arXiv subjects

Zhao Xu

Publications and source records attributed to Zhao Xu.

At least 55 records · Page 3Linked to original sources

arXiv2025

TransBench: Benchmarking Machine Translation for Industrial-Scale Applications

Machine translation (MT) has become indispensable for cross-border communication in globalized industries like e-commerce, finance, and legal services, with recent advancements in large language models (LLMs) significantly enhancing translation quality. However, applying general-purpose MT models to industrial scenarios reveals critical limitations due to domain-specific terminology, cultural nuances, and stylistic conventions absent in generic benchmarks. Existing evaluation frameworks inadequately assess performance in specialized contexts, creating a gap between academic benchmarks and real-world efficacy. To address this, we propose a three-level translation capability framework: (1) Basic Linguistic Competence, (2) Domain-Specific Proficiency, and (3) Cultural Adaptation, emphasizing the need for holistic evaluation across these dimensions. We introduce TransBench, a benchmark tailored for industrial MT, initially targeting international e-commerce with 17,000 professionally translated sentences spanning 4 main scenarios and 33 language pairs. TransBench integrates traditional metrics (BLEU, TER) with Marco-MOS, a domain-specific evaluation model, and provides guidelines for reproducible benchmark construction. Our contributions include: (1) a structured framework for industrial MT evaluation, (2) the first publicly available benchmark for e-commerce translation, (3) novel metrics probing multi-level translation quality, and (4) open-sourced evaluation tools. This work bridges the evaluation gap, enabling researchers and practitioners to systematically assess and enhance MT systems for industry-specific needs.

cs.CL↗

arXiv2025

Advancing Tool-Augmented Large Language Models: Integrating Insights from Errors in Inference Trees

Tool-augmented large language models (LLMs) leverage tools, often in the form of APIs, to improve their reasoning capabilities on complex tasks. This enables them to act as intelligent agents interacting with the real world. The recently introduced ToolLLaMA model by Qin et al. [2023] utilizes the depth-first search-based decision tree (DFSDT) mechanism for multi-step reasoning with $16000+$ real-world APIs, effectively enhancing the performance of tool-augmented LLMs compared to traditional chain reasoning mechanisms. However, their approach only employs successful paths from decision trees (also called inference trees) for supervised fine-tuning (SFT), missing out on the potential learning opportunities from failed paths. Inspired by this, we propose an inference trajectory optimization framework based on preference learning to address this limitation. We first introduce a novel method for constructing step-wise preference data from tree-like expert trajectories, which leverages the previously ignored failed explorations in the decision trees. In the subsequent training phase, we first fine-tune the LLM with successful tool-usage expert trajectories and then apply direct preference optimization (DPO) with the preference data to update the LLM's policy, resulting in our ToolPrefer-LLaMA (TP-LLaMA) model. This approach not only enhances the utilization of original expert data but also broadens the learning space of the model. Our experiments demonstrate that by obtaining insights from errors in inference trees, TP-LLaMA significantly outperforms the baselines across almost all test scenarios by a large margin and exhibits better generalization capabilities with unseen APIs. At the same time, TP-LLaMA has also demonstrated superior reasoning efficiency compared to the baselines, making it more suitable for complex tool-usage reasoning tasks.

cs.CL↗

arXiv2025

A Text-guided Protein Design Framework

Current AI-assisted protein design mainly utilizes protein sequential and structural information. Meanwhile, there exists tremendous knowledge curated by humans in the text format describing proteins' high-level functionalities. Yet, whether the incorporation of such text data can help protein design tasks has not been explored. To bridge this gap, we propose ProteinDT, a multi-modal framework that leverages textual descriptions for protein design. ProteinDT consists of three subsequent steps: ProteinCLAP which aligns the representation of two modalities, a facilitator that generates the protein representation from the text modality, and a decoder that creates the protein sequences from the representation. To train ProteinDT, we construct a large dataset, SwissProtCLAP, with 441K text and protein pairs. We quantitatively verify the effectiveness of ProteinDT on three challenging tasks: (1) over 90% accuracy for text-guided protein generation; (2) best hit ratio on 12 zero-shot text-guided protein editing tasks; (3) superior performance on four out of six protein property prediction benchmarks.

cs.LG↗

arXiv2025

Evaluating Image Caption via Cycle-consistent Text-to-Image Generation

Evaluating image captions typically relies on reference captions, which are costly to obtain and exhibit significant diversity and subjectivity. While reference-free evaluation metrics have been proposed, most focus on cross-modal evaluation between captions and images. Recent research has revealed that the modality gap generally exists in the representation of contrastive learning-based multi-modal systems, undermining the reliability of cross-modality metrics like CLIPScore. In this paper, we propose CAMScore, a cyclic reference-free automatic evaluation metric for image captioning models. To circumvent the aforementioned modality gap, CAMScore utilizes a text-to-image model to generate images from captions and subsequently evaluates these generated images against the original images. Furthermore, to provide fine-grained information for a more comprehensive evaluation, we design a three-level evaluation framework for CAMScore that encompasses pixel-level, semantic-level, and objective-level perspectives. Extensive experiment results across multiple benchmark datasets show that CAMScore achieves a superior correlation with human judgments compared to existing reference-based and reference-free metrics, demonstrating the effectiveness of the framework.

cs.CV↗

arXiv2024

UNIC-Adapter: Unified Image-instruction Adapter with Multi-modal Transformer for Image Generation

Recently, text-to-image generation models have achieved remarkable advancements, particularly with diffusion models facilitating high-quality image synthesis from textual descriptions. However, these models often struggle with achieving precise control over pixel-level layouts, object appearances, and global styles when using text prompts alone. To mitigate this issue, previous works introduce conditional images as auxiliary inputs for image generation, enhancing control but typically necessitating specialized models tailored to different types of reference inputs. In this paper, we explore a new approach to unify controllable generation within a single framework. Specifically, we propose the unified image-instruction adapter (UNIC-Adapter) built on the Multi-Modal-Diffusion Transformer architecture, to enable flexible and controllable generation across diverse conditions without the need for multiple specialized models. Our UNIC-Adapter effectively extracts multi-modal instruction information by incorporating both conditional images and task instructions, injecting this information into the image generation process through a cross-attention mechanism enhanced by Rotary Position Embedding. Experimental results across a variety of tasks, including pixel-level spatial control, subject-driven image generation, and style-image-based image synthesis, demonstrate the effectiveness of our UNIC-Adapter in unified controllable image generation.

cs.CV↗

arXiv2024

LLMLight: Large Language Models as Traffic Signal Control Agents

Traffic Signal Control (TSC) is a crucial component in urban traffic management, aiming to optimize road network efficiency and reduce congestion. Traditional TSC methods, primarily based on transportation engineering and reinforcement learning (RL), often struggle with generalization abilities across varied traffic scenarios and lack interpretability. This paper presents LLMLight, a novel framework employing Large Language Models (LLMs) as decision-making agents for TSC. Specifically, the framework begins by instructing the LLM with a knowledgeable prompt detailing real-time traffic conditions. Leveraging the advanced generalization capabilities of LLMs, LLMLight engages a reasoning and decision-making process akin to human intuition for effective traffic control. Moreover, we build LightGPT, a specialized backbone LLM tailored for TSC tasks. By learning nuanced traffic patterns and control strategies, LightGPT enhances the LLMLight framework cost-effectively. Extensive experiments conducted on ten real-world and synthetic datasets, along with evaluations by fifteen human experts, demonstrate the exceptional effectiveness, generalization ability, and interpretability of LLMLight with LightGPT, outperforming nine baseline methods and ten advanced LLMs.

cs.AI↗

arXiv2024

Bag of Tricks: Benchmarking of Jailbreak Attacks on LLMs

Although Large Language Models (LLMs) have demonstrated significant capabilities in executing complex tasks in a zero-shot manner, they are susceptible to jailbreak attacks and can be manipulated to produce harmful outputs. Recently, a growing body of research has categorized jailbreak attacks into token-level and prompt-level attacks. However, previous work primarily overlooks the diverse key factors of jailbreak attacks, with most studies concentrating on LLM vulnerabilities and lacking exploration of defense-enhanced LLMs. To address these issues, we introduced $\textbf{JailTrickBench}$ to evaluate the impact of various attack settings on LLM performance and provide a baseline for jailbreak attacks, encouraging the adoption of a standardized evaluation framework. Specifically, we evaluate the eight key factors of implementing jailbreak attacks on LLMs from both target-level and attack-level perspectives. We further conduct seven representative jailbreak attacks on six defense methods across two widely used datasets, encompassing approximately 354 experiments with about 55,000 GPU hours on A800-80G. Our experimental results highlight the need for standardized benchmarking to evaluate these attacks on defense-enhanced LLMs. Our code is available at https://github.com/usail-hkust/JailTrickBench.

cs.CR↗

arXiv2024

Equivariant Graph Network Approximations of High-Degree Polynomials for Force Field Prediction

Recent advancements in equivariant deep models have shown promise in accurately predicting atomic potentials and force fields in molecular dynamics simulations. Using spherical harmonics (SH) and tensor products (TP), these equivariant networks gain enhanced physical understanding, like symmetries and many-body interactions. Beyond encoding physical insights, SH and TP are also crucial to represent equivariant polynomial functions. In this work, we analyze the equivariant polynomial functions for the equivariant architecture, and introduce a novel equivariant network, named PACE. The proposed PACE utilizes edge booster and the Atomic Cluster Expansion (ACE) technique to approximate a greater number of $SE(3) \times S_n$ equivariant polynomial functions with enhanced degrees. As experimented in commonly used benchmarks, PACE demonstrates state-of-the-art performance in predicting atomic energy and force fields, with robust generalization capability across various geometric distributions under molecular dynamics (MD) across different temperature conditions. Our code is publicly available as part of the AIRS library https://github.com/divelab/AIRS/.

cs.LG↗

arXiv2024

JAILJUDGE: A Comprehensive Jailbreak Judge Benchmark with Multi-Agent Enhanced Explanation Evaluation Framework

Despite advancements in enhancing LLM safety against jailbreak attacks, evaluating LLM defenses remains a challenge, with current methods often lacking explainability and generalization to complex scenarios, leading to incomplete assessments (e.g., direct judgment without reasoning, low F1 score of GPT-4 in complex cases, bias in multilingual scenarios). To address this, we present JAILJUDGE, a comprehensive benchmark featuring diverse risk scenarios, including synthetic, adversarial, in-the-wild, and multilingual prompts, along with high-quality human-annotated datasets. The JAILJUDGE dataset includes over 35k+ instruction-tune data with reasoning explainability and JAILJUDGETEST, a 4.5k+ labeled set for risk scenarios, and a 6k+ multilingual set across ten languages. To enhance evaluation with explicit reasoning, we propose the JailJudge MultiAgent framework, which enables explainable, fine-grained scoring (1 to 10). This framework supports the construction of instruction-tuning ground truth and facilitates the development of JAILJUDGE Guard, an end-to-end judge model that provides reasoning and eliminates API costs. Additionally, we introduce JailBoost, an attacker-agnostic attack enhancer, and GuardShield, a moderation defense, both leveraging JAILJUDGE Guard. Our experiments demonstrate the state-of-the-art performance of JailJudge methods (JailJudge MultiAgent, JAILJUDGE Guard) across diverse models (e.g., GPT-4, Llama-Guard) and zero-shot scenarios. JailBoost and GuardShield significantly improve jailbreak attack and defense tasks under zero-shot settings, with JailBoost enhancing performance by 29.24% and GuardShield reducing defense ASR from 40.46% to 0.15%.

cs.CL↗

arXiv2024

An Interpretable Alternative to Neural Representation Learning for Rating Prediction -- Transparent Latent Class Modeling of User Reviews

Nowadays, neural network (NN) and deep learning (DL) techniques are widely adopted in many applications, including recommender systems. Given the sparse and stochastic nature of collaborative filtering (CF) data, recent works have critically analyzed the effective improvement of neural-based approaches compared to simpler and often transparent algorithms for recommendation. Previous results showed that NN and DL models can be outperformed by traditional algorithms in many tasks. Moreover, given the largely black-box nature of neural-based methods, interpretable results are not naturally obtained. Following on this debate, we first present a transparent probabilistic model that topologically organizes user and product latent classes based on the review information. In contrast to popular neural techniques for representation learning, we readily obtain a statistical, visualization-friendly tool that can be easily inspected to understand user and product characteristics from a textual-based perspective. Then, given the limitations of common embedding techniques, we investigate the possibility of using the estimated interpretable quantities as model input for a rating prediction task. To contribute to the recent debates, we evaluate our results in terms of both capacity for interpretability and predictive performances in comparison with popular text-based neural approaches. The results demonstrate that the proposed latent class representations can yield competitive predictive performances, compared to popular, but difficult-to-interpret approaches.

cs.IR↗

arXiv2024

Ovis: Structural Embedding Alignment for Multimodal Large Language Model

Current Multimodal Large Language Models (MLLMs) typically integrate a pre-trained LLM with another pre-trained vision transformer through a connector, such as an MLP, endowing the LLM with visual capabilities. However, the misalignment between two embedding strategies in MLLMs -- the structural textual embeddings based on an embedding look-up table and the continuous embeddings generated directly by the vision encoder -- makes challenges for a more seamless fusion of visual and textual information. We propose Ovis, a novel MLLM architecture designed to structurally align visual and textual embeddings. Ovis integrates an additional learnable visual embedding table into the visual encoder's process. To capture rich visual semantics, each image patch indexes the visual embedding table multiple times, resulting in a final visual embedding that is a probabilistic combination of the indexed embeddings. This structural approach mirrors the method used for generating textual embeddings. Empirical evaluations on various multimodal benchmarks show that Ovis outperforms open-source MLLMs of similar parameter scales and even surpasses the proprietary model Qwen-VL-Plus overall. These results highlight the potential of Ovis' structured visual representation for advancing MLLM architectural design and promoting more effective multimodal learning. Code, datasets, and models are available at https://github.com/AIDC-AI/Ovis.

cs.CV↗

arXiv2024

Adversarial Tuning: Defending Against Jailbreak Attacks for LLMs

Although safely enhanced Large Language Models (LLMs) have achieved remarkable success in tackling various complex tasks in a zero-shot manner, they remain susceptible to jailbreak attacks, particularly the unknown jailbreak attack. To enhance LLMs' generalized defense capabilities, we propose a two-stage adversarial tuning framework, which generates adversarial prompts to explore worst-case scenarios by optimizing datasets containing pairs of adversarial prompts and their safe responses. In the first stage, we introduce the hierarchical meta-universal adversarial prompt learning to efficiently and effectively generate token-level adversarial prompts. In the second stage, we propose the automatic adversarial prompt learning to iteratively refine semantic-level adversarial prompts, further enhancing LLM's defense capabilities. We conducted comprehensive experiments on three widely used jailbreak datasets, comparing our framework with six defense baselines under five representative attack scenarios. The results underscore the superiority of our proposed methods. Furthermore, our adversarial tuning framework exhibits empirical generalizability across various attack strategies and target LLMs, highlighting its potential as a transferable defense mechanism.

cs.CL↗

arXiv2024

Wings: Learning Multimodal LLMs without Text-only Forgetting

Multimodal large language models (MLLMs), initiated with a trained LLM, first align images with text and then fine-tune on multimodal mixed inputs. However, the MLLM catastrophically forgets the text-only instructions, which do not include images and can be addressed within the initial LLM. In this paper, we present Wings, a novel MLLM that excels in both text-only dialogues and multimodal comprehension. Analyzing MLLM attention in multimodal instructions reveals that text-only forgetting is related to the attention shifts from pre-image to post-image text. From that, we construct extra modules that act as the boosted learner to compensate for the attention shift. The complementary visual and textual learners, like "wings" on either side, are connected in parallel within each layer's attention block. Initially, image and text inputs are aligned with visual learners operating alongside the main attention, balancing focus on visual elements. Textual learners are later collaboratively integrated with attention-based routing to blend the outputs of the visual and textual learners. We design the Low-Rank Residual Attention (LoRRA) to guarantee high efficiency for learners. Our experimental results demonstrate that Wings outperforms equally-scaled MLLMs in both text-only and visual question-answering tasks. On a newly constructed Interleaved Image-Text (IIT) benchmark, Wings exhibits superior performance from text-only-rich to multimodal-rich question-answering tasks.

cs.CL↗

arXiv2024

Adapting to climate change: Long-term impact of wind resource changes on China's power system resilience

Modern society's reliance on power systems is at risk from the escalating effects of wind-related climate change. Yet, failure to identify the intricate relationship between wind-related climate risks and power systems could lead to serious short- and long-term issues, including partial or complete blackouts. Here, we develop a comprehensive framework to assess China's power system resilience across various climate change scenarios, enabling a holistic evaluation of the repercussions induced by wind-related climate change. Our findings indicate that China's current wind projects and planning strategies could be jeopardized by wind-related climate change, with up to a 12\% decline in regional wind power availability. Moreover, our results underscore a pronounced vulnerability of power system resilience amidst the rigors of hastened climate change, unveiling a potential amplification of resilience deterioration, even approaching fourfold by 2060 under the most severe scenario, relative to the 2020 benchmark. This work advocates for strategic financial deployment within the power sector aimed at climate adaptation, enhancing power system resilience to avert profound losses from long-term, wind-influenced climatic fluctuations.

eess.SY↗

arXiv2024

Applying Large Language Models to Power Systems: Potential Security Threats

Applying large language models (LLMs) to modern power systems presents a promising avenue for enhancing decision-making and operational efficiency. However, this action may also incur potential security threats, which have not been fully recognized so far. To this end, this article analyzes potential threats incurred by applying LLMs to power systems, emphasizing the need for urgent research and development of countermeasures.

cs.AI↗

arXiv2023

Efficient and Equivariant Graph Networks for Predicting Quantum Hamiltonian

We consider the prediction of the Hamiltonian matrix, which finds use in quantum chemistry and condensed matter physics. Efficiency and equivariance are two important, but conflicting factors. In this work, we propose a SE(3)-equivariant network, named QHNet, that achieves efficiency and equivariance. Our key advance lies at the innovative design of QHNet architecture, which not only obeys the underlying symmetries, but also enables the reduction of number of tensor products by 92\%. In addition, QHNet prevents the exponential growth of channel dimension when more atom types are involved. We perform experiments on MD17 datasets, including four molecular systems. Experimental results show that our QHNet can achieve comparable performance to the state of the art methods at a significantly faster speed. Besides, our QHNet consumes 50\% less memory due to its streamlined architecture. Our code is publicly available as part of the AIRS library (\url{https://github.com/divelab/AIRS}).

cs.LG↗

arXiv2023

Entire Space Cascade Delayed Feedback Modeling for Effective Conversion Rate Prediction

Conversion rate (CVR) prediction is an essential task for large-scale e-commerce platforms. However, refund behaviors frequently occur after conversion in online shopping systems, which drives us to pay attention to effective conversion for building healthier shopping services. This paper defines the probability of item purchasing without any subsequent refund as an effective conversion rate (ECVR). A simple paradigm for ECVR prediction is to decompose it into two sub-tasks: CVR prediction and post-conversion refund rate (RFR) prediction. However, RFR prediction suffers from data sparsity (DS) and sample selection bias (SSB) issues, as the refund behaviors are only available after user purchase. Furthermore, there is delayed feedback in both conversion and refund events and they are sequentially dependent, named cascade delayed feedback (CDF), which significantly harms data freshness for model training. Previous studies mainly focus on tackling DS and SSB or delayed feedback for a single event. To jointly tackle these issues in ECVR prediction, we propose an Entire space CAscade Delayed feedback modeling (ECAD) method. Specifically, ECAD deals with DS and SSB by constructing two tasks including CVR prediction and conversion \& refund rate (CVRFR) prediction using the entire space modeling framework. In addition, it carefully schedules auxiliary tasks to leverage both conversion and refund time within data to alleviate CDF. Experimental results on the offline industrial dataset and online A/B testing demonstrate the effectiveness of ECAD. In addition, ECAD has been deployed in one of the recommender systems in Alibaba, contributing to a significant improvement of ECVR.

cs.IR↗

arXiv2023

3D Molecular Geometry Analysis with 2D Graphs

Ground-state 3D geometries of molecules are essential for many molecular analysis tasks. Modern quantum mechanical methods can compute accurate 3D geometries but are computationally prohibitive. Currently, an efficient alternative to computing ground-state 3D molecular geometries from 2D graphs is lacking. Here, we propose a novel deep learning framework to predict 3D geometries from molecular graphs. To this end, we develop an equilibrium message passing neural network (EMPNN) to better capture ground-state geometries from molecular graphs. To provide a testbed for 3D molecular geometry analysis, we develop a benchmark that includes a large-scale molecular geometry dataset, data splits, and evaluation protocols. Experimental results show that EMPNN can efficiently predict more accurate ground-state 3D geometries than RDKit and other deep learning methods. Results also show that the proposed framework outperforms self-supervised learning methods on property prediction tasks.

physics.chem-ph↗