arXiv subjects

Tong Zhang

Publications and source records attributed to Tong Zhang.

At least 181 records · Page 10Linked to original sources

arXiv2025

Towards Better Generalization via Distributional Input Projection Network

As overparameterized models become increasingly prevalent, training loss alone offers limited insight into generalization performance. While smoothness has been linked to improved generalization across various settings, directly enforcing smoothness in neural networks remains challenging. To address this, we introduce Distributional Input Projection Networks (DIPNet), a novel framework that projects inputs into learnable distributions at each layer. This distributional representation induces a smoother loss landscape with respect to the input, promoting better generalization. We provide theoretical analysis showing that DIPNet reduces both local smoothness measures and the Lipschitz constant of the network, contributing to improved generalization performance. Empirically, we validate DIPNet across a wide range of architectures and tasks, including Vision Transformers (ViTs), Large Language Models (LLMs), ResNet and MLPs. Our method consistently enhances test performance under standard settings, adversarial attacks, out-of-distribution inputs, and reasoning benchmarks. We demonstrate that the proposed input projection strategy can be seamlessly integrated into existing models, providing a general and effective approach for boosting generalization performance in modern deep learning.

cs.LG↗

arXiv2025

VideoAnchor: Reinforcing Subspace-Structured Visual Cues for Coherent Visual-Spatial Reasoning

Multimodal Large Language Models (MLLMs) have achieved impressive progress in vision-language alignment, yet they remain limited in visual-spatial reasoning. We first identify that this limitation arises from the attention mechanism: visual tokens are overshadowed by language tokens, preventing the model from consistently recognizing the same visual cues across frames. To address this challenge, we draw a novel connection between the self-expressiveness property in sparse subspace clustering and the attention mechanism in Transformers. Building on this insight, we propose VideoAnchor, a plug-and-play module that leverages subspace affinities to reinforce visual cues across frames without retraining, effectively anchoring attention to shared visual structures. Extensive experiments across benchmarks and backbone models show consistent performance gains -- $e.g.$, 3.2% and 4.6% improvements on VSI-Bench and Video-MME (spatial-related tasks) with InternVL2-8B and Qwen2.5VL-72B -- while qualitative analyses demonstrate more coherent subspace partitions and stronger visual grounding. Our codes will be made public available at https://github.com/feufhd/VideoAnchor.

cs.CV↗

arXiv2025

Reformulation is All You Need: Addressing Malicious Text Features in DNNs

Human language encompasses a wide range of intricate and diverse implicit features, which attackers can exploit to launch adversarial or backdoor attacks, compromising DNN models for NLP tasks. Existing model-oriented defenses often require substantial computational resources as model size increases, whereas sample-oriented defenses typically focus on specific attack vectors or schemes, rendering them vulnerable to adaptive attacks. We observe that the root cause of both adversarial and backdoor attacks lies in the encoding process of DNN models, where subtle textual features, negligible for human comprehension, are erroneously assigned significant weight by less robust or trojaned models. Based on it we propose a unified and adaptive defense framework that is effective against both adversarial and backdoor attacks. Our approach leverages reformulation modules to address potential malicious features in textual inputs while preserving the original semantic integrity. Extensive experiments demonstrate that our framework outperforms existing sample-oriented defense baselines across a diverse range of malicious textual features.

cs.LG↗

arXiv2025

Pre-training CLIP against Data Poisoning with Optimal Transport-based Matching and Alignment

Recent studies have shown that Contrastive Language-Image Pre-training (CLIP) models are threatened by targeted data poisoning and backdoor attacks due to massive training image-caption pairs crawled from the Internet. Previous defense methods correct poisoned image-caption pairs by matching a new caption for each image. However, the matching process relies solely on the global representations of images and captions, overlooking fine-grained features of visual and textual features. It may introduce incorrect image-caption pairs and harm the CLIP pre-training. To address their limitations, we propose an Optimal Transport-based framework to reconstruct image-caption pairs, named OTCCLIP. We propose a new optimal transport-based distance measure between fine-grained visual and textual feature sets and re-assign new captions based on the proposed optimal transport distance. Additionally, to further reduce the negative impact of mismatched pairs, we encourage the inter- and intra-modality fine-grained alignment by employing optimal transport-based objective functions. Our experiments demonstrate that OTCCLIP can successfully decrease the attack success rates of poisoning attacks. Also, compared to previous methods, OTCCLIP significantly improves CLIP's zero-shot and linear probing performance trained on poisoned datasets.

cs.CV↗

arXiv2025

GraphTorque: Torque-Driven Rewiring Graph Neural Network

Graph Neural Networks (GNNs) have emerged as powerful tools for learning from graph-structured data, leveraging message passing to diffuse information and update node representations. However, most efforts have suggested that native interactions encoded in the graph may not be friendly for this process, motivating the development of graph rewiring methods. In this work, we propose a torque-driven hierarchical rewiring strategy, inspired by the notion of torque in classical mechanics, dynamically modulating message passing to improve representation learning in heterophilous and homophilous graphs. Specifically, we define the torque by treating the feature distance as a lever arm vector and the neighbor feature as a force vector weighted by the homophily disparity between nodes. We use the metric to hierarchically reconfigure receptive field of each layer by judiciously pruning high-torque edges and adding low-torque links, suppressing the impact of irrelevant information and boosting pertinent signals during message passing. Extensive evaluations on benchmark datasets show that the proposed approach surpasses state-of-the-art rewiring methods on both heterophilous and homophilous graphs.

cs.LG↗

arXiv2025

Entropy-Regularized Process Reward Model

Large language models (LLMs) have shown promise in performing complex multi-step reasoning, yet they continue to struggle with mathematical reasoning, often making systematic errors. A promising solution is reinforcement learning (RL) guided by reward models, particularly those focusing on process rewards, which score each intermediate step rather than solely evaluating the final outcome. This approach is more effective at guiding policy models towards correct reasoning trajectories. In this work, we propose an entropy-regularized process reward model (ER-PRM) that integrates KL-regularized Markov Decision Processes (MDP) to balance policy optimization with the need to prevent the policy from shifting too far from its initial distribution. We derive a novel reward construction method based on the theoretical results. Our theoretical analysis shows that we could derive the optimal reward model from the initial policy sampling. Our empirical experiments on the MATH and GSM8K benchmarks demonstrate that ER-PRM consistently outperforms existing process reward models, achieving 1% improvement on GSM8K and 2-3% improvement on MATH under best-of-N evaluation, and more than 1% improvement under RLHF. These results highlight the efficacy of entropy-regularization in enhancing LLMs' reasoning capabilities.

cs.LG↗

arXiv2025

Generalizable Geometric Image Caption Synthesis

Multimodal large language models have various practical applications that demand strong reasoning abilities. Despite recent advancements, these models still struggle to solve complex geometric problems. A key challenge stems from the lack of high-quality image-text pair datasets for understanding geometric images. Furthermore, most template-based data synthesis pipelines typically fail to generalize to questions beyond their predefined templates. In this paper, we bridge this gap by introducing a complementary process of Reinforcement Learning with Verifiable Rewards (RLVR) into the data generation pipeline. By adopting RLVR to refine captions for geometric images synthesized from 50 basic geometric relations and using reward signals derived from mathematical problem-solving tasks, our pipeline successfully captures the key features of geometry problem-solving. This enables better task generalization and yields non-trivial improvements. Furthermore, even in out-of-distribution scenarios, the generated dataset enhances the general reasoning capabilities of multimodal large language models, yielding accuracy improvements of $2.8\%\text{-}4.8\%$ in statistics, arithmetic, algebraic, and numerical tasks with non-geometric input images of MathVista and MathVerse, along with $2.4\%\text{-}3.9\%$ improvements in Art, Design, Tech, and Engineering tasks in MMMU.

cs.AI↗

arXiv2025

Radio Map-Enabled 3D Trajectory and Communication Optimization for Low-Altitude Air-Ground Cooperation

Low-altitude economy includes the application of unmanned aerial vehicles (UAVs) serving ground robots. This paper investigates the 3-dimensional (3D) trajectory and communication optimization for low-altitude air-ground cooperation systems, where mobile unmanned ground vehicles (UGVs) upload data to UAVs. We propose a joint optimization algorithm to maximize the minimal sum-rate of UGVs while ensuring quality of service and navigation constraints. The proposed algorithm integrates a successive convex approximation (SCA)-penalty method for UGV-UAV scheduling, an SCA-based approach for UGV transmit power control, and a novel warm-start particle swarm optimization with cross mutation (WS-PSO-CM). The WS-PSO-CM leverages convex optimization results from a statistical channel model to initialize particle swarm, significantly improving the performance, compared with celebrated PSO-CM. Simulation results demonstrate that the proposed algorithm achieves a $45.8$\% higher minimal sum-rate compared to the baseline PSO-CM under the same iterations. This gain can be translated to reducing computational time by $46.7$\% of PSO-CM. Furthermore, our simulation results reveal that UAVs dynamically adjust trajectories to avoid interference by buildings, and maintain proximity to UGVs to mitigate path-loss.

cs.IT↗

arXiv2025

Accelerating LLM Inference via Dynamic KV Cache Placement in Heterogeneous Memory System

Large Language Model (LLM) inference is increasingly constrained by memory bandwidth, with frequent access to the key-value (KV) cache dominating data movement. While attention sparsity reduces some memory traffic, the relevance of past tokens varies over time, requiring the full KV cache to remain accessible and sustaining pressure on both bandwidth and capacity. With advances in interconnects such as NVLink and LPDDR5X, modern AI hardware now integrates high-bandwidth memory (HBM) with high-speed off-package DRAM, making heterogeneous memory systems a practical solution. This work investigates dynamic KV cache placement across such systems to maximize aggregated bandwidth utilization under capacity constraints. Rather than proposing a specific scheduling policy, we formulate the placement problem mathematically and derive a theoretical upper bound, revealing substantial headroom for runtime optimization. To our knowledge, this is the first formal treatment of dynamic KV cache scheduling in heterogeneous memory systems for LLM inference.

cs.AR↗

arXiv2025

ALRPHFS: Adversarially Learned Risk Patterns with Hierarchical Fast \& Slow Reasoning for Robust Agent Defense

LLM Agents are becoming central to intelligent systems. However, their deployment raises serious safety concerns. Existing defenses largely rely on "Safety Checks", which struggle to capture the complex semantic risks posed by harmful user inputs or unsafe agent behaviors - creating a significant semantic gap between safety checks and real-world risks. To bridge this gap, we propose a novel defense framework, ALRPHFS (Adversarially Learned Risk Patterns with Hierarchical Fast & Slow Reasoning). ALRPHFS consists of two core components: (1) an offline adversarial self-learning loop to iteratively refine a generalizable and balanced library of risk patterns, substantially enhancing robustness without retraining the base LLM, and (2) an online hierarchical fast & slow reasoning engine that balances detection effectiveness with computational efficiency. Experimental results demonstrate that our approach achieves superior overall performance compared to existing baselines, achieving a best-in-class average accuracy of 80% and exhibiting strong generalizability across agents and tasks.

cs.CR↗

arXiv2025

Theoretical Analysis on how Learning Rate Warmup Accelerates Convergence

Learning rate warmup is a popular and practical technique in training large-scale deep neural networks. Despite the huge success in practice, the theoretical advantages of this strategy of gradually increasing the learning rate at the beginning of the training process have not been fully understood. To resolve this gap between theory and practice, we first propose a novel family of generalized smoothness assumptions, and validate its applicability both theoretically and empirically. Under the novel smoothness assumption, we study the convergence properties of gradient descent (GD) in both deterministic and stochastic settings. It is shown that learning rate warmup consistently accelerates GD, and GD with warmup can converge at most $Θ(T)$ times faster than with a non-increasing learning rate schedule in some specific cases, providing insights into the benefits of this strategy from an optimization theory perspective.

cs.LG↗

arXiv2025

Surface reconstruction and orthogonal decoupling in SrAl4 and EuAl4

Surface-induced symmetry breaking in quantum materials can stabilize exotic electronic phases distinct from those in the bulk, yet its momentum-space manifestations remain elusive due to domain-averaging effects. Here, using angle-resolved photoemission spectroscopy (ARPES) and scanning tunneling microscopy (STM), we present a microscopic investigation of the electronic structures of SrAl4 and EuAl4, layered tetragonal intermetallic compounds that exhibit well-characterized incommensurate charge-density-wave (CDW) transitions. Below the CDW transition temperatures, we uncover linearly dispersing electronic states and pronounced unidirectional replica bands orthogonal to the bulk CDW wave vector, evidencing the emergence of an in-plane C4 symmetry-breaking electronic order that is not dictated by the bulk incommensurate CDW. STM measurements further reveal a 1 times 2 surface reconstruction with quasi-one-dimensional modulations and half-unit-cell steps, traced to ordered 50 percent Sr/Eu vacancies, which vanish irreversibly upon thermal cycling, indicating decoupled surface and bulk orders. These findings establish SrAl4 and EuAl4 as model platforms for exploring surface-confined nematicity and emergent low-dimensional phases in quantum materials.

cond-mat.mtrl-sci↗

arXiv2025

CANDY: Benchmarking LLMs' Limitations and Assistive Potential in Chinese Misinformation Fact-Checking

The effectiveness of large language models (LLMs) to fact-check misinformation remains uncertain, despite their growing use. To this end, we present CANDY, a benchmark designed to systematically evaluate the capabilities and limitations of LLMs in fact-checking Chinese misinformation. Specifically, we curate a carefully annotated dataset of ~20k instances. Our analysis shows that current LLMs exhibit limitations in generating accurate fact-checking conclusions, even when enhanced with chain-of-thought reasoning and few-shot prompting. To understand these limitations, we develop a taxonomy to categorize flawed LLM-generated explanations for their conclusions and identify factual fabrication as the most common failure mode. Although LLMs alone are unreliable for fact-checking, our findings indicate their considerable potential to augment human performance when deployed as assistive tools in scenarios. Our dataset and code can be accessed at https://github.com/SCUNLP/CANDY

cs.CL↗

arXiv2025

Breaking the HBM Bit Cost Barrier: Domain-Specific ECC for AI Inference Infrastructure

High-Bandwidth Memory (HBM) delivers exceptional bandwidth and energy efficiency for AI workloads, but its high cost per bit, driven in part by stringent on-die reliability requirements, poses a growing barrier to scalable deployment. This work explores a system-level approach to cost reduction by eliminating on-die ECC and shifting all fault management to the memory controller. We introduce a domain-specific ECC framework combining large-codeword Reed--Solomon~(RS) correction with lightweight fine-grained CRC detection, differential parity updates to mitigate write amplification, and tunable protection based on data importance. Our evaluation using LLM inference workloads shows that, even under raw HBM bit error rates up to $10^{-3}$, the system retains over 78\% of throughput and 97\% of model accuracy compared with systems equipped with ideal error-free HBM. By treating reliability as a tunable system parameter rather than a fixed hardware constraint, our design opens a new path toward low-cost, high-performance HBM deployment in AI infrastructure.

cs.AR↗

arXiv2025

StepWiser: Stepwise Generative Judges for Wiser Reasoning

As models increasingly leverage multi-step reasoning strategies to solve complex problems, supervising the logical validity of these intermediate steps has become a critical research challenge. Process reward models address this by providing step-by-step feedback, but current approaches have two major drawbacks: they typically function as classifiers without providing explanations, and their reliance on supervised fine-tuning with static datasets limits generalization. Inspired by recent advances, we reframe stepwise reward modeling from a classification task to a reasoning task itself. We thus propose a generative judge that reasons about the policy model's reasoning steps (i.e., meta-reasons), outputting thinking tokens before delivering a final verdict. Our model, StepWiser, is trained by reinforcement learning using relative outcomes of rollouts. We show it provides (i) better judgment accuracy on intermediate steps than existing methods; (ii) can be used to improve the policy model at training time; and (iii) improves inference-time search.

cs.AI↗

arXiv2025

Hyperbolic Graph Neural Networks: A Review of Methods and Applications

Graph representation learning in Euclidean space, despite its widespread adoption and proven utility in many domains, often struggles to effectively capture the inherent hierarchical and complex relational structures prevalent in real-world data, particularly for datasets exhibiting a highly non-Euclidean latent anatomy or power-law distributions. Hyperbolic geometry, with its constant negative curvature and exponential growth property, naturally accommodates such structures, offering a promising alternative for learning rich graph representations. This survey paper provides a comprehensive review of the rapidly evolving field of Hyperbolic Graph Learning (HGL). We systematically categorize and analyze existing methods broadly dividing them into (1) hyperbolic graph embedding-based techniques, (2) graph neural network-based hyperbolic models, and (3) emerging paradigms. Beyond methodologies, we extensively discuss diverse applications of HGL across multiple domains, including recommender systems, knowledge graphs, bioinformatics, and other relevant scenarios, demonstrating the broad applicability and effectiveness of hyperbolic geometry in real-world graph learning tasks. Most importantly, we identify several key challenges that serve as directions for advancing HGL, including handling complex data structures, developing geometry-aware learning objectives, ensuring trustworthy and scalable implementations, and integrating with foundation models, e.g., large language models. We highlight promising research opportunities in this exciting interdisciplinary area. A comprehensive repository can be found at https://github.com/digailab/awesome-hyperbolic-graph-learning.

cs.LG↗

arXiv2025

A Sharp KL-Convergence Analysis for Diffusion Models under Minimal Assumptions

Diffusion-based generative models have emerged as highly effective methods for synthesizing high-quality samples. Recent works have focused on analyzing the convergence of their generation process with minimal assumptions, either through reverse SDEs or Probability Flow ODEs. The best known guarantees, without any smoothness assumptions, for the KL divergence so far achieve a linear dependence on the data dimension $d$ and an inverse quadratic dependence on $\varepsilon$. In this work, we present a refined analysis that improves the dependence on $\varepsilon$. We model the generation process as a composition of two steps: a reverse ODE step, followed by a smaller noising step along the forward process. This design leverages the fact that the ODE step enables control in Wasserstein-type error, which can then be converted into a KL divergence bound via noise addition, leading to a better dependence on the discretization step size. We further provide a novel analysis to achieve the linear $d$-dependence for the error due to discretizing this Probability Flow ODE in absence of any smoothness assumptions. We show that $\tilde{O}\left(\tfrac{d\log^{3/2}(\frac{1}δ)}{\varepsilon}\right)$ steps suffice to approximate the target distribution corrupted with Gaussian noise of variance $δ$ within $O(\varepsilon^2)$ in KL divergence, improving upon the previous best result, requiring $\tilde{O}\left(\tfrac{d\log^2(\frac{1}δ)}{\varepsilon^2}\right)$ steps.

stat.ML↗

arXiv2025

Reverse Markov Learning: Multi-Step Generative Models for Complex Distributions

Learning complex distributions is a fundamental challenge in contemporary applications. Shen and Meinshausen (2024) introduced engression, a generative approach based on scoring rules that maps noise (and covariates, if available) directly to data. While effective, engression can struggle with highly complex distributions, such as those encountered in image data. In this work, we propose reverse Markov learning (RML), a framework that defines a general forward process transitioning from the target distribution to a known distribution (e.g., Gaussian) and then learns a reverse Markov process using multiple engression models. This reverse process reconstructs the target distribution step by step. This framework accommodates general forward processes, allows for dimension reduction, and naturally discretizes the generative process. In the special case of diffusion-based forward processes, RML provides an efficient discretization strategy for both training and inference in diffusion models. We further introduce an alternating sampling scheme to enhance post-training performance. Our statistical analysis establishes error bounds for RML and elucidates its advantages in estimation efficiency and flexibility in forward process design. Empirical results on simulated and climate data corroborate the theoretical findings, demonstrating the effectiveness of RML in capturing complex distributions.

cs.LG↗