Thomas Beckers
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.21234v1

Diffusion models provide expressive priors for forecasting trajectories of dynamical systems, but are typically unreliable in the sparse data regime. Physics-informed machine learning (PIML) improves reliability in such settings; however, most methods require \emph{explicit governing equations} during training, which are often only partially known due to complex and nonlinear dynamics. We introduce \textbf{PHDME}, a port-Hamiltonian diffusion framework designed for \emph{sparse observations} and \emph{incomplete physics}. PHDME leverages port-Hamiltonian structural prior but does not require full knowledge of the closed-form governing equations. Our approach first trains a Gaussian process distributed Port-Hamiltonian system (GP-dPHS) on limited observations to capture an energy-based representation of the dynamics. The GP-dPHS is then used to generate a physically consistent artificial dataset for diffusion training, and to inform the diffusion model with a structured physics residual loss. After training, the diffusion model acts as an amortized sampler and forecaster for fast trajectory generation. Finally, we apply split conformal calibration to provide uncertainty statements for the generated predictions. Experiments on PDE benchmarks and a real-world spring system show improved accuracy and physical consistency under data scarcity.

---

### Just Ask: Curious Code Agents Reveal System Prompts in Frontier LLMs

Authors: Xiang Zheng, Yutao Wu, Hanxun Huang, Yige Li, Xingjun Ma, Bo Li, Yu-Gang Jiang, Cong Wang
Published: 2026-01-29
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.21233v2

Autonomous code agents built on large language models are reshaping software and AI development through tool use, long-horizon reasoning, and self-directed interaction. However, this autonomy introduces a previously unrecognized security risk: agentic interaction fundamentally expands the LLM attack surface, enabling systematic probing and recovery of hidden system prompts that guide model behavior. We identify system prompt extraction as an emergent vulnerability intrinsic to code agents and present \textbf{\textsc{JustAsk}}, a self-evolving framework that autonomously discovers effective extraction strategies through interaction alone. Unlike prior prompt-engineering or dataset-based attacks, \textsc{JustAsk} requires no handcrafted prompts, labeled supervision, or privileged access beyond standard user interaction. It formulates extraction as an online exploration problem, using Upper Confidence Bound-based strategy selection and a hierarchical skill space spanning atomic probes and high-level orchestration. These skills exploit imperfect system-instruction generalization and inherent tensions between helpfulness and safety. Evaluated on \textbf{41} black-box commercial models across multiple providers, \textsc{JustAsk} consistently achieves full or near-complete system prompt recovery, revealing recurring design- and architecture-level vulnerabilities. Our results expose system prompts as a critical yet largely unprotected attack surface in modern agent systems.

---

### Data-Driven Generation of Neutron Star Equations of State Using Variational Autoencoders

Authors: Alex Ross, Tianqi Zhao, Sanjay Reddy
Published: 2026-01-29
Categories: astro-ph.HE, astro-ph.IM, astro-ph.SR, cs.LG
arXiv: https://arxiv.org/abs/2601.21231v1

We develop a machine learning model based on a structured variational autoencoder (VAE) framework to reconstruct and generate neutron star (NS) equations of state (EOS). The VAE consists of an encoder network that maps high-dimensional EOS data into a lower-dimensional latent space and a decoder network that reconstructs the full EOS from the latent representation. The latent space includes supervised NS observables derived from the training EOS data, as well as latent random variables corresponding to additional unspecified EOS features learned automatically. Sampling the latent space enables the generation of new, causal, and stable EOS models that satisfy astronomical constraints on the supervised NS observables, while allowing Bayesian inference of the EOS incorporating additional multimessenger data, including gravitational waves from LIGO/Virgo and mass and radius measurements of pulsars. Based on a VAE trained on a Skyrme EOS dataset, we find that a latent space with two supervised NS observables, the maximum mass $(M_{\max})$ and the canonical radius $(R_{1.4})$, together with one latent random variable controlling the EOS near the crust--core transition, can already reconstruct Skyrme EOSs with high fidelity, achieving mean absolute percentage errors of approximately $(0.15\%)$ for $(M_{\max})$ and $(R_{1.4})$ derived from the decoder-reconstructed EOS.

---

### Delegation Without Living Governance

Authors: Wolfgang Rohde
Published: 2026-01-29
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.21226v1

Most governance frameworks assume that rules can be defined in advance, systems can be engineered to comply, and accountability can be applied after outcomes occur. This model worked when machines replaced physical labor or accelerated calculation. It no longer holds when judgment itself is delegated to agentic AI systems operating at machine speed. The central issue here is not safety, efficiency, or employment. It is whether humans remain relevant participants in systems that increasingly shape social, economic, and political outcomes. This paper argues that static, compliance-based governance fails once decision-making moves to runtime and becomes opaque. It further argues that the core challenge is not whether AI is conscious, but whether humans can maintain meaningful communication, influence, and co-evolution with increasingly alien forms of intelligence. We position runtime governance, specifically, a newly proposed concept called the Governance Twin (1); as a strong candidate for preserving human relevance, while acknowledging that accountability, agency, and even punishment must be rethought in this transition.

---

### MGSM-Pro: A Simple Strategy for Robust Multilingual Mathematical Reasoning Evaluation

Authors: Tianyi Xu, Kosei Uemura, Alfred Malengo Kondoro, Tadesse Destaw Belay, Catherine Nana Nyaah Essuman, Ifeoma Okoh, Ganiyat Afolabi, Ayodele Awokoya, David Ifeoluwa Adelani
Published: 2026-01-29
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.21225v3

Large language models have made substantial progress in mathematical reasoning. However, benchmark development for multilingual evaluation has lagged behind English in both difficulty and recency. Recently, GSM-Symbolic showed a strong evidence of high variance when models are evaluated on different instantiations of the same question; however, the evaluation was conducted only in English. In this paper, we introduce MGSM-Pro, an extension of MGSM dataset with GSM-Symbolic approach. Our dataset provides five instantiations per MGSM question by varying names, digits and irrelevant context. Evaluations across nine languages reveal that many low-resource languages suffer large performance drops when tested on digit instantiations different from those in the original test set. We further find that models robustness in HRL setting do not necessarily translate to LRL. Moreover, proprietary models, such as Gemini 2.5 Flash and GPT-4.1 are less robust to digit, whereas Gemini 3.0 Pro is more robust. Among open models, GPT-OSS 120B and DeepSeek v3 show stronger robustness. Based on these findings, we recommend evaluating each problem using at least five digit-varying instantiations to obtain a more robust and realistic assessment of math reasoning.

---

### Causal Discovery for Explainable AI: A Dual-Encoding Approach

Authors: Henry Salgado, Meagan R. Kendall, Martine Ceberio
Published: 2026-01-29
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.21221v1

Understanding causal relationships among features is fundamental for explaining machine learning model decisions. However, traditional causal discovery methods face challenges with categorical variables due to numerical instability in conditional independence testing. We propose a dual-encoding causal discovery approach that addresses these limitations by running constraint-based algorithms with complementary encoding strategies and merging results through majority voting. Applied to the Titanic dataset, our method identifies causal structures that align with established explainable methods.

---

### LAMP: Learning Universal Adversarial Perturbations for Multi-Image Tasks via Pre-trained Models

Authors: Alvi Md Ishmam, Najibul Haque Sarker, Zaber Ibn Abdul Hakim, Chris Thomas
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.21220v1

Multimodal Large Language Models (MLLMs) have achieved remarkable performance across vision-language tasks. Recent advancements allow these models to process multiple images as inputs. However, the vulnerabilities of multi-image MLLMs remain unexplored. Existing adversarial attacks focus on single-image settings and often assume a white-box threat model, which is impractical in many real-world scenarios. This paper introduces LAMP, a black-box method for learning Universal Adversarial Perturbations (UAPs) targeting multi-image MLLMs. LAMP applies an attention-based constraint that prevents the model from effectively aggregating information across images. LAMP also introduces a novel cross-image contagious constraint that forces perturbed tokens to influence clean tokens, spreading adversarial effects without requiring all inputs to be modified. Additionally, an index-attention suppression loss enables a robust position-invariant attack. Experimental results show that LAMP outperforms SOTA baselines and achieves the highest attack success rates across multiple vision-language tasks and models.

---

### Soft Quantization: Model Compression Via Weight Coupling

Authors: Daniel T. Bernstein, Luca Di Carlo, David Schwab
Published: 2026-01-29
Categories: cs.LG, cond-mat.dis-nn
arXiv: https://arxiv.org/abs/2601.21219v1

We show that introducing short-range attractive couplings between the weights of a neural network during training provides a novel avenue for model quantization. These couplings rapidly induce the discretization of a model's weight distribution, and they do so in a mixed-precision manner despite only relying on two additional hyperparameters. We demonstrate that, within an appropriate range of hyperparameters, our "soft quantization'' scheme outperforms histogram-equalized post-training quantization on ResNet-20/CIFAR-10. Soft quantization provides both a new pipeline for the flexible compression of machine learning models and a new tool for investigating the trade-off between compression and generalization in high-dimensional loss landscapes.

---

### Parametric Knowledge is Not All You Need: Toward Honest Large Language Models via Retrieval of Pretraining Data

Authors: Christopher Adrian Kusuma, Muhammad Reza Qorib, Hwee Tou Ng
Published: 2026-01-29
Categories: cs.CL
arXiv: https://arxiv.org/abs/2601.21218v2

Large language models (LLMs) are highly capable of answering questions, but they are often unaware of their own knowledge boundary, i.e., knowing what they know and what they don't know. As a result, they can generate factually incorrect responses on topics they do not have enough knowledge of, commonly known as hallucination. Rather than hallucinating, a language model should be more honest and respond with "I don't know" when it does not have enough knowledge about a topic. Many methods have been proposed to improve LLM honesty, but their evaluations lack robustness, as they do not take into account the knowledge that the LLM has ingested during its pretraining. In this paper, we propose a more robust evaluation benchmark dataset for LLM honesty by utilizing Pythia, a truly open LLM with publicly available pretraining data. In addition, we also propose a novel method for harnessing the pretraining data to build a more honest LLM.

---

### A Flexible Empirical Bayes Approach to Generalized Linear Models, with Applications to Sparse Logistic Regression

Authors: Dongyue Xie, Matthew Stephens
Published: 2026-01-29
Categories: stat.ML, cs.LG, stat.CO, stat.ME
arXiv: https://arxiv.org/abs/2601.21217v2

We introduce a flexible empirical Bayes approach for fitting Bayesian generalized linear models. Specifically, we adopt a novel mean-field variational inference (VI) method and the prior is estimated within the VI algorithm, making the method tuning-free. Unlike traditional VI methods that optimize the posterior density function, our approach directly optimizes the posterior mean and prior parameters. This formulation reduces the number of parameters to optimize and enables the use of scalable algorithms such as L-BFGS and stochastic gradient descent. Furthermore, our method automatically determines the optimal posterior based on the prior and likelihood, distinguishing it from existing VI methods that often assume a Gaussian variational. Our approach represents a unified framework applicable to a wide range of exponential family distributions, removing the need to develop unique VI methods for each combination of likelihood and prior distributions. We apply the framework to solve sparse logistic regression and demonstrate the superior predictive performance of our method in extensive numerical studies, by comparing it to prevalent sparse logistic regression approaches.

---

### Temporal Context and Architecture: A Benchmark for Naturalistic EEG Decoding

Authors: Mehmet Ergezer
Published: 2026-01-29
Categories: cs.LG, cs.AI, eess.SP
arXiv: https://arxiv.org/abs/2601.21215v1

We study how model architecture and temporal context interact in naturalistic EEG decoding. Using the HBN movie-watching dataset, we benchmark five architectures, CNN, LSTM, a stabilized Transformer (EEGXF), S4, and S5, on a 4-class task across segment lengths from 8s to 128s. Accuracy improves with longer context: at 64s, S5 reaches 98.7%+/-0.6 and CNN 98.3%+/-0.3, while S5 uses ~20x fewer parameters than CNN. To probe real-world robustness, we evaluate zero-shot cross-frequency shifts, cross-task OOD inputs, and leave-one-subject-out generalization. S5 achieves stronger cross-subject accuracy but makes over-confident errors on OOD tasks; EEGXF is more conservative and stable under frequency shifts, though less calibrated in-distribution. These results reveal a practical efficiency-robustness trade-off: S5 for parameter-efficient peak accuracy; EEGXF when robustness and conservative uncertainty are critical.

---

### Scaling Reasoning Hop Exposes Weaknesses: Demystifying and Improving Hop Generalization in Large Language Models

Authors: Zhaoyi Li, Jiatong Li, Gangwei Jiang, Linqi Song, Defu Lian, Ying Wei
Published: 2026-01-29
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2601.21214v2

Chain-of-thought (CoT) reasoning has become the standard paradigm for enabling Large Language Models (LLMs) to solve complex problems. However, recent studies reveal a sharp performance drop in reasoning hop generalization scenarios, where the required number of reasoning steps exceeds training distributions while the underlying algorithm remains unchanged. The internal mechanisms driving this failure remain poorly understood. In this work, we conduct a systematic study on tasks from multiple domains, and find that errors concentrate at token positions of a few critical error types, rather than being uniformly distributed. Closer inspection reveals that these token-level erroneous predictions stem from internal competition mechanisms: certain attention heads, termed erroneous processing heads (ep heads), tip the balance by amplifying incorrect reasoning trajectories while suppressing correct ones. Notably, removing individual ep heads during inference can often restore the correct predictions. Motivated by these insights, we propose test-time correction of reasoning, a lightweight intervention method that dynamically identifies and deactivates ep heads in the reasoning process. Extensive experiments across different tasks and LLMs show that it consistently improves reasoning hop generalization, highlighting both its effectiveness and potential.

---

### Intelli-Planner: Towards Customized Urban Planning via Large Language Model Empowered Reinforcement Learning

Authors: Xixian Yong, Peilin Sun, Zihe Wang, Xiao Zhou
Published: 2026-01-29
Categories: cs.AI, cs.CY
arXiv: https://arxiv.org/abs/2601.21212v1

Effective urban planning is crucial for enhancing residents' quality of life and ensuring societal stability, playing a pivotal role in the sustainable development of cities. Current planning methods heavily rely on human experts, which are time-consuming and labor-intensive, or utilize deep learning algorithms, often limiting stakeholder involvement. To bridge these gaps, we propose Intelli-Planner, a novel framework integrating Deep Reinforcement Learning (DRL) with large language models (LLMs) to facilitate participatory and customized planning scheme generation. Intelli-Planner utilizes demographic, geographic data, and planning preferences to determine high-level planning requirements and demands for each functional type. During training, a knowledge enhancement module is employed to enhance the decision-making capability of the policy network. Additionally, we establish a multi-dimensional evaluation system and employ LLM-based stakeholders for satisfaction scoring. Experimental validation across diverse urban settings shows that Intelli-Planner surpasses traditional baselines and achieves comparable performance to state-of-the-art DRL-based methods in objective metrics, while enhancing stakeholder satisfaction and convergence speed. These findings underscore the effectiveness and superiority of our framework, highlighting the potential for integrating the latest advancements in LLMs with DRL approaches to revolutionize tasks related to functional areas planning.

---

### Uncovering Hidden Correctness in LLM Causal Reasoning via Symbolic Verification

Authors: Paul He, Yinya Huang, Mrinmaya Sachan, Zhijing Jin
Published: 2026-01-29
Categories: cs.AI
arXiv: https://arxiv.org/abs/2601.21210v1

Large language models (LLMs) are increasingly being applied to tasks that involve causal reasoning. However, current benchmarks often rely on string matching or surface-level metrics that do not capture whether the output of a model is formally valid under the semantics of causal reasoning. To address this, we propose DoVerifier, a simple symbolic verifier that checks whether LLM-generated causal expressions are derivable from a given causal graph using rules from do-calculus and probability theory. This allows us to recover correct answers to causal queries that would otherwise be marked incorrect due to superficial differences in their causal semantics. Our evaluations on synthetic data and causal QA benchmarks show that DoVerifier more accurately captures semantic correctness of causal reasoning traces, offering a more rigorous and informative way to evaluate LLMs on causal reasoning.

---

### When should I search more: Adaptive Complex Query Optimization with Reinforcement Learning

Authors: Wei Wen, Sihang Deng, Tianjun Wei, Keyu Chen, Ruizhi Qiao, Xing Sun
Published: 2026-01-29
Categories: cs.AI, cs.IR
arXiv: https://arxiv.org/abs/2601.21208v1

Query optimization is a crucial component for the efficacy of Retrieval-Augmented Generation (RAG) systems. While reinforcement learning (RL)-based agentic and reasoning methods have recently emerged as a promising direction on query optimization, most existing approaches focus on the expansion and abstraction of a single query. However, complex user queries are prevalent in real-world scenarios, often requiring multiple parallel and sequential search strategies to handle disambiguation and decomposition. Directly applying RL to these complex cases introduces significant hurdles. Determining the optimal number of sub-queries and effectively re-ranking and merging retrieved documents vastly expands the search space and complicates reward design, frequently leading to training instability. To address these challenges, we propose a novel RL framework called Adaptive Complex Query Optimization (ACQO). Our framework is designed to adaptively determine when and how to expand the search process. It features two core components: an Adaptive Query Reformulation (AQR) module that dynamically decides when to decompose a query into multiple sub-queries, and a Rank-Score Fusion (RSF) module that ensures robust result aggregation and provides stable reward signals for the learning agent. To mitigate training instabilities, we adopt a Curriculum Reinforcement Learning (CRL) approach, which stabilizes the training process by progressively introducing more challenging queries through a two-stage strategy. Our comprehensive experiments demonstrate that ACQO achieves state-of-the-art performance on three complex query benchmarks, significantly outperforming established baselines. The framework also showcases improved computational efficiency and broad compatibility with different retrieval architectures, establishing it as a powerful and generalizable solution for next-generation RAG systems.

---

### A Sheaf-Theoretic and Topological Perspective on Complex Network Modeling and Attention Mechanisms in Graph Neural Models

Authors: Chuan-Shen Hu
Published: 2026-01-29
Categories: cs.LG, cs.AI, math.AT
arXiv: https://arxiv.org/abs/2601.21207v4

Combinatorial and topological structures, such as graphs, simplicial complexes, and cell complexes, form the foundation of geometric and topological deep learning (GDL and TDL) architectures. These models aggregate signals over such domains, integrate local features, and generate representations for diverse real-world applications. However, the distribution and diffusion behavior of GDL and TDL features during training remains an open and underexplored problem. Motivated by this gap, we introduce a cellular sheaf theoretic framework for modeling and analyzing the local consistency and harmonicity of node features and edge weights in graph-based architectures. By tracking local feature alignments and agreements through sheaf structures, the framework offers a topological perspective on feature diffusion and aggregation. Furthermore, a multiscale extension inspired by topological data analysis (TDA) is proposed to capture hierarchical feature interactions in graph models. This approach enables a joint characterization of GDL and TDL architectures based on their underlying geometric and topological structures and the learned signals defined on them, providing insights for future studies on conventional tasks such as node classification, substructure detection, and community detection.

---

### Multilingual Dysarthric Speech Assessment Using Universal Phone Recognition and Language-Specific Phonemic Contrast Modeling

Authors: Eunjung Yeo, Julie M. Liss, Visar Berisha, David R. Mortensen
Published: 2026-01-29
Categories: cs.CL, cs.SD, eess.AS
arXiv: https://arxiv.org/abs/2601.21205v2

The growing prevalence of neurological disorders associated with dysarthria motivates the need for automated intelligibility assessment methods that are applicalbe across languages. However, most existing approaches are either limited to a single language or fail to capture language-specific factors shaping intelligibility. We present a multilingual phoneme-production assessment framework that integrates universal phone recognition with language-specific phoneme interpretation using contrastive phonological feature distances for phone-to-phoneme mapping and sequence alignment. The framework yields three metrics: phoneme error rate (PER), phonological feature error rate (PFER), and a newly proposed alignment-free measure, phoneme coverage (PhonCov). Analysis on English, Spanish, Italian, and Tamil show that PER benefits from the combination of mapping and alignment, PFER from alignment alone, and PhonCov from mapping. Further analyses demonstrate that the proposed framework captures clinically meaningful patterns of intelligibility degradation consistent with established observations of dysarthric speech.

---

### Scaling Embeddings Outperforms Scaling Experts in Language Models

Authors: Hong Liu, Jiaqi Zhang, Chao Wang, Xing Hu, Linkun Lyu, Jiaqi Sun, Xurui Yang, Bo Wang, Fengcun Li, Yulei Qian, Lingtong Si, Yerui Sun et al.
Published: 2026-01-29
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.21204v2

While Mixture-of-Experts (MoE) architectures have become the standard for sparsity scaling in large language models, they increasingly face diminishing returns and system-level bottlenecks. In this work, we explore embedding scaling as a potent, orthogonal dimension for scaling sparsity. Through a comprehensive analysis and experiments, we identify specific regimes where embedding scaling achieves a superior Pareto frontier compared to expert scaling. We systematically characterize the critical architectural factors governing this efficacy -- ranging from parameter budgeting to the interplay with model width and depth. Moreover, by integrating tailored system optimizations and speculative decoding, we effectively convert this sparsity into tangible inference speedups. Guided by these insights, we introduce LongCat-Flash-Lite, a 68.5B parameter model with ~3B activated trained from scratch. Despite allocating over 30B parameters to embeddings, LongCat-Flash-Lite not only surpasses parameter-equivalent MoE baselines but also exhibits exceptional competitiveness against existing models of comparable scale, particularly in agentic and coding domains.

---

### Rethinking Self-Training Based Cross-Subject Domain Adaptation for SSVEP Classification

Authors: Weiguang Wang, Yong Liu, Yingjie Gao, Guangyuan Xu
Published: 2026-01-29
Categories: cs.LG
arXiv: https://arxiv.org/abs/2601.21203v1

Steady-state visually evoked potentials (SSVEP)-based brain-computer interfaces (BCIs) are widely used due to their high signal-to-noise ratio and user-friendliness. Accurate decoding of SSVEP signals is crucial for interpreting user intentions in BCI applications. However, signal variability across subjects and the costly user-specific annotation limit recognition performance. Therefore, we propose a novel cross-subject domain adaptation method built upon the self-training paradigm. Specifically, a Filter-Bank Euclidean Alignment (FBEA) strategy is designed to exploit frequency information from SSVEP filter banks. Then, we propose a Cross-Subject Self-Training (CSST) framework consisting of two stages: Pre-Training with Adversarial Learning (PTAL), which aligns the source and target distributions, and Dual-Ensemble Self-Training (DEST), which refines pseudo-label quality. Moreover, we introduce a Time-Frequency Augmented Contrastive Learning (TFA-CL) module to enhance feature discriminability across multiple augmented views. Extensive experiments on the Benchmark and BETA datasets demonstrate that our approach achieves state-of-the-art performance across varying signal lengths, highlighting its superiority.

---

### Provably Reliable Classifier Guidance via Cross-Entropy Control

Authors: Sharan Sahu, Arisina Banerjee, Yuchen Wu
Published: 2026-01-29
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2601.21200v2

Classifier-guided diffusion models generate conditional samples by augmenting the reverse-time score with the gradient of the log-probability predicted by a probabilistic classifier. In practice, this classifier is usually obtained by minimizing an empirical loss function. While existing statistical theory guarantees good generalization performance when the sample size is sufficiently large, it remains unclear whether such training yields an effective guidance mechanism. We study this question in the context of cross-entropy loss, which is widely used for classifier training. Under mild smoothness assumptions on the classifier, we show that controlling the cross-entropy at each diffusion model step is sufficient to control the corresponding guidance error. In particular, probabilistic classifiers achieving conditional KL divergence $\varepsilon^2$ induce guidance vectors with mean squared error $\widetilde O(d \varepsilon )$, up to constant and logarithmic factors. Our result yields an upper bound on the sampling error of classifier-guided diffusion models and bears resemblance to a reverse log-Sobolev--type inequality. To the best of our knowledge, this is the first result that quantitatively links classifier training to guidance alignment in diffusion models, providing both a theoretical explanation for the empirical success of classifier guidance, and principled guidelines for selecting classifiers that induce effective guidance.

---

### Thinker: A vision-language foundation model for embodied intelligence

Authors: Baiyu Pan, Daqin Luo, Junpeng Yang, Jiyuan Wang, Yixuan Zhang, Hailin Shi, Jichao Jiao
Published: 2026-01-29
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2601.21199v1

When large vision-language models are applied to the field of robotics, they encounter problems that are simple for humans yet error-prone for models. Such issues include confusion between third-person and first-person perspectives and a tendency to overlook information in video endings during temporal reasoning. To address these challenges, we propose Thinker, a large vision-language foundation model designed for embodied intelligence. We tackle the aforementioned issues from two perspectives. Firstly, we construct a large-scale dataset tailored for robotic perception and reasoning, encompassing ego-view videos, visual grounding, spatial understanding, and chain-of-thought data. Secondly, we introduce a simple yet effective approach that substantially enhances the model's capacity for video comprehension by jointly incorporating key frames and full video sequences as inputs. Our model achieves state-of-the-art results on two of the most commonly used benchmark datasets in the field of task planning.

---

### ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling

Authors: Yuchen Yang, Yaru Zhao, Pu Yang, Shaowei Wang, Zhi-Hua Zhou
Published: 2026-01-29
Categories: cs.DC, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.21198v2

While Mixture-of-Experts (MoE) architectures substantially bolster the expressive power of large-language models, their prohibitive memory footprint severely impedes the practical deployment on resource-constrained edge devices, especially when model behavior must be preserved without relying on lossy quantization. In this paper, we present ZipMoE, an efficient and semantically lossless on-device MoE serving system. ZipMoE exploits the synergy between the hardware properties of edge devices and the statistical redundancy inherent to MoE parameters via a caching-scheduling co-design with provable performance guarantee. Fundamentally, our design shifts the paradigm of on-device MoE inference from an I/O-bound bottleneck to a compute-centric workflow that enables efficient parallelization. We implement a prototype of ZipMoE and conduct extensive experiments on representative edge computing platforms using popular open-source MoE models and real-world workloads. Our evaluation reveals that ZipMoE achieves up to $72.77\%$ inference latency reduction and up to $6.76\times$ higher throughput than the state-of-the-art systems.Our code is available at: https://github.com/npnothard/ZipMoE-ICML26.

---

### Generative Recall, Dense Reranking: Learning Multi-View Semantic IDs for Efficient Text-to-Video Retrieval

Authors: Zecheng Zhao, Zhi Chen, Zi Huang, Shazia Sadiq, Tong Chen
Published: 2026-01-29
Categories: cs.CV
arXiv: https://arxiv.org/abs/2601.21193v1

Text-to-Video Retrieval (TVR) is essential in video platforms. Dense retrieval with dual-modality encoders leads in accuracy, but its computation and storage scale poorly with corpus size. Thus, real-time large-scale applications adopt two-stage retrieval, where a fast recall model gathers a small candidate pool, which is reranked by an advanced dense retriever. Due to hugely reduced candidates, the reranking model can use any off-the-shelf dense retriever without hurting efficiency, meaning the recall model bounds two-stage TVR performance. Recently, generative retrieval (GR) replaces dense video embeddings with discrete semantic IDs and retrieves by decoding text queries into ID tokens. GR offers near-constant inference and storage complexity, and its semantic IDs capture high-level video features via quantization, making it ideal for quickly eliminating irrelevant candidates during recall. However, as a recall model in two-stage TVR, GR suffers from (i) semantic ambiguity, where each video satisfies diverse queries but is forced into one semantic ID; and (ii) cross-modal misalignment, as semantic IDs are solely derived from visual features without text supervision. We propose Generative Recall and Dense Reranking (GRDR), designing a novel GR method to uplift recalled candidate quality. GRDR assigns multiple semantic IDs to each video using a query-guided multi-view tokenizer exposing diverse semantic access paths, and jointly trains the tokenizer and generative retriever via a shared codebook to cast semantic IDs as the semantic bridge between texts and videos. At inference, trie-constrained decoding generates a compact candidate set reranked by a dense model for fine-grained matching. Experiments on TVR benchmarks show GRDR matches strong dense retrievers in accuracy while reducing index storage by an order of magnitude and accelerating up to 300$\times$ in full-corpus retrieval.

---

### Do Reasoning Models Enhance Embedding Models?

Authors: Wun Yu Chan, Shaojin Chen, Huihao Jing, Kwun Hang Lau, Elton Chun-Chai Li, Zihao Wang, Haoran Li, Yangqiu Song
Published: 2026-01-29
Categories: cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2601.21192v1

State-of-the-art embedding models are increasingly derived from decoder-only Large Language Model (LLM) backbones adapted via contrastive learning. Given the emergence of reasoning models trained via Reinforcement Learning with Verifiable Rewards (RLVR), a natural question arises: do enhanced reasoning translate to superior semantic representations when these models serve as embedding initializations? Contrary to expectation, our evaluation on MTEB and BRIGHT reveals a **null effect**: embedding models initialized from RLVR-tuned backbones yield no consistent performance advantage over their base counterparts when subjected to identical training recipes. To unpack this paradox, we introduce **H**ierarchical **R**epresentation **S**imilarity **A**nalysis (HRSA), a framework that decomposes similarity across representation, geometry, and function levels. HRSA reveals that while RLVR induces irreversible latent manifold's local geometry reorganization and reversible coordinate basis drift, it preserves the global manifold geometry and linear readout. Consequently, subsequent contrastive learning drives strong alignment between base- and reasoning-initialized models, a phenomenon we term **Manifold Realignment**. Empirically, our findings suggest that unlike Supervised Fine-Tuning (SFT), RLVR optimizes trajectories within an existing semantic landscape rather than fundamentally restructuring the landscape itself.

---

### Function Words as Statistical Cues for Language Learning

Authors: Xiulin Yang, Heidi Getz, Ethan Gotlieb Wilcox
Published: 2026-01-29
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2601.21191v2

What statistical properties might support learning abstract grammatical knowledge from linear input? We address this question by examining the statistical distribution of function words. Function words have been argued to aid acquisition through three distributional properties: high frequency, reliable syntactic association, and phrase-boundary alignment. We conduct a cross-linguistic corpus analysis of 186 languages, which confirms that all three properties are universal. Using counterfactual language modeling and ablation experiments on English, we show that preserving these properties facilitates acquisition in neural learners, with a Goldilocks effect: function words must be frequent enough to be reliable, yet diverse enough to remain informative to structural dependency. Probing analyses further reveal that different learning conditions produce systematically different reliance on function words.

---

### Adaptive and Robust Cost-Aware Proof of Quality for Decentralized LLM Inference Networks

Authors: Arther Tian, Alex Ding, Frank Chen, Simon Wu, Aaron Chan
Published: 2026-01-29
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2601.21189v1

Decentralized large language model inference networks require lightweight mechanisms to reward high quality outputs under heterogeneous latency and cost. Proof of Quality provides scalable verification by sampling evaluator nodes that score candidate outputs, then aggregating their scores into a consensus signal that determines rewards. However, evaluator heterogeneity and malicious score manipulation can distort consensus and inflate payouts, which weakens incentive alignment in open participation settings. This paper extends a cost-aware Proof of Quality mechanism by adding adversary-resilient consensus formation. We study robust aggregation rules, including median and trimmed mean, and an adaptive trust-weighted consensus that updates evaluator weights from deviation signals. Using question answering and summarization workloads with a ground truth proxy for offline analysis, we quantify evaluator reliability and show strong variance across evaluators, including task-dependent misalignment that can invert correlations. We then evaluate robustness under four adversarial strategies, including noise injection, boosting, sabotage, and intermittent manipulation, across a sweep of malicious ratios and evaluator sample sizes. Our results show that robust aggregation improves consensus alignment with the ground truth proxy and reduces sensitivity to noisy and strategic attacks compared with simple averaging. We further characterize the operational trade-off introduced by evaluator sampling, where larger evaluator sets reduce evaluator rewards and increase payoff variance while inference rewards remain relatively stable in our configuration. These findings motivate robust consensus as a default component for cost-aware Proof of Quality and provide practical guidance for selecting evaluator sampling parameters under adversarial risk and resource constraints.

---

### FRISM: Fine-Grained Reasoning Injection via Subspace-Level Model Merging for Vision-Language Models

Authors: Chenyu Huang, Peng Ye, Xudong Tan, Jinhan Mu, Shenghe Zheng, Li Shen, Tao Chen
Published: 2026-01-29
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2601.21187v2

Efficiently enhancing the reasoning capabilities of Vision-Language Models (VLMs) by merging them with Large Reasoning Models (LRMs) has emerged as a promising direction. However, existing methods typically operate at a coarse-grained layer level, which often leads to a trade-off between injecting reasoning capabilities and preserving visual capabilities. To address this limitation, we propose FRISM (Fine-grained Reasoning Injection via Subspace-level model Merging), a fine-grained reasoning injection framework based on subspace-level model merging. Observing that different SVD subspaces contribute differently to reasoning and perception, FRISM decomposes LRM task vectors via Singular Value Decomposition (SVD) and adaptively tunes the scaling coefficients of each subspace through learning to realize fine-grained reasoning injection. Furthermore, we introduce a label-free self-distillation learning strategy with dual-objective optimization using common vision-language perception datasets. Extensive experiments demonstrate that FRISM effectively improves reasoning capabilities while largely preserving the model's visual capabilities by consistently achieving strong performance across diverse visual-language reasoning benchmarks.

---

### Sycophantic Anchors: Localizing and Quantifying User Agreement in Reasoning Models

Authors: Jacek Duszenko
Published: 2026-01-29
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2601.21183v2

Reasoning models frequently agree with incorrect user suggestions -- a behavior known as sycophancy. However, it is unclear where in the reasoning trace this agreement originates and how strong the commitment is. We introduce \emph{sycophantic anchors} -- sentences identified via counterfactual analysis that commit models to user agreement. Across four reasoning models spanning three architecture families (Llama, Qwen, Falcon-hybrid) and 1.5B--8B parameters, we analyze over 200,000 counterfactual rollouts and show that linear probes reliably detect sycophantic anchors (74--85\% balanced accuracy), outperforming text-only baselines at high commitment levels -- confirming they capture internal states beyond surface vocabulary. Regressors further predict commitment strength from activations ($R^2$ up to 0.74). We observe a consistent asymmetry: sycophancy leaves a stronger mechanistic footprint than correct reasoning. We also find that sycophancy builds gradually during generation rather than being determined by the prompt. These findings enable sentence-level detection and quantification of model misalignment mid-inference.

---

### Rethinking Refinement: Correcting Generative Bias without Noise Injection

Authors: Xin Peng, Ang Gao
Published: 2026-01-29
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2601.21182v1

Generative models, including diffusion and flow-based models, often exhibit systematic biases that degrade sample quality, particularly in high-dimensional settings. We revisit refinement methods and show that effective bias correction can be achieved as a post-hoc procedure, without noise injection or multi-step resampling of the sampling process. We propose a flow-matching-based \textbf{Bi-stage Flow Refinement (BFR)} framework with two refinement strategies operating at different stages: latent space alignment for approximately invertible generators and data space refinement trained with lightweight augmentations. Unlike previous refiners that perturb sampling dynamics, BFR preserves the original ODE trajectory and applies deterministic corrections to generated samples. Experiments on MNIST, CIFAR-10, and FFHQ at 256x256 resolution demonstrate consistent improvements in fidelity and coverage; notably, starting from base samples with FID 3.95, latent space refinement achieves a \textbf{state-of-the-art} FID of \textbf{1.46} on MNIST using only a single additional function evaluation (1-NFE), while maintaining sample diversity.

---