arXiv subjects

Tian Zhang

Publications and source records attributed to Tian Zhang.

At least 55 records · Page 3Linked to original sources

arXiv2025

LetheViT: Selective Machine Unlearning for Vision Transformers via Attention-Guided Contrastive Learning

Vision Transformers (ViTs) have revolutionized computer vision tasks with their exceptional performance. However, the introduction of privacy regulations such as GDPR and CCPA has brought new challenges to them. These laws grant users the right to withdraw their data, necessitating not only the deletion of data but also the complete removal of its influence from trained models. Machine unlearning emerges as a critical solution, with exact unlearning being computationally prohibitive and approximate methods offering a more practical approach. This work addresses the particularly challenging scenario of random data forgetting in ViTs, where the model must forget specific samples while retaining others, even within the same class. We first reveal the core characteristics of ViTs through selective masking experiments: when high-attention areas are masked, the model retains its recognition capability but significantly weakens its memorization ability. Based on the above insights, we propose LetheViT, a contrastive unlearning method tailored for ViTs. LetheViT uses masked image inputs to generate positive logits and original image inputs to generate negative logits, guiding the model to forget specific details while retaining the general cl category outlines. Experimental results demonstrate that LetheViT achieves state-of-the-art performance, effectively balancing privacy compliance with model efficacy.

cs.CV↗

arXiv2025

MultiKernelBench: A Multi-Platform Benchmark for Kernel Generation

The automatic generation of deep learning (DL) kernels using large language models (LLMs) has emerged as a promising approach to reduce the manual effort and hardware-specific expertise required for writing high-performance operator implementations. However, existing benchmarks for evaluating LLMs in this domain suffer from limited hardware support, coarse-grained kernel categorization, and imbalanced task coverage. To address these limitations, we introduce MultiKernelBench, the first comprehensive, multi-platform benchmark for LLM-based DL kernel generation. MultiKernelBench spans 285 tasks across 14 well-defined kernel categories and supports three major hardware platforms: Nvidia GPUs, Huawei NPUs, and Google TPUs. To enable future extensibility, we design a modular backend abstraction layer that decouples platform-specific logic from the core benchmarking infrastructure, allowing easy integration of new hardware platforms. We further propose a simple yet effective category-aware one-shot prompting method that improves generation quality by providing in-category exemplars. Through systematic evaluations of seven state-of-the-art LLMs, we reveal significant variation in task difficulty, poor generalization to platforms with less training exposure, and the effectiveness of targeted prompting strategies. MultiKernelBench is publicly available at https://github.com/wzzll123/MultiKernelBench.

cs.DC↗

arXiv2025

DFQ-ViT: Data-Free Quantization for Vision Transformers without Fine-tuning

Data-Free Quantization (DFQ) enables the quantization of Vision Transformers (ViTs) without requiring access to data, allowing for the deployment of ViTs on devices with limited resources. In DFQ, the quantization model must be calibrated using synthetic samples, making the quality of these synthetic samples crucial. Existing methods fail to fully capture and balance the global and local features within the samples, resulting in limited synthetic data quality. Moreover, we have found that during inference, there is a significant difference in the distributions of intermediate layer activations between the quantized and full-precision models. These issues lead to a severe performance degradation of the quantized model. To address these problems, we propose a pipeline for Data-Free Quantization for Vision Transformers (DFQ-ViT). Specifically, we synthesize samples in order of increasing difficulty, effectively enhancing the quality of synthetic data. During the calibration and inference stage, we introduce the activation correction matrix for the quantized model to align the intermediate layer activations with those of the full-precision model. Extensive experiments demonstrate that DFQ-ViT achieves remarkable superiority over existing DFQ methods and its performance is on par with models quantized through real data. For example, the performance of DeiT-T with 3-bit weights quantization is 4.29% higher than the state-of-the-art. Our method eliminates the need for fine-tuning, which not only reduces computational overhead but also lowers the deployment barriers for edge devices. This characteristic aligns with the principles of Green Learning by improving energy efficiency and facilitating real-world applications in resource-constrained environments.

cs.CV↗

arXiv2025

Hunyuan-TurboS: Advancing Large Language Models through Mamba-Transformer Synergy and Adaptive Chain-of-Thought

As Large Language Models (LLMs) rapidly advance, we introduce Hunyuan-TurboS, a novel large hybrid Transformer-Mamba Mixture of Experts (MoE) model. It synergistically combines Mamba's long-sequence processing efficiency with Transformer's superior contextual understanding. Hunyuan-TurboS features an adaptive long-short chain-of-thought (CoT) mechanism, dynamically switching between rapid responses for simple queries and deep "thinking" modes for complex problems, optimizing computational resources. Architecturally, this 56B activated (560B total) parameter model employs 128 layers (Mamba2, Attention, FFN) with an innovative AMF/MF block pattern. Faster Mamba2 ensures linear complexity, Grouped-Query Attention minimizes KV cache, and FFNs use an MoE structure. Pre-trained on 16T high-quality tokens, it supports a 256K context length and is the first industry-deployed large-scale Mamba model. Our comprehensive post-training strategy enhances capabilities via Supervised Fine-Tuning (3M instructions), a novel Adaptive Long-short CoT Fusion method, Multi-round Deliberation Learning for iterative improvement, and a two-stage Large-scale Reinforcement Learning process targeting STEM and general instruction-following. Evaluations show strong performance: overall top 7 rank on LMSYS Chatbot Arena with a score of 1356, outperforming leading models like Gemini-2.0-Flash-001 (1352) and o4-mini-2025-04-16 (1345). TurboS also achieves an average of 77.9% across 23 automated benchmarks. Hunyuan-TurboS balances high performance and efficiency, offering substantial capabilities at lower inference costs than many reasoning models, establishing a new paradigm for efficient large-scale pre-trained models.

cs.CL↗

arXiv2025

Efficient training for large-scale optical neural network using an evolutionary strategy and attention pruning

MZI-based block optical neural networks (BONNs), which can achieve large-scale network models, have increasingly drawn attentions. However, the robustness of the current training algorithm is not high enough. Moreover, large-scale BONNs usually contain numerous trainable parameters, resulting in expensive computation and power consumption. In this article, by pruning matrix blocks and directly optimizing the individuals in population, we propose an on-chip covariance matrix adaptation evolution strategy and attention-based pruning (CAP) algorithm for large-scale BONNs. The calculated results demonstrate that the CAP algorithm can prune 60% and 80% of the parameters for MNIST and Fashion-MNIST datasets, respectively, while only degrades the performance by 3.289% and 4.693%. Considering the influence of dynamic noise in phase shifters, our proposed CAP algorithm (performance degradation of 22.327% for MNIST dataset and 24.019% for Fashion-MNIST dataset utilizing a poor fabricated chip and electrical control with a standard deviation of 0.5) exhibits strongest robustness compared with both our previously reported block adjoint training algorithm (43.963% and 41.074%) and the covariance matrix adaptation evolution strategy (25.757% and 32.871%), respectively. Moreover, when 60% of the parameters are pruned, the CAP algorithm realizes 88.5% accuracy in experiment for the simplified MNIST dataset, which is similar to the simulation result without noise (92.1%). Additionally, we simulationally and experimentally demonstrate that using MZIs with only internal phase shifters to construct BONNs is an efficient way to reduce both the system area and the required trainable parameters. Notably, our proposed CAP algorithm show excellent potential for larger-scale network models and more complex tasks.

cs.LG↗

arXiv2025

RoNo: A novel way in generating reconfigurable on-chip nonlinear activation functions

Due to the limitations of Moore's Law and the increasing demand of computing, optical neural network (ONNs) are gradually coming to the stage as an alternative to electrical neural networks. The control of nonlinear activation functions in optical environments, as an important component of neural networks, has always been a challenge. In this work, firstly, we use inverse design tools to design a optical patterned area in silicon-carbide-on-insulator. This patterned area could generate two different nonlinear responses of the amplitude. Secondly, the patterned region is integrated with a control network to form a reconfigurable on-chip nonlinear activation function generator for wave-based analog computing. Experiment shows that neural network that uses such a system as an activation function performs well in the MNIST handwritten dataset and CIFAR-10, respectively. Compared to previous works, we propose a novel approach to generate on-chip reconfigurable activation functions in optical neural networks, which achieves compact footprint and enables high-quality activation function generation.

physics.optics↗

arXiv2025

Knowledge Distillation for Enhancing Walmart E-commerce Search Relevance Using Large Language Models

Ensuring the products displayed in e-commerce search results are relevant to users queries is crucial for improving the user experience. With their advanced semantic understanding, deep learning models have been widely used for relevance matching in search tasks. While large language models (LLMs) offer superior ranking capabilities, it is challenging to deploy LLMs in real-time systems due to the high-latency requirements. To leverage the ranking power of LLMs while meeting the low-latency demands of production systems, we propose a novel framework that distills a high performing LLM into a more efficient, low-latency student model. To help the student model learn more effectively from the teacher model, we first train the teacher LLM as a classification model with soft targets. Then, we train the student model to capture the relevance margin between pairs of products for a given query using mean squared error loss. Instead of using the same training data as the teacher model, we significantly expand the student model dataset by generating unlabeled data and labeling it with the teacher model predictions. Experimental results show that the student model performance continues to improve as the size of the augmented training data increases. In fact, with enough augmented data, the student model can outperform the teacher model. The student model has been successfully deployed in production at Walmart.com with significantly positive metrics.

cs.IR↗

arXiv2025

MSCoT: Structured Chain-of-Thought Generation for Multiple Programming Languages

With the rapid development of code intelligence, the application of multiple programming languages is becoming increasingly widespread. However, most existing code generation models mainly focus on a single or a few programming languages, resulting in unsatisfactory performance in a multilingual environment. Chain-of-Thought (CoT) reasoning can significantly improve the performance of the model without the need for retraining or fine-tuning the code generation model by reasonably decomposing complex code generation tasks into multiple subtasks and gradually deriving solutions for each subtask. Nevertheless, the existing CoT generation methods mainly concentrate on Python code, and the performance on other programming languages remains unclear. To fill this gap, we first constructed a CoT generation dataset for 12 programming languages through multi-agent technology. On this basis, we proposed a CoT generation method MSCoT applicable to multiple programming languages. By introducing CoT into the code generation large model, the performance of the code generation large model in a multilingual environment can be improved. Through large-scale empirical research, we compared the generalization abilities of MSCoT and the existing CoT generation methods on multiple programming languages and proved the effectiveness of MSCoT for multiple programming languages. In addition, we also designed a human study to prove the quality of the CoT generated by MSCoT. Finally, we opensourced the model and dataset of MSCoT to promote the research on CoT generation for multiple programming languages.

cs.SE↗

arXiv2025

ECG-Chat: A Large ECG-Language Model for Cardiac Disease Diagnosis

The success of Multimodal Large Language Models (MLLMs) in the medical auxiliary field shows great potential, allowing patients to engage in conversations using physiological signal data. However, general MLLMs perform poorly in cardiac disease diagnosis, particularly in the integration of ECG data analysis and medical report generation, mainly due to the complexity of ECG data analysis and the gap between text and ECG signal modalities. To address these issues, we propose ECG-Chat, a multitask MLLMs focused on ECG medical report generation, providing multimodal conversational capabilities based on cardiology knowledge. We propose a contrastive learning approach that integrates ECG waveform data with text reports, aligning ECG features with reports in a fine-grained manner. This method also results in an ECG encoder that excels in zero-shot report retrieval tasks. Additionally, expanding existing datasets, we constructed a 19k ECG diagnosis dataset and a 25k multi-turn dialogue dataset for training and fine-tuning ECG-Chat, which provides professional diagnostic and conversational capabilities. Furthermore, ECG-Chat can generate comprehensive ECG analysis reports through an automated LaTeX generation pipeline. We established a benchmark for the ECG report generation task and tested our model on multiple baselines. ECG-Chat achieved the best performance in classification, retrieval, and medical report generation tasks. Our code is available at https://github.com/YubaoZhao/ECG-Chat.

eess.SP↗

arXiv2025

Average-Distortion Sketching

We introduce average-distortion sketching for metric spaces. As in (worst-case) sketching, these algorithms compress points in a metric space while approximately recovering pairwise distances. The novelty is studying average-distortion: for any fixed (yet, arbitrary) distribution $μ$ over the metric, the sketch should not over-estimate distances, and it should (approximately) preserve the average distance with respect to draws from $μ$. The notion generalizes average-distortion embeddings into $\ell_1$ [Rabinovich '03, Kush-Nikolov-Tang '21] as well as data-dependent locality-sensitive hashing [Andoni-Razenshteyn '15, Andoni-Naor-Nikolov-et-al. '18], which have been recently studied in the context of nearest neighbor search. $\bullet$ For all $p \in (2, \infty)$ and any $c$ larger than a fixed constant, we give an average-distortion sketch for $([Δ]^d, \ell_p)$ with approximation $c$ and bit-complexity $\text{poly}(2^{p/c} \cdot \log(dΔ))$, which is provably impossible in (worst-case) sketching. $\bullet$ As an application, we improve on the approximation of sublinear-time data structures for nearest neighbor search over $\ell_p$ (for large $p > 2$). The prior best approximation was $O(p)$ [Andoni-Naor-Nikolov-et-al. '18, Kush-Nikolov-Tang '21], and we show it can be any $c$ larger than a fixed constant (irrespective of $p$) by using $n^{O(p/c)}$ space. We give some evidence that $2^{Ω(p/c)}$ space may be necessary by giving a lower bound on average-distortion sketches which produce a certain probabilistic certificate of farness (which our sketches crucially rely on).

cs.DS↗

arXiv2025

SABER: Model-agnostic Backdoor Attack on Chain-of-Thought in Neural Code Generation

Recent studies have proposed integrating Chain-of-Thought (CoT) reasoning to further enhance the reliability of Code Language Models (CLMs) in generating code, a step-by-step approach that breaks down complex programming tasks into manageable sub-problems. Advances in this area have introduced CoT models, specifically designed to integrate CoT reasoning effectively into language models, achieving notable improvements in code generation. Despite these advancements, the security of CoT models has not been systematically studied. In this study, we aim to fill this gap by investigating the vulnerability of CoT models to backdoor injection in code generation tasks. To address this, we propose a model-agnostic backdoor attack method SABER (Self-Attention-BasEd backdooR) based on the self-attention mechanism. SABER begins by selecting a malicious output as the backdoor using code mutation operations. It then identifies the tokens most relevant to poisoned content by analyzing self-attention scores in the CodeBERT model. Finally, it mimicks user behavior to generate adaptive and natural triggers. Our experiments on HumanEval-CoT and OpenEval-CoT test sets demonstrate that CoT models are susceptible to backdoor attacks via data poisoning. Taking the HumanEval-CoT dataset as an example, SABER achieves an ASR of 80.95%, representing an improvement of 33.33% over RIPPLe and a substantial 4.76% enhancement compared to BadPre. Further evaluations using ONION for automated detection and human studies reveal that SABER is stealthier and harder to detect, bypassing 61.90% of automated detection, with a human detection rate of just 3.17%. Our findings reveal that backdoors can be injected into CoT models to manipulate downstream code generation tasks. This highlights the urgent need for further research to understand and mitigate the security vulnerabilities in CoT models.

cs.SE↗

arXiv2025

PolicySimEval: A Benchmark for Evaluating Policy Outcomes through Agent-Based Simulation

With the growing adoption of agent-based models in policy evaluation, a pressing question arises: Can such systems effectively simulate and analyze complex social scenarios to inform policy decisions? Addressing this challenge could significantly enhance the policy-making process, offering researchers and practitioners a systematic way to validate, explore, and refine policy outcomes. To advance this goal, we introduce PolicySimEval, the first benchmark designed to evaluate the capability of agent-based simulations in policy assessment tasks. PolicySimEval aims to reflect the real-world complexities faced by social scientists and policymakers. The benchmark is composed of three categories of evaluation tasks: (1) 20 comprehensive scenarios that replicate end-to-end policy modeling challenges, complete with annotated expert solutions; (2) 65 targeted sub-tasks that address specific aspects of agent-based simulation (e.g., agent behavior calibration); and (3) 200 auto-generated tasks to enable large-scale evaluation and method development. Experiments show that current state-of-the-art frameworks struggle to tackle these tasks effectively, with the highest-performing system achieving only 24.5\% coverage rate on comprehensive scenarios, 15.04\% on sub-tasks, and 14.5\% on auto-generated tasks. These results highlight the difficulty of the task and the gap between current capabilities and the requirements for real-world policy evaluation.

cs.MA↗

arXiv2024

Global-Local Medical SAM Adaptor Based on Full Adaption

Emerging of visual language models, such as the segment anything model (SAM), have made great breakthroughs in the field of universal semantic segmentation and significantly aid the improvements of medical image segmentation, in particular with the help of Medical SAM adaptor (Med-SA). However, Med-SA still can be improved, as it fine-tunes SAM in a partial adaption manner. To resolve this problem, we present a novel global medical SAM adaptor (GMed-SA) with full adaption, which can adapt SAM globally. We further combine GMed-SA and Med-SA to propose a global-local medical SAM adaptor (GLMed-SA) to adapt SAM both globally and locally. Extensive experiments have been performed on the challenging public 2D melanoma segmentation dataset. The results show that GLMed-SA outperforms several state-of-the-art semantic segmentation methods on various evaluation metrics, demonstrating the superiority of our methods.

cs.AI↗

arXiv2024

A Cascade Dual-Decoder Model for Joint Entity and Relation Extraction

In knowledge graph construction, a challenging issue is how to extract complex (e.g., overlapping) entities and relationships from a small amount of unstructured historical data. The traditional pipeline methods are to divide the extraction into two separate subtasks, which misses the potential interaction between the two subtasks and may lead to error propagation. In this work, we propose an effective cascade dual-decoder method to extract overlapping relational triples, which includes a text-specific relation decoder and a relation-corresponded entity decoder. Our approach is straightforward and it includes a text-specific relation decoder and a relation-corresponded entity decoder. The text-specific relation decoder detects relations from a sentence at the text level. That is, it does this according to the semantic information of the whole sentence. For each extracted relation, which is with trainable embedding, the relation-corresponded entity decoder detects the corresponding head and tail entities using a span-based tagging scheme. In this way, the overlapping triple problem can be tackled naturally. We conducted experiments on a real-world open-pit mine dataset and two public datasets to verify the method's generalizability. The experimental results demonstrate the effectiveness and competitiveness of our proposed method and achieve better F1 scores under strict evaluation metrics. Our implementation is available at https://github.com/prastunlp/DualDec.

cs.CL↗

arXiv2024

Some dynamical properties of delayed weakly reversible mass-action systems

This paper focuses on the dynamical properties of delayed complex balanced systems. We first study the relationship between the stoichiometric compatibility classes of delayed and non-delayed systems. Using this relation we give another way to derive the existence of positive equilibrium in each stoichiometric compatibility class for delayed complex balanced systems. And if time delays are constant, the result can be generalized to weakly reversible networks. Also, by utilizing the Lyapunov-Krasovskii functional, we can obtain a long-time dynamical property about $ω$-limit set of the complex balanced system with constant time delays. An example is also provided to support our results.

math.DS↗

arXiv2024

Data-Dependent LSH for the Earth Mover's Distance

We give new data-dependent locality sensitive hashing schemes (LSH) for the Earth Mover's Distance ($\mathsf{EMD}$), and as a result, improve the best approximation for nearest neighbor search under $\mathsf{EMD}$ by a quadratic factor. Here, the metric $\mathsf{EMD}_s(\mathbb{R}^d,\ell_p)$ consists of sets of $s$ vectors in $\mathbb{R}^d$, and for any two sets $x,y$ of $s$ vectors the distance $\mathsf{EMD}(x,y)$ is the minimum cost of a perfect matching between $x,y$, where the cost of matching two vectors is their $\ell_p$ distance. Previously, Andoni, Indyk, and Krauthgamer gave a (data-independent) locality-sensitive hashing scheme for $\mathsf{EMD}_s(\mathbb{R}^d,\ell_p)$ when $p \in [1,2]$ with approximation $O(\log^2 s)$. By being data-dependent, we improve the approximation to $\tilde{O}(\log s)$. Our main technical contribution is to show that for any distribution $μ$ supported on the metric $\mathsf{EMD}_s(\mathbb{R}^d, \ell_p)$, there exists a data-dependent LSH for dense regions of $μ$ which achieves approximation $\tilde{O}(\log s)$, and that the data-independent LSH actually achieves a $\tilde{O}(\log s)$-approximation outside of those dense regions. Finally, we show how to "glue" together these two hashing schemes without any additional loss in the approximation. Beyond nearest neighbor search, our data-dependent LSH also gives optimal (distributional) sketches for the Earth Mover's Distance. By known sketching lower bounds, this implies that our LSH is optimal (up to $\mathrm{poly}(\log \log s)$ factors) among those that collide close points with constant probability.

cs.DS↗

arXiv2023

High performance artificial visual system with plasmon-enhanced 2D material neural network

Artificial visual systems (AVS) have gained tremendous momentum because of its huge potential in areas such as autonomous vehicles and robotics as part of artificial intelligence (AI) in recent years. However, current machine visual systems composed of complex circuits based on complementary metal oxide semiconductor (CMOS) platform usually contains photosensor array, format conversion, memory and processing module. The large amount of redundant data shuttling between each unit, resulting in large latency and high power consumption, which greatly limits the performance of the AVS. Here, we demonstrate an AVS based on a new design concept, which consists of hardware devices connected in an artificial neural network (ANN) that can simultaneously sense, pre-process and recognize optical images without latency. The Ag nanograting and the two-dimensional (2D) heterostructure integrated plasmonic phototransistor array (PPTA) constitute the hardware ANN, and its synaptic weight is determined by the adjustable regularized photoresponsivity matrix. The eye-inspired pre-processing function of the device under photoelectric synergy ensures the considerable improvement of the efficiency and accuracy of subsequent image recognition. The comprehensive performance of the proof-of-concept device demonstrates great potential for machine vision applications in terms of large dynamic range (180 dB), high speed (500 ns) and ultralow energy consumption per spike (2.4e(-17) J).

physics.optics↗

arXiv2023

A simplified method characterizing magnetic ordering modulated photo-thermoelectric response in noncentrosymmetric semimetal Ca3Ru2O7

Photo-Thermoelectric (PTE) response is usually one of the main working mechanisms for photodetectors. However, as another fast and easier way to measure thermoelectric characteristics of materials, it can also reveal important physics such as electric-phonon coupling, electron-electron correlation, etc. Recently, the spin entropy related to magnetic order transition which contributes to thermoelectric power is attracting more and more attention. Here, we demonstrate the PTE response can be reshaped when Ca3Ru2O7 undergoes meta-magnetic phase (MMP) transition driven by both temperature and magnetic field. Firstly, a sign change is observed crossing TS = 48 K and the linear polarization angle dependent PTE current maximizes along a-axis above TS while maximizes along b-axis below TS, which indicates that the antiferromagnetic spin order contributes to such spatial anisotropy. Secondly, in the temperature range of around 40 ~ 50 K, the PTE current is found to be sharply suppressed when external magnetic field is applied in plane along a-axis but is only gradually suppressed when applied field is along b-axis which gives out two critical fields. We attribute such suppression of PTE current under magnetic field to the suppression of the spin entropy in the phase transition between the antiferromagnetic state and the MMP state and the H-T phase diagrams of Ca3Ru2O7 is redrawn accordingly. Compared to previously work which trying to understand the magnetic phase transition in Ca3Ru2O7, such as neutron scattering, specific heat, and other advanced transport measurements, our work provides a more convenient yet efficient method, which may also find applications in other correlated spin materials in general.

cond-mat.str-el↗