Searcharxiv⌕ Search

arXiv subjects

Zheng Zhang

Publications and source records attributed to Zheng Zhang.

At least 19 recordsLinked to original sources

AutoRL: A Multi-Agent System towards Autonomous Reinforcement Learning

Recursive self-improvement (RSI) is one of the most promising paths toward an AI singularity for superintelligence. Achieving this vision requires AI systems to undertake the full training pipeline for iterative improvement. Reinforcement learning (RL) plays an essential role in this cycle to improve model capabilities and generalization, yet automating RL remains difficult due to unstable training dynamics and the need for sustained supervision. We introduce AutoRL, a multi-agent system for automating RL training, as a cornerstone for future RSI. Different from conducting research by end-to-end complete experiments, AutoRL monitors the evolving training state and integrates process-level supervision and research into an ongoing RL training, following the practical workflow of an RL researcher. Specifically, AutoRL consists of three RL-specific agentic roles. Perceptors track training dynamics, rollouts, evaluation, and efficiency, triggering deliberation when they detect anomalies, degradation, or stagnation. Three complementary Proposers diagnose problems and refine candidate interventions through shared evidence and checks. The Orchestrator reviews their recommendation and applies approved changes through online hot updates or checkpoint-based resumption. Experiments on Qwen3 and LongCat-Flash-Lite demonstrate the superiority of AutoRL over static RL recipes and a general agent. The research records show that existing agents can handle unattended RL supervision and dynamic adjustment. The difficulty for future RSI lies in developing fundamental methods that are valuable for resource-consuming validation.

cs.AI↗

KUDEX: A Zero-Shot Decompositional Framework for Logical Analysis in Natural Language

While Large Language Models (LLMs) excel at generating fluent text, they often struggle with multi-step logical reasoning with inferential connections among statements. We introduce KUDEX (Knowledge-Upward-Downward Reasoning Extension), a novel zero-shot, decompositional framework for analyzing and verifying logical relationships in natural language. Unlike traditional approaches that rely on surface-level entailment checks, KUDEX leverages LLMs to decompose reasoning into atomic primitives and constructs transparent reasoning graphs that serve both as execution traces and audit logs. Our framework includes a general prompt adaptation pipeline that adapts to diverse reasoning tasks without additional fine-tuning or domain-specific pipelines. Through comprehensive experiments on benchmarks across four domains, we demonstrate that KUDEX provides superior performance compared to existing methods, offering a more granular and robust framework for assessing logical relationships between statements.

cs.CL↗

UniSpark: Exploring Hybrid Attention for Multimodal Understanding and Generation

We present UniSpark-9B, a unified multimodal model that investigates hybrid attention for visual understanding and generation. Built on a Mixture-of-Transformer-Experts backbone with approximately 9B parameters in each branch, UniSpark combines full attention for explicit token interactions with GatedDeltaNet linear attention for compact context propagation. Autoregressive text prediction and visual flow matching jointly support visual question answering, text-to-image generation, instruction-guided editing, and interleaved text-image prediction. In interleaved generation, noisy visual targets are isolated from subsequent predictions, while clean visual representations provide context for later outputs. A blockwise protocol enforces this distinction through attention masks in full-attention layers and recurrent-state retention rules in linear-attention layers. The selected training sources contain 159.7M generation samples, 213.4M understanding samples, and 624.8K interleaved sequences. A four-stage curriculum jointly trains multimodal understanding and visual generation, refining answer accuracy and instruction following while extending image synthesis to 2K. UniSpark scores 2470.3 on MME-S and 0.83 on GenEval. Qualitative examples further illustrate its capabilities in interleaved image generation, video synthesis, and video editing. Together, these results demonstrate that hybrid attention provides an effective backbone for unified multimodal understanding and generation.

cs.CV↗

SpecXMaster Technical Report

Intelligent spectroscopy serves as a pivotal element in AI-driven closed-loop scientific discovery, functioning as the critical bridge between matter structure and artificial intelligence. However, conventional expert-dependent spectral interpretation encounters substantial hurdles, including susceptibility to human bias and error, dependence on limited specialized expertise, and variability across interpreters. To address these challenges, we propose SpecXMaster, an intelligent framework leveraging Agentic Reinforcement Learning (RL) for NMR molecular spectral interpretation. SpecXMaster enables automated extraction of multiplicity information from both 1H and 13C spectra directly from raw FID (free induction decay) data. This end-to-end pipeline enables fully automated interpretation of NMR spectra into chemical structures. It demonstrates superior performance across multiple public NMR interpretation benchmarks and has been refined through iterative evaluations by professional chemical spectroscopists. We believe that SpecXMaster, as a novel methodological paradigm for spectral interpretation, will have a profound impact on the organic chemistry community.

cs.LG↗

Textual Response-Aware Guidance for Zero-Shot Adversarial Robustness of Vision-Language Models

Vision-language models (VLMs) exhibit strong zero-shot generalization but remain vulnerable to adversarial attacks. Existing adversarial fine-tuning methods often use pre-trained model guidance to preserve image--text alignment while improving robustness. Our analysis of visual representation drift reveals complementary limitations of two representative forms of guidance. Matching visual features constrains all directions equally, providing broad coverage without semantic emphasis. Matching prediction distributions prioritizes semantically relevant directions, but remains limited by the candidate texts and prediction probabilities. These limitations can hinder preservation of pre-trained image--text relationships and degrade zero-shot performance. We therefore propose Textual Response-Aware Guidance (TRAG), which constrains visual representation drift using the pre-trained text encoder's responses to local variations in input embeddings. TRAG constructs anchor-level constraints from the response directions and magnitudes around semantic anchors, then aggregates them using pre-trained image--anchor relevance, yielding an image-adaptive constraint. Extensive experiments across architectures and tasks demonstrate state-of-the-art clean and robust performance. Further analysis shows that TRAG helps preserve pre-trained image--text relationships.

cs.CV↗

Adaptive Entropy-Sparing for Efficient Reasoning

Large Reasoning Models (LRMs) produce explicit chains of thought to improve transparency, but they frequently overthink, generating long yet inefficient reasoning chains that inflate computational costs and risk introducing errors. While numerous reinforcement learning methods struggle to penalize verbosity to achieve conciseness, they typically apply the penalty uniformly across lengthy sequences, which leads to credit confounding, that is, the failure to distinguish essential reasoning steps from superfluous ones. In this paper, we study a measurable proxy for this confounding: high-entropy tokens that often coincide with surface transition points where reasoning re-enters verification. Based on this insight, we propose Adaptive Entropy-Sparing (AES), which selectively spares high-entropy tokens that occur before an efficient in-group reference, while penalizing overlength high-entropy suffix tokens associated with superfluous branching. Our AES thereby navigates the suppression-exploration trade-off, reducing overthinking while preserving average reasoning accuracy. We also present an entropy-length model that motivates why local uncertainty can serve as a proxy for future reasoning cost. Across eight reasoning benchmarks, AES reduces reasoning length by up to 52.9% while increasing mean accuracy by up to 3.8% over the base models, consistently outperforming existing efficiency-oriented methods. Our code is available at https://github.com/MobiusDai/AES.

cs.CL↗

Quotient Index for Symmetry-Protected Zeros and the Tenfold Way in Mackey's Construction of Super Corepresentations

Symmetries that anticommute with a Hamiltonian can protect zero modes by obstructing their removal through symmetry-compatible perturbations. We formulate this obstruction as a quotient of the additive group of symmetry representations by the image of the map that forgets a compatible representation grading. To compute the quotient, we use Mackey's method to construct all irreducible super corepresentations of finite graded symmetry groups, allowing antiunitary operations and projective factor systems. Starting from the unitary Hamiltonian-preserving normal subgroup, this construction yields two complex and ten real structural types, characterized by residual Altland--Zirnbauer symmetry algebras. The residual and induced graded modules have isomorphic graded commutants, and their represented symmetry algebras share a graded Morita class that is super opposite to that of the commutant. We derive character criteria that identify these types. Exhaustive enumeration of the restriction relations gives independent sector contributions $0$, $Z$, or $Z_2$, with only integer factors in the purely unitary case. In the finite semisimple setting, the complete index vanishes if and only if all zero modes can be lifted by a symmetry-compatible perturbation. Applications to one-dimensional end states and interpolation cylinders of disordered boundary Dirac theories recover protected multiplet counts and connect the quotient to bulk topology and sigma-model topological terms using bulk--boundary correspondence, texture-index theorems, and fermion-phase calculations.

cond-mat.mes-hall↗

From Learner Behavior to Reusable Skills for Effective and Efficient Learner Simulation

Learner simulation aims to reproduce how a particular learner behaves on new tasks. Although Large Language Models (LLMs) can generate increasingly fine-grained learning behaviors, existing approaches often need to repeatedly process a growing interaction history to reconstruct the learner. This introduces additional context and inference costs and makes the acquired learner-specific simulation capability difficult to reuse across different LLMs. We therefore propose Learner2Skill, which externalizes the simulation capability acquired from historical interactions into a persistent and reusable Simulation Skill. The Skill captures the learner's current learning state and recurring response patterns, evolves as new real interactions arrive, and can be adapted to a new LLM through lightweight executor calibration without reconstructing the learner from scratch. Experiments show that Learner2Skill more faithfully reproduces fine-grained learner behavior while reducing overall token cost, and that the same constructed Skills can be effectively reused across different LLM executors.

cs.AI↗

SSFMamba: Learning Symmetry-driven Spatial-Frequency Modeling for Physically Consistent 3D Medical Image Segmentation

Accurate 3D medical image segmentation requires a delicate balance between fine-grained local details and global contextual understanding. While spatial-domain models often struggle with long-range dependencies, existing frequency-based approaches frequently overlook intrinsic spectral properties such as Hermitian symmetry, leading to suboptimal feature integration. In this paper, we propose SSFMamba, a Mamba-based Symmetry-driven Spatial-Frequency fusion framework tailored for 3D medical imaging. Our architecture employs a complementary dual-branch design: the spatial branch models voxel sequences to preserve intricate anatomical textures, while the frequency branch operates on spectral components, each of which aggregates information from the entire volume, to capture global context. A core innovation is the 3D Multi-Directional Scanning Mechanism (MDSM), which integrates Hermitian symmetry with the causal nature of State Space Models (SSMs) to enable direction-aware global modeling. Crucially, by shifting the modeling focus to frequency-domain spectral components, SSFMamba captures the underlying structural characteristics of anatomical tissues. This leads to a highly adaptable framework that excels in both MRI and CT applications, regardless of the significant variations in intensity distributions. Extensive evaluations on the BraTS2020, BraTS2023, and BTCV datasets demonstrate that SSFMamba consistently outperforms state-of-the-art methods. Notably, our approach achieves exceptional performance on low-contrast organs such as the pancreas (81.97% Dice), underscoring its potential as a unified and physically consistent perception framework for diverse 3D clinical applications.

cs.CV↗

From Outcome to Representation: Tracing Reasoning Mechanisms through Integrated Policy Gradient

Large Language Models (LLMs) demonstrate remarkable reasoning capabilities, yet the internal mechanisms driving these multi-step processes remain opaque. Existing mechanistic interpretability approaches often rely on superficial text-pattern co-occurrences or capture only short-term effects, struggling to trace the long-horizon, outcome-driven influence of internal components such as neurons or sparse features in multi-step reasoning.In this paper, we introduce Integrated Policy Gradient (IPG), a novel, training-free framework that brings policy-based attribution into mechanistic interpretability for LLM reasoning. Grounded in \textit{outcome-oriented} and \textit{sequential-influence-aware} principles, it backpropagates outcome-level, non-differentiable behavioral signals (e.g., reasoning outcome correctness) through entire inference trajectories, utilizing path integration to isolate outcome-relevant mediators at component level for reliable attribution. This provides a linear approximation of Causal Mediation Analysis (CMA) in reasoning. Empirical evaluations demonstrate that \method achieves precise localization and reveals mechanistic insights by identifying sparse, concentrated subsets of internal components associated with specific reasoning behaviors. Our method also enables modulation of reasoning capabilities, offering a powerful and reliable framework for understanding LLM reasoning. The project page is at https://foundation-model-research.github.io/IPG .

cs.AI↗

Training Specialist LLMs on Your Phone: Forward-Only Optimization via Block-Spectral Coordinates

Large language models (LLMs) on personal devices keep user data local, but limited memory restricts their general capability. Since most user-specific tasks do not require the breadth of frontier models, a practical solution is to adapt a shared on-device model into lightweight specialists. However, backpropagation is prohibitive in memory, and users rarely provide labeled data. We present MobileSpec, a system that trains specialists with forward passes alone and learns from automatically checked task outcomes, as in reinforcement learning with verifiable rewards (RLVR). Because forward-only updates are built from random perturbations, training is efficient only when these perturbations resemble the weight changes of RLVR, which largely preserve the singular values of pretrained weights and rotate their singular vectors. The core algorithm of MobileSpec, FuRA-S, produces such perturbations while training only the block-spectral coordinates of pretrained weights, namely the singular values of their column blocks. As we show theoretically and empirically, rescaling these blocks independently shifts their relative contributions to each weight matrix and thereby rotates its singular vectors. Since this update can be rewritten as a transformation of the layer input, FuRA-S also adapts INT4 models without modifying their weights. With specialists of only 1-5 MB that update 0.015-0.05% of the parameters, MobileSpec raises the success rate of a phone-use agent from 6-23% up to 89-100% and trains a personalized 8B model on an iPhone 16 Pro overnight. On RLVR benchmarks, FuRA-S is competitive with full-parameter evolution strategies (ES) and outperforms ES over LoRA adapters and quantized ES on INT4 models.

cs.LG↗

Beyond MACs: Deployment-Constrained Design Space Exploration for Tensorized Neural Networks

Tensor-train (TT) compression reduces neural-network parameters, but the latency of a tensorized layer is set by its contraction order and its mapping onto hardware: a contraction path with fewer multiply-accumulate operations (MACs) is not necessarily faster on a systolic array. We present a deployment-constrained, latency-guided design space exploration (DSE) framework for fixed-rank tensorized neural networks (TNNs) under a fixed processing-element (PE) budget. Using MAC-guided path pruning, it evaluates feasible combinations of contraction paths, three PE-array partitions, and input-, output-, and weight-stationary dataflows. By minimizing the sum of inference and training latency estimates, it selects a single physical PE partition shared across layers and both modes, while allowing layer- and mode-specific paths and dataflows. On ResNet-18 and ViT-Ti/4, 25%-50% of the layers select a path other than the minimum-MAC one, the best dataflow varies across layers and models, and on a representative ViT layer the latency-selected candidate reduces simulated latency by 23.5% relative to the minimum-MAC candidate. Deployed on a Xilinx VU9P FPGA accelerator that supports both inference and training, the complete TT+INT8 system achieves 3.28x-4.00x inference and 3.42x-3.85x training speedups over dense implementations with the same PE budget.

cs.AR↗

Controlling Action-Head Co-Adaptation to VLM Representation Anisotropy

Modern learning systems increasingly combine pretrained models with downstream heads. Despite the growing importance of such modular designs, how the downstream model co-adapts to the geometry of pretrained representations remains relatively underexplored. We study this problem through vision-language-action models (VLAs), where a pretrained vision-language model (VLM) conditions an action model for control. We find that the action model progressively concentrates its sensitivity on high-variance dimensions of the VLM outputs. Through controlled experiments across different backbone-head models, we further show that counteracting this sensitivity concentration can be beneficial under representation--task mismatch settings. Motivated by these observations, we introduce \emph{Adaptive Interface Equalizer} (AdaIE), a parameter-free mechanism that reweights pretrained representations according to the evolving downstream sensitivity during training. AdaIE improves VLA success rates on simulation benchmarks, with gains of up to $+5.28$ percentage points (pp) on RoboCasa-GR1 and $+1.68$ pp on RoboTwin 2.0, and achieves an average gain of $+19$ pp across five real-world manipulation tasks. It further generalizes to diffusion policies and world models, improving downstream performance in both settings. Our results highlight interface co-adaptation as an important optimization problem when coupling pretrained representations with downstream models.

cs.AI↗

Humans' ALMANAC: A Human Collaboration Dataset of Action-Level Mental Model Annotations for Agent Collaboration

Recent advances in LLM agents have enabled complex cognitive capabilities, such as multi-step reasoning, planning, and tool use, that increasingly position these agents as collaborative partners in human workflows. Effective collaboration, however, requires collaborators to continuously maintain and align mental models of their own reasoning, partners' intentions, and shared goals during the collaborative process. Today's agents rarely develop such capabilities since they are primarily optimized for task completion rather than for maintaining the underlying mental models required for collaborative processes. Moreover, the community lacks authentic human collaboration data with action-level mental model annotations that could guide agents toward collaborative competence. To bridge this gap, we present ALMANAC, a dataset of Action-Level Mental model ANnotations for Agent Collaboration built from the Map Task, a classic dyadic routing task from social science. ALMANAC contains 2,987 collaboration actions, each paired with theory-informed mental model annotations that record the participants' self-reasoning, perceived partner intent, and perceived team goal. We benchmark six LLMs on predicting humans' next-turn behavior and mental models. Our results demonstrate ALMANAC's utility in evaluating models' ability to simulate human collaborative behaviors and infer their underlying mental models. The project page is available at https://human-agent-collab.hailab.io/almanac.

cs.AI↗

FuRA: Full-Rank Parameter-Efficient Fine-Tuning with Spectral Preconditioning

Both full fine-tuning (Full FT) and parameter-efficient fine-tuning methods such as LoRA introduce weight updates without accounting for the spectral structure established during pretraining. As a result, noisy gradients from limited fine-tuning data can perturb robust pretrained features. We identify spectral preconditioning as the missing ingredient: reparameterizing each weight matrix through its full-rank singular value decomposition (SVD) and freezing one singular basis constrains updates to the pretrained column space, yielding a preconditioned optimization scheme that outperforms unconstrained Full FT at the same trainable parameter count. Building on this insight, we propose FuRA (Full-Rank Adaptation), an efficient full-rank adaptation framework based on a block tensor-train factorization W = LSR, where the large core L is fixed to the pretrained block-wise SVD basis, while only the compact core R and the block-wise singular values S are optimized. This design simultaneously provides full-rank spectral preconditioning, preserves full-rank update expressivity, and achieves parameter, memory, and step-time efficiency comparable to LoRA. FuRA consistently outperforms Full FT across multiple settings, including LLM fine-tuning (+1.37 on LLaMA-3-8B commonsense reasoning), LLM reinforcement learning for mathematical reasoning, and visual instruction tuning for VLMs. Furthermore, the 4-bit quantized variant, QFuRA, also surpasses QLoRA. Code is available at https://github.com/yequanzhao/FuRA

cs.LG↗

Breaking Information Islands in Sparse Tuning via Small-World Connectivity

Sparse tuning is widely used to adapt large language models due to its parameter efficiency. However, its effectiveness depends not only on the number of trainable parameters, but also on the feature-interaction topology of the sparse update's structural support. Limited or uneven connectivity in this support may isolate subsets of feature dimensions, forming information islands that restrict direct task-specific interactions through the trainable update. To address this issue, we propose Small-World Induced Fine-Tuning (SWIFT), a sparse tuning framework inspired by small-world connectivity. SWIFT combines block-diagonal sparse updates for local feature mixing with a fixed deterministic scattering permutation that routes the mixed outputs across feature groups without additional trainable parameters. We further show theoretically that certain random sparse-update models and block-diagonal updates can exhibit information islands, whereas a sufficiently scattering permutation guarantees positive block-level expansion when the trainable block-diagonal factor consists of equal-sized, structurally dense square blocks. Extensive experiments on commonsense reasoning, natural language generation, and image classification demonstrate that SWIFT outperforms the evaluated parameter-efficient fine-tuning (PEFT) baselines on average and can match or surpass full fine-tuning. Implementation is available at https://github.com/xianchaoguan/SWIFT

cs.AI↗

PASA-KV: Attacking Action-Facing Visual Routes in World Action Models

World action models (WAMs) increasingly couple internal visual or world representations with action generation, making their world-to-action pathways a critical yet poorly understood source of robustness. We investigate when perturbations to these internal representations actually become behaviorally consequential, and uncover a pronounced mismatch between representation perturbability and closed-loop vulnerability: attack effectiveness varies sharply across visual K/V routes, while larger representation displacement does not reliably imply stronger control failure. Our analysis attributes this mismatch to the action-conditioned readout, where keys determine where action queries read and values determine what content they receive. This motivates an action-facing view of vulnerability, in which the effect of a perturbation depends on how it is routed into action generation rather than on representation-space magnitude alone. We instantiate this view with Tail-AV, a short-probe signal for identifying vulnerable routes, and operationalize it in PASA-KV, which selects action-facing K/V targets before training a fresh task-shared attack. Experiments across multiple WAM systems and a physical robot validate readout-informed target selection, demonstrate strong closed-loop attacks, and reveal substantial system-level differences in resistance.

cs.RO↗

Powering Up Zeroth-Order Training via Subspace Gradient Orthogonalization

Zeroth-order (ZO) optimization provides a gradient-free alternative to first-order (FO) methods by estimating gradients via finite differences of function evaluations, and has recently emerged as a memory-efficient paradigm for fine-tuning large-scale models by avoiding backpropagation. However, ZO optimization has a fundamental tension between accuracy and query efficiency. In this work, we show that ZO optimization can be substantially improved by unifying two complementary principles: (i) a projection-based subspace view that reduces gradient estimation variance by exploiting the intrinsic low-rank structure of model updates, and (ii) Muon-style spectral optimization that applies gradient orthogonalization to extract informative spectral structure from noisy ZO gradients. These findings form a unified framework of subspace gradient orthogonalization, which we instantiate in a new method, ZO-Muon, admitting a natural interpretation as a low-rank Muon optimizer in the ZO setting. Extensive experiments on large language models (LLMs) and vision transformers (ViTs) demonstrate that ZO-Muon significantly accelerates convergence and achieves a win-win improvement in accuracy and query/runtime efficiency. Notably, compared to the popular MeZO baseline, ZO-Muon requires only 24.7% of the queries to reach the same SST-2 performance for LLM fine-tuning, and improves accuracy by 25.1% on ViT-L fine-tuning on CIFAR-100.

cs.LG↗