arXiv subjects

Tong Zhang

Publications and source records attributed to Tong Zhang.

At least 217 records · Page 12Linked to original sources

arXiv2025

Demystifying Singular Defects in Large Language Models

Large transformer models are known to produce high-norm tokens. In vision transformers (ViTs), such tokens have been mathematically modeled through the singular vectors of the linear approximations of layers. However, in large language models (LLMs), the underlying causes of high-norm tokens remain largely unexplored, and their different properties from those of ViTs require a new analysis framework. In this paper, we provide both theoretical insights and empirical validation across a range of recent models, leading to the following observations: i) The layer-wise singular direction predicts the abrupt explosion of token norms in LLMs. ii) The negative eigenvalues of a layer explain its sudden decay. iii) The computational pathways leading to high-norm tokens differ between initial and noninitial tokens. iv) High-norm tokens are triggered by the right leading singular vector of the matrix approximating the corresponding modules. We showcase two practical applications of these findings: the improvement of quantization schemes and the design of LLM signatures. Our findings not only advance the understanding of singular defects in LLMs but also open new avenues for their application. We expect that this work will stimulate further research into the internal mechanisms of LLMs. Code is released at https://github.com/haoqiwang/singular_defect.

cs.CL↗

arXiv2025

ProCaliper: functional and structural analysis, visualization, and annotation of proteins

Understanding protein function at the molecular level requires connecting residue-level annotations with physical and structural properties. This can be cumbersome and error-prone when functional annotation, computation of physico-chemical properties, and structure visualization are separated. To address this, we introduce ProCaliper, an open-source Python library for computing and visualizing physico-chemical properties of proteins. It can retrieve annotation and structure data from UniProt and AlphaFold databases, compute residue-level properties such as charge, solvent accessibility, and protonation state, and interactively visualize the results of these computations along with user-supplied residue-level data. Additionally, ProCaliper incorporates functional and structural information to construct and optionally sparsify networks that encode the distance between residues and/or annotated functional sites or regions. The package ProCaliper and its source code, along with the code used to generate the figures in this manuscript, are freely available at https://github.com/PNNL-Predictive-Phenomics/ProCaliper.

q-bio.BM↗

arXiv2025

Chain-of-Experts: Unlocking the Communication Power of Mixture-of-Experts Models

We propose Chain-of-Experts (CoE), a new Mixture-of-Experts (MoE) architecture that introduces sequential expert communication within each layer. Unlike traditional MoE models, where experts operate independently in parallel, CoE processes tokens iteratively across a chain of experts inside a layer. To support dynamic expert selection across iterations, CoE employs a dedicated router at each iteration step within a layer. This design allows tokens to re-evaluate and select different experts during each iteration, rather than being statically assigned. As a result, CoE introduces a flexible routing mechanism that increases the diversity of expert combinations and enriches the model's representational capacity. CoE demonstrates improved performance under fixed compute: on math reasoning tasks, it reduces validation loss from 1.20 to 1.12 compared to a standard MoE. Beyond performance, CoE offers a new scaling axis: depth through expert iteration, which complements conventional width/depth scaling. For example, using 2x iterations matches the performance of 3x expert selections (in width), while reducing memory usage by 17.6-42% relative to other scaling strategies. Our analysis reveals that CoE's benefits stem from its iterative residual structure and enhanced expert specialization empowered by iterative routing, which together unlock more expressive representations. Code is available at https://github.com/ZihanWang314/coe.

cs.LG↗

arXiv2025

Linear and Numerical SDoF Bounds of Active RIS-Assisted MIMO Wiretap Interference Channel

The multiple-input multiple-output (MIMO) wiretap interference channel (IC) serves as a canonical model for information-theoretic security, where a multiple-antenna eavesdropper attempts to intercept communications in a two-user MIMO IC system. The secure degrees-of-freedom (SDoF) of an active reconfigurable intelligent surface (RIS)-assisted MIMO wiretap IC is with practical interests but remains unexplored. In this paper, we establish both sum-SDoF lower and upper bounds through linear beamforming conditions and numerical methods. Specifically, our proposed lower bound is derived from transmission scheme design and corresponding solutions to the sum-SDoF maximization problem, formulated by linear integer programming. The solutions to this optimization problem addresses RIS element allocation for leakage and interference cancellation. The proposed upper bound is obtained by solving a nuclear norm minimization problem, leveraging the fact that nuclear norm serves as a convex relaxation of the rank function. For symmetry antenna configurations, we derive a closed-form lower bound. Extensive numerical simulations show that our proposed lower and upper bounds coincide across many antenna configurations, and our proposed lower bound outperforms the existing benchmark.

cs.IT↗

arXiv2025

DuaShepherd: Integrating Stepwise Correctness and Potential Rewards for Mathematical Reasoning

In this paper, we propose DuaShepherd, a novel reward modeling framework that integrates two complementary reward signals, correctness and potential, to enhance the mathematical reasoning capabilities of Large Language Models (LLMs). While correctness-based signals emphasize identification of stepwise errors, potential-based signals focus on the likelihood of reaching the correct final answer. We developed an automated pipeline for constructing large-scale reward modeling dataset with both signals. A unified, multi-head architecture was explored to train the two reward models in a multi-task setup, demonstrating benefits from learning both correctness and potential in parallel. By combining these two signals into a compound probability, our model achieves consistent performance improvements across multiple benchmarks. Empirical evaluations on MATH500 and ProcessBench confirm that this combined reward significantly outperforms models trained on either reward type alone, achieving state-of-the-art performance under comparable resource constraints.

cs.CL↗

arXiv2025

CRIA: A Cross-View Interaction and Instance-Adapted Pre-training Framework for Generalizable EEG Representations

The difficulty of extracting deep features from EEG data and effectively integrating information from multiple views presents significant challenges for developing a generalizable pretraining framework for EEG representation learning. However, most existing pre-training methods rely solely on the contextual semantics of a single view, failing to capture the complex and synergistic interactions among different perspectives, limiting the expressiveness and generalization of learned representations. To address these issues, this paper proposes CRIA, an adaptive framework that utilizes variable-length and variable-channel coding to achieve a unified representation of EEG data across different datasets. In this work, we define cross-view information as the integrated representation that emerges from the interaction among temporal, spectral, and spatial views of EEG signals. The model employs a cross-attention mechanism to fuse temporal, spectral, and spatial features effectively, and combines an attention matrix masking strategy based on the information bottleneck principle with a novel viewpoint masking pre-training scheme. Experimental results on the Temple University EEG corpus and the CHB-MIT dataset show that CRIA outperforms existing methods with the same pre-training conditions, achieving a balanced accuracy of 57.02% for multi-class event classification and 80.03% for anomaly detection, highlighting its strong generalization ability.

cs.LG↗

arXiv2025

VL-GenRM: Enhancing Vision-Language Verification via Vision Experts and Iterative Training

Reinforcement Fine-Tuning (RFT) with verifiable rewards has advanced large language models but remains underexplored for Vision-Language (VL) models. The Vision-Language Reward Model (VL-RM) is key to aligning VL models by providing structured feedback, yet training effective VL-RMs faces two major challenges. First, the bootstrapping dilemma arises as high-quality training data depends on already strong VL models, creating a cycle where self-generated supervision reinforces existing biases. Second, modality bias and negative example amplification occur when VL models hallucinate incorrect visual attributes, leading to flawed preference data that further misguides training. To address these issues, we propose an iterative training framework leveraging vision experts, Chain-of-Thought (CoT) rationales, and Margin-based Rejection Sampling. Our approach refines preference datasets, enhances structured critiques, and iteratively improves reasoning. Experiments across VL-RM benchmarks demonstrate superior performance in hallucination detection and multimodal reasoning, advancing VL model alignment with reinforcement learning.

cs.CL↗

arXiv2025

A Minimalist Approach to LLM Reasoning: from Rejection Sampling to Reinforce

Reinforcement learning (RL) has become a prevailing approach for fine-tuning large language models (LLMs) on complex reasoning tasks. Among recent methods, GRPO stands out for its empirical success in training models such as DeepSeek-R1, yet the sources of its effectiveness remain poorly understood. In this work, we revisit GRPO from a reinforce-like algorithm perspective and analyze its core components. Surprisingly, we find that a simple rejection sampling baseline, RAFT, which trains only on positively rewarded samples, yields competitive performance than GRPO and PPO. Our ablation studies reveal that GRPO's main advantage arises from discarding prompts with entirely incorrect responses, rather than from its reward normalization. Motivated by this insight, we propose Reinforce-Rej, a minimal extension of policy gradient that filters both entirely incorrect and entirely correct samples. Reinforce-Rej improves KL efficiency and stability, serving as a lightweight yet effective alternative to more complex RL algorithms. We advocate RAFT as a robust and interpretable baseline, and suggest that future advances should focus on more principled designs for incorporating negative samples, rather than relying on them indiscriminately. Our findings provide guidance for future work in reward-based LLM post-training.

cs.LG↗

arXiv2025

Canonical Latent Representations in Conditional Diffusion Models

Conditional diffusion models (CDMs) have shown impressive performance across a range of generative tasks. Their ability to model the full data distribution has opened new avenues for analysis-by-synthesis in downstream discriminative learning. However, this same modeling capacity causes CDMs to entangle the class-defining features with irrelevant context, posing challenges to extracting robust and interpretable representations. To this end, we identify Canonical LAtent Representations (CLAReps), latent codes whose internal CDM features preserve essential categorical information while discarding non-discriminative signals. When decoded, CLAReps produce representative samples for each class, offering an interpretable and compact summary of the core class semantics with minimal irrelevant details. Exploiting CLAReps, we develop a novel diffusion-based feature-distillation paradigm, CaDistill. While the student has full access to the training set, the CDM as teacher transfers core class knowledge only via CLAReps, which amounts to merely 10 % of the training data in size. After training, the student achieves strong adversarial robustness and generalization ability, focusing more on the class signals instead of spurious background cues. Our findings suggest that CDMs can serve not just as image generators but also as compact, interpretable teachers that can drive robust representation learning.

cs.LG↗

arXiv2025

Thinking vs. Doing: Agents that Reason by Scaling Test-Time Interaction

The current paradigm of test-time scaling relies on generating long reasoning traces ("thinking" more) before producing a response. In agent problems that require interaction, this can be done by generating thinking traces before acting in the world. However, this process does not allow agents to acquire new information from the environment or adapt their behavior over time. In this work, we propose to scale test-time interaction, an untapped dimension of test-time scaling that increases the agent's interaction horizon to enable running rich behaviors such as exploration, backtracking, and dynamic re-planning within a single rollout. To demonstrate the promise of this scaling dimension, we study the domain of web agents. We first show that even prompting-based interaction scaling without any training can improve task success on web benchmarks non-trivially. Building on this, we introduce TTI (Test-Time Interaction), a curriculum-based online reinforcement learning (RL) approach that trains agents by adaptively adjusting their rollout lengths. Using a Gemma 3 12B model, TTI produces state-of-the-art open-source, open-data web agents on WebVoyager and WebArena benchmarks. We further show that TTI enables agents to balance exploration and exploitation adaptively. Our results establish interaction scaling as a powerful, complementary axis to scaling per-step compute, offering new avenues for training adaptive agents.

cs.LG↗

arXiv2025

EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents

Leveraging Multi-modal Large Language Models (MLLMs) to create embodied agents offers a promising avenue for tackling real-world tasks. While language-centric embodied agents have garnered substantial attention, MLLM-based embodied agents remain underexplored due to the lack of comprehensive evaluation frameworks. To bridge this gap, we introduce EmbodiedBench, an extensive benchmark designed to evaluate vision-driven embodied agents. EmbodiedBench features: (1) a diverse set of 1,128 testing tasks across four environments, ranging from high-level semantic tasks (e.g., household) to low-level tasks involving atomic actions (e.g., navigation and manipulation); and (2) six meticulously curated subsets evaluating essential agent capabilities like commonsense reasoning, complex instruction understanding, spatial awareness, visual perception, and long-term planning. Through extensive experiments, we evaluated 24 leading proprietary and open-source MLLMs within EmbodiedBench. Our findings reveal that: MLLMs excel at high-level tasks but struggle with low-level manipulation, with the best model, GPT-4o, scoring only 28.9\% on average. EmbodiedBench provides a multifaceted standardized evaluation platform that not only highlights existing challenges but also offers valuable insights to advance MLLM-based embodied agents. Our code and dataset are available at https://embodiedbench.github.io.

cs.AI↗

arXiv2025

GUI-Actor: Coordinate-Free Visual Grounding for GUI Agents

One of the principal challenges in building VLM-powered GUI agents is visual grounding, i.e., localizing the appropriate screen region for action execution based on both the visual content and the textual plans. Most existing work formulates this as a text-based coordinate generation task. However, these approaches suffer from several limitations: weak spatial-semantic alignment, inability to handle ambiguous supervision targets, and a mismatch between the dense nature of screen coordinates and the coarse, patch-level granularity of visual features extracted by models like Vision Transformers. In this paper, we propose GUI-Actor, a VLM-based method for coordinate-free GUI grounding. At its core, GUI-Actor introduces an attention-based action head that learns to align a dedicated token with all relevant visual patch tokens, enabling the model to propose one or more action regions in a single forward pass. In line with this, we further design a grounding verifier to evaluate and select the most plausible action region from the candidates proposed for action execution. Extensive experiments show that GUI-Actor outperforms prior state-of-the-art methods on multiple GUI action grounding benchmarks, with improved generalization to unseen screen resolutions and layouts. Notably, GUI-Actor-7B even surpasses UI-TARS-72B (38.1) on ScreenSpot-Pro, achieving scores of 40.7 with Qwen2-VL and 44.6 with Qwen2.5-VL as backbones. Furthermore, by incorporating the verifier, we find that fine-tuning only the newly introduced action head (~100M parameters for 7B model) while keeping the VLM backbone frozen is sufficient to achieve performance comparable to previous state-of-the-art models, highlighting that GUI-Actor can endow the underlying VLM with effective grounding capabilities without compromising its general-purpose strengths.

cs.CL↗

arXiv2025

Understanding Overadaptation in Supervised Fine-Tuning: The Role of Ensemble Methods

Supervised fine-tuning (SFT) on domain-specific data is the dominant approach for adapting foundation models to specialized tasks. However, it has been observed that SFT models tend to forget knowledge acquired during pretraining. In vision models, ensembling a pretrained model with its fine-tuned counterpart has been shown to mitigate this issue. In this work, we demonstrate that the same holds for language models, and, more strikingly, we observe an overadaptation phenomenon: the ensemble model not only retains general knowledge from the foundation model but also outperforms the fine-tuned model even on the fine-tuning domain itself. Despite the empirical success of ensembling, a theoretical understanding of its benefits remains underexplored. We develop a formal theoretical analysis of the overadaptation phenomenon. Ensembling mitigates this by balancing two primary sources of error: bias, caused by insufficient fine-tuning, and variance, introduced by overfitting to fine-tuning data. While regularization techniques aim to address this trade-off, we show that ensembling provides a more effective solution. We analyze this phenomenon in over-parameterized linear settings and demonstrate that interpolating between pretrained and fine-tuned weights significantly improves performance. These findings offer theoretical justification for the observed advantages of model ensembling, supported by empirical experiments consistent with our analysis.

cs.AI↗

arXiv2025

Transformers as Multi-task Learners: Decoupling Features in Hidden Markov Models

Transformer based models have shown remarkable capabilities in sequence learning across a wide range of tasks, often performing well on specific task by leveraging input-output examples. Despite their empirical success, a comprehensive theoretical understanding of this phenomenon remains limited. In this work, we investigate the layerwise behavior of Transformers to uncover the mechanisms underlying their multi-task generalization ability. Taking explorations on a typical sequence model, i.e, Hidden Markov Models, which are fundamental to many language tasks, we observe that: first, lower layers of Transformers focus on extracting feature representations, primarily influenced by neighboring tokens; second, on the upper layers, features become decoupled, exhibiting a high degree of time disentanglement. Building on these empirical insights, we provide theoretical analysis for the expressiveness power of Transformers. Our explicit constructions align closely with empirical observations, providing theoretical support for the Transformer's effectiveness and efficiency on sequence learning across diverse tasks.

cs.LG↗

arXiv2025

Transcriptome and Redox Proteome Reveal Temporal Scales of Carbon Metabolism Regulation in Model Cyanobacteria Under Light Disturbance

We develop a systems approach based on an energy-landscape concept to differentiate interactions involving redox activities and conformational changes of proteins and nucleic acids interactions in multi-layered protein-DNA regulatory networks under light disturbance. Our approach is a data-driven modeling workflow using a physics-informed machine learning algorithm to train a non-linear mathematical model for interpreting gene expression dynamics and to lead discovery for protein regulators using redox proteome analysis. We distinguish light-responsive elements within central carbon metabolism pathways from independent variables like circadian time using the publicly available transcriptome datasets of Synechococcus elongatus over diel cycles responding to light perturbations. Our approach provides interpretable de novo models for elucidating events of reactions in complex regulatory pathways in response to stressful disturbance from the environment. We discovered protein regulators in response to light disturbance in the proteome analysis involving shifts in protein abundance as well as cysteine redox states under constant illumination and after two hours of darkness. We discovered significant shifts in cysteine redox states in regulatory proteins such as transcription sigma factors and metabolic enzymes in the oxidative pentose phosphate pathway and the Calvin-Benson cycle, while the changes in their protein abundance were minimal. These results indicate that regulatory dynamics in reductant generation link photo-induced electron transport pathways and redox metabolic pathways with circadian rhythms through fast redox-induced conformational changes or slow expression regulations across networks.

q-bio.MN↗

arXiv2025

Decoding Cellular Temperature via Neural Network-Aided Fluorescent Thermometry

The temperature distribution within cells, especially the debates on mitochondrial temperature, has recently attracted widespread attention. Some studies have claimed that the temperature of mitochondria can reach up to 50-53 degrees Celsius. Yet others have questioned that this is due to measurement errors from fluorescent thermometry caused by other factors, like cell viscosity. Here we present a neural network-aided fluorescent thermometry and decouple the effect of cellular viscosity on temperature measurements. We found that cellular viscosity may cause significant deviations in temperature measurements. We investigated the dynamic temperature changes in different organelles within the cell under stimulation and observed a distinct temperature gradient within the cell. Eliminating the influence of viscosity, the upper limit of mitochondrial temperature does not exceed 42-43 degrees Celsius, supporting our knowledge about the inactivation temperature of enzymes. The temperature of mitochondria is closely related to their functions and morphology, such as fission and fusion. Our results help to clarify the question of "how hot are mitochondria?" and promote a better understanding on cellular thermodynamics.

physics.bio-ph↗

arXiv2025

Daunce: Data Attribution through Uncertainty Estimation

Training data attribution (TDA) methods aim to identify which training examples influence a model's predictions on specific test data most. By quantifying these influences, TDA supports critical applications such as data debugging, curation, and valuation. Gradient-based TDA methods rely on gradients and second-order information, limiting their applicability at scale. While recent random projection-based methods improve scalability, they often suffer from degraded attribution accuracy. Motivated by connections between uncertainty and influence functions, we introduce Daunce - a simple yet effective data attribution approach through uncertainty estimation. Our method operates by fine-tuning a collection of perturbed models and computing the covariance of per-example losses across these models as the attribution score. Daunce is scalable to large language models (LLMs) and achieves more accurate attribution compared to existing TDA methods. We validate Daunce on tasks ranging from vision tasks to LLM fine-tuning, and further demonstrate its compatibility with black-box model access. Applied to OpenAI's GPT models, our method achieves, to our knowledge, the first instance of data attribution on proprietary LLMs.

cs.LG↗

arXiv2025

Almost Linear Convergence under Minimal Score Assumptions: Quantized Transition Diffusion

Continuous diffusion models have demonstrated remarkable performance in data generation across various domains, yet their efficiency remains constrained by two critical limitations: (1) the local adjacency structure of the forward Markov process, which restricts long-range transitions in the data space, and (2) inherent biases introduced during the simulation of time-inhomogeneous reverse denoising processes. To address these challenges, we propose Quantized Transition Diffusion (QTD), a novel approach that integrates data quantization with discrete diffusion dynamics. Our method first transforms the continuous data distribution $p_*$ into a discrete one $q_*$ via histogram approximation and binary encoding, enabling efficient representation in a structured discrete latent space. We then design a continuous-time Markov chain (CTMC) with Hamming distance-based transitions as the forward process, which inherently supports long-range movements in the original data space. For reverse-time sampling, we introduce a \textit{truncated uniformization} technique to simulate the reverse CTMC, which can provably provide unbiased generation from $q_*$ under minimal score assumptions. Through a novel KL dynamic analysis of the reverse CTMC, we prove that QTD can generate samples with $O(d\ln^2(d/ε))$ score evaluations in expectation to approximate the $d$--dimensional target distribution $p_*$ within an $ε$ error tolerance. Our method not only establishes state-of-the-art inference efficiency but also advances the theoretical foundations of diffusion-based generative modeling by unifying discrete and continuous diffusion paradigms.

stat.ML↗