arXiv subjects

Peiran Li

Publications and source records attributed to Peiran Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

ToxiGAN: Toxic Data Augmentation via LLM-Guided Directional Adversarial Generation

Augmenting toxic language data in a controllable and class-specific manner is crucial for improving robustness in toxicity classification, yet remains challenging due to limited supervision and distributional skew. We propose ToxiGAN, a class-aware text augmentation framework that combines adversarial generation with semantic guidance from large language models (LLMs). To address common issues in GAN-based augmentation such as mode collapse and semantic drift, ToxiGAN introduces a two-step directional training strategy and leverages LLM-generated neutral texts as semantic ballast. Unlike prior work that treats LLMs as static generators, our approach dynamically selects neutral exemplars to provide balanced guidance. Toxic samples are explicitly optimized to diverge from these exemplars, reinforcing class-specific contrastive signals. Experiments on four hate speech benchmarks show that ToxiGAN achieves the strongest average performance in both macro-F1 and hate-F1, consistently outperforming traditional and LLM-based augmentation methods. Ablation and sensitivity analyses further confirm the benefits of semantic ballast and directional training in enhancing classifier robustness.

cs.CL↗

arXiv2025

GRACE: Generative Representation Learning via Contrastive Policy Optimization

Prevailing methods for training Large Language Models (LLMs) as text encoders rely on contrastive losses that treat the model as a black box function, discarding its generative and reasoning capabilities in favor of static embeddings. We introduce GRACE (Generative Representation Learning via Contrastive Policy Optimization), a novel framework that reimagines contrastive signals not as losses to be minimized, but as rewards that guide a generative policy. In GRACE, the LLM acts as a policy that produces explicit, human-interpretable rationales--structured natural language explanations of its semantic understanding. These rationales are then encoded into high-quality embeddings via mean pooling. Using policy gradient optimization, we train the model with a multi-component reward function that maximizes similarity between query positive pairs and minimizes similarity with negatives. This transforms the LLM from an opaque encoder into an interpretable agent whose reasoning process is transparent and inspectable. On MTEB benchmark, GRACE yields broad cross category gains: averaged over four backbones, the supervised setting improves overall score by 11.5% over base models, and the unsupervised variant adds 6.9%, while preserving general capabilities. This work treats contrastive objectives as rewards over rationales, unifying representation learning with generation to produce stronger embeddings and transparent rationales. The model, data and code are available at https://github.com/GasolSun36/GRACE.

cs.CL↗

arXiv2025

Re-Align: Aligning Vision Language Models via Retrieval-Augmented Direct Preference Optimization

The emergence of large Vision Language Models (VLMs) has broadened the scope and capabilities of single-modal Large Language Models (LLMs) by integrating visual modalities, thereby unlocking transformative cross-modal applications in a variety of real-world scenarios. Despite their impressive performance, VLMs are prone to significant hallucinations, particularly in the form of cross-modal inconsistencies. Building on the success of Reinforcement Learning from Human Feedback (RLHF) in aligning LLMs, recent advancements have focused on applying direct preference optimization (DPO) on carefully curated datasets to mitigate these issues. Yet, such approaches typically introduce preference signals in a brute-force manner, neglecting the crucial role of visual information in the alignment process. In this paper, we introduce Re-Align, a novel alignment framework that leverages image retrieval to construct a dual-preference dataset, effectively incorporating both textual and visual preference signals. We further introduce rDPO, an extension of the standard direct preference optimization that incorporates an additional visual preference objective during fine-tuning. Our experimental results demonstrate that Re-Align not only mitigates hallucinations more effectively than previous methods but also yields significant performance gains in general visual question-answering (VQA) tasks. Moreover, we show that Re-Align maintains robustness and scalability across a wide range of VLM sizes and architectures. This work represents a significant step forward in aligning multimodal LLMs, paving the way for more reliable and effective cross-modal applications. We release all the code in https://github.com/taco-group/Re-Align.

cs.CV↗

arXiv2025

Heavy QCD Axions at High-Energy Muon Colliders

We study the physics potential of heavy QCD axions at high-energy muon colliders. Unlike typical axion-like particles, heavy QCD axions solve the strong CP problem with phenomenology driven by the anomalous gluon ($aG\widetilde G$) couplings. Several ultraviolet scenarios are presented in which QCD axions with TeV-scale masses and decay constants arise consistently with a solution to both the strong CP problem and the axion quality problem. We perform a detailed collider analysis for both a 3 and 10~TeV muon collider, focusing on hadronic axion decays that gives rise to a dijet-resonance signature. Our projections for the axion discovery reach in the multi-TeV mass range demonstrate that a muon collider can significantly extend sensitivity to heavy QCD axions compared to existing experiments.

hep-ph↗

arXiv2025

Demystifying the Visual Quality Paradox in Multimodal Large Language Models

Recent Multimodal Large Language Models (MLLMs) excel on benchmark vision-language tasks, yet little is known about how input visual quality shapes their responses. Does higher perceptual quality of images already translate to better MLLM understanding? We conduct the first systematic study spanning leading MLLMs and a suite of vision-language benchmarks, applying controlled degradations and stylistic shifts to each image. Surprisingly, we uncover a visual-quality paradox: model, task, and even individual-instance performance can improve when images deviate from human-perceived fidelity. Off-the-shelf restoration pipelines fail to reconcile these idiosyncratic preferences. To close the gap, we introduce Visual-Quality Test-Time Tuning (VQ-TTT)-a lightweight adaptation module that: (1) inserts a learnable, low-rank kernel before the frozen vision encoder to modulate frequency content; and (2) fine-tunes only shallow vision-encoder layers via LoRA. VQ-TTT dynamically adjusts each input image in a single forward pass, aligning it with task-specific model preferences. Across the evaluated MLLMs and all datasets, VQ-TTT lifts significant average accuracy, with no external models, cached features, or extra training data. These findings redefine ``better'' visual inputs for MLLMs and highlight the need for adaptive, rather than universally ``clean'', imagery, in the new era of AI being the main data customer.

cs.CV↗

arXiv2025

SAFEFLOW: A Principled Protocol for Trustworthy and Transactional Autonomous Agent Systems

Recent advances in large language models (LLMs) and vision-language models (VLMs) have enabled powerful autonomous agents capable of complex reasoning and multi-modal tool use. Despite their growing capabilities, today's agent frameworks remain fragile, lacking principled mechanisms for secure information flow, reliability, and multi-agent coordination. In this work, we introduce SAFEFLOW, a new protocol-level framework for building trustworthy LLM/VLM-based agents. SAFEFLOW enforces fine-grained information flow control (IFC), precisely tracking provenance, integrity, and confidentiality of all the data exchanged between agents, tools, users, and environments. By constraining LLM reasoning to respect these security labels, SAFEFLOW prevents untrusted or adversarial inputs from contaminating high-integrity decisions. To ensure robustness in concurrent multi-agent settings, SAFEFLOW introduces transactional execution, conflict resolution, and secure scheduling over shared state, preserving global consistency across agents. We further introduce mechanisms, including write-ahead logging, rollback, and secure caches, that further enhance resilience against runtime errors and policy violations. To validate the performances, we built SAFEFLOWBENCH, a comprehensive benchmark suite designed to evaluate agent reliability under adversarial, noisy, and concurrent operational conditions. Extensive experiments demonstrate that agents built with SAFEFLOW maintain impressive task performance and security guarantees even in hostile environments, substantially outperforming state-of-the-art. Together, SAFEFLOW and SAFEFLOWBENCH lay the groundwork for principled, robust, and secure agent ecosystems, advancing the frontier of reliable autonomy.

cs.AI↗

arXiv2025

AutoDAN-Turbo: A Lifelong Agent for Strategy Self-Exploration to Jailbreak LLMs

In this paper, we propose AutoDAN-Turbo, a black-box jailbreak method that can automatically discover as many jailbreak strategies as possible from scratch, without any human intervention or predefined scopes (e.g., specified candidate strategies), and use them for red-teaming. As a result, AutoDAN-Turbo can significantly outperform baseline methods, achieving a 74.3% higher average attack success rate on public benchmarks. Notably, AutoDAN-Turbo achieves an 88.5 attack success rate on GPT-4-1106-turbo. In addition, AutoDAN-Turbo is a unified framework that can incorporate existing human-designed jailbreak strategies in a plug-and-play manner. By integrating human-designed strategies, AutoDAN-Turbo can even achieve a higher attack success rate of 93.4 on GPT-4-1106-turbo.

cs.CR↗

arXiv2025

Multi-label classification for multi-temporal, multi-spatial coral reef condition monitoring using vision foundation model with adapter learning

Coral reef ecosystems provide essential ecosystem services, but face significant threats from climate change and human activities. Although advances in deep learning have enabled automatic classification of coral reef conditions, conventional deep models struggle to achieve high performance when processing complex underwater ecological images. Vision foundation models, known for their high accuracy and cross-domain generalizability, offer promising solutions. However, fine-tuning these models requires substantial computational resources and results in high carbon emissions. To address these challenges, adapter learning methods such as Low-Rank Adaptation (LoRA) have emerged as a solution. This study introduces an approach integrating the DINOv2 vision foundation model with the LoRA fine-tuning method. The approach leverages multi-temporal field images collected through underwater surveys at 15 dive sites at Koh Tao, Thailand, with all images labeled according to universal standards used in citizen science-based conservation programs. The experimental results demonstrate that the DINOv2-LoRA model achieved superior accuracy, with a match ratio of 64.77%, compared to 60.34% achieved by the best conventional model. Furthermore, incorporating LoRA reduced the trainable parameters from 1,100M to 5.91M. Transfer learning experiments conducted under different temporal and spatial settings highlight the exceptional generalizability of DINOv2-LoRA across different seasons and sites. This study is the first to explore the efficient adaptation of foundation models for multi-label classification of coral reef conditions under multi-temporal and multi-spatial settings. The proposed method advances the classification of coral reef conditions and provides a tool for monitoring, conserving, and managing coral reef ecosystems.

cs.CV↗

arXiv2025

Probing Millicharged Particles at an Electron Beam Dump with Ultralow-Threshold Sensors

We propose to search for millicharged particles produced in high-intensity electron beam dumps using small ultralow-threshold sensors. As a concrete example, we consider a Skipper-CCD placed behind the beam dump in Hall A at Jefferson Lab. We compute the millicharged particle flux, including both electromagnetic cascade and meson productions emanating from an aluminum target. We find that the sensitivity of a modest 2x14 array of Skipper-CCDs can exceed the sensitivity of all existing searches for millicharged particle masses below 1.5 GeV, and is either competitive or world leading when compared to other proposed experiments. Our results demonstrate that small-scale ultralow threshold silicon devices can enhance the reach of accelerator-based experiments, while fitting comfortably within existing experimental halls.

hep-ph↗

arXiv2024

Pooling Image Datasets With Multiple Covariate Shift and Imbalance

Small sample sizes are common in many disciplines, which necessitates pooling roughly similar datasets across multiple institutions to study weak but relevant associations between images and disease outcomes. Such data often manifest shift/imbalance in covariates (i.e., secondary non-imaging data). Controlling for such nuisance variables is common within standard statistical analysis, but the ideas do not directly apply to overparameterized models. Consequently, recent work has shown how strategies from invariant representation learning provides a meaningful starting point, but the current repertoire of methods is limited to accounting for shifts/imbalances in just a couple of covariates at a time. In this paper, we show how viewing this problem from the perspective of Category theory provides a simple and effective solution that completely avoids elaborate multi-stage training pipelines that would otherwise be needed. We show the effectiveness of this approach via extensive experiments on real datasets. Further, we discuss how this style of formulation offers a unified perspective on at least 5+ distinct problem settings, from self-supervised learning to matching problems in 3D reconstruction.

cs.LG↗

arXiv2024

$W$-Boson Exotic Decay into Three Charged Leptons at the LHC

We investigate the $W$ boson's exotic decay channel, $W \rightarrow \ell\ell\ell ν$, at the LHC. Although the four-body final states suppress the decay branching ratio, the large production of $W$ bosons makes detecting and precisely measuring this decay probability entirely feasible. Our simulation study indicates that this tiny branching ratio can be measured with sub-percent precision at the HL-LHC. This decay channel can also constrain Standard Model extensions. Using the $ L_μ-L_τ$ model as a benchmark, we find that the current bound on the gauge coupling for $Z'$ mass in the range of $[4,75]$ GeV can significantly improve.

hep-ph↗

arXiv2024

GUI Action Narrator: Where and When Did That Action Take Place?

The advent of Multimodal LLMs has significantly enhanced image OCR recognition capabilities, making GUI automation a viable reality for increasing efficiency in digital tasks. One fundamental aspect of developing a GUI automation system is understanding primitive GUI actions. This comprehension is crucial as it enables agents to learn from user demonstrations, an essential element of automation. To rigorously evaluate such capabilities, we developed a video captioning benchmark for GUI actions, comprising 4,189 diverse video captioning samples. This task presents unique challenges compared to natural scene video captioning: 1) GUI screenshots typically contain denser information than natural scenes, and 2) events within GUIs are subtler and occur more rapidly, requiring precise attention to the appropriate time span and spatial region for accurate understanding. To address these challenges, we introduce our GUI action dataset \textbf{Act2Cap} as well as a simple yet effective framework, \textbf{GUI Narrator}, for GUI video captioning that utilizes the cursor as a visual prompt to enhance the interpretation of high-resolution screenshots. Specifically, a cursor detector is trained on our dataset, and a multimodal LLM model with mechanisms for selecting keyframes and key regions generates the captions. Experimental results indicate that even for today's most advanced multimodal models, such as GPT-4o, the task remains highly challenging. Additionally, our evaluations show that our strategy effectively enhances model performance, whether integrated into the fine-tuning of open-source models or employed as a prompting strategy in closed-source models.

cs.CV↗

arXiv2024

Higgs Width and Couplings at High Energy Muon Colliders with Forward Muon Detection

We propose a novel method using the $ZZ$-fusion channel and forward muon detection at high-energy muon colliders to address the challenge of the Higgs coupling-width degeneracy. Our approach enables inclusive Higgs rate measurement to 0.75% at 10~TeV muon collider, breaking the coupling-width degeneracy. Results indicate the potential to refine Higgs coupling to sub-percent levels and estimate its total width within (-0.41%, +2.1%). Key insights include the effectiveness of forward muon tagging in signal-background separation despite broad recoil mass distribution due to muon energy reconstruction and beam energy spread. The study emphasizes the significance of muon rapidity coverage up to $|η(μ)|<6$, enhancing measurement precision. Our findings highlight the unique capabilities of high-energy lepton colliders for model-independent Higgs coupling determination and lay the groundwork for future advancements in muon collider technology and Higgs physics research.

hep-ph↗

arXiv2024

ASSISTGUI: Task-Oriented Desktop Graphical User Interface Automation

Graphical User Interface (GUI) automation holds significant promise for assisting users with complex tasks, thereby boosting human productivity. Existing works leveraging Large Language Model (LLM) or LLM-based AI agents have shown capabilities in automating tasks on Android and Web platforms. However, these tasks are primarily aimed at simple device usage and entertainment operations. This paper presents a novel benchmark, AssistGUI, to evaluate whether models are capable of manipulating the mouse and keyboard on the Windows platform in response to user-requested tasks. We carefully collected a set of 100 tasks from nine widely-used software applications, such as, After Effects and MS Word, each accompanied by the necessary project files for better evaluation. Moreover, we propose an advanced Actor-Critic Embodied Agent framework, which incorporates a sophisticated GUI parser driven by an LLM-agent and an enhanced reasoning mechanism adept at handling lengthy procedural tasks. Our experimental results reveal that our GUI Parser and Reasoning mechanism outshine existing methods in performance. Nevertheless, the potential remains substantial, with the best model attaining only a 46% success rate on our benchmark. We conclude with a thorough analysis of the current methods' limitations, setting the stage for future breakthroughs in this domain.

cs.CV↗

arXiv2023

Learning to Generate Pseudo Personal Mobility

The importance of personal mobility data is widely recognized in various fields. However, the utilization of real personal mobility data raises privacy concerns. Therefore, it is crucial to generate pseudo personal mobility data that accurately reflects real-world mobility patterns while safeguarding user privacy. Nevertheless, existing methods for generating pseudo mobility data, such as mechanism-based and deep-learning-based approaches, have limitations in capturing sufficient individual heterogeneity. To address these gaps, taking pseudo-person(avatar) as ground-zero, a novel individual-based human mobility generator called GeoAvatar has been proposed - which considers individual heterogeneity in spatial and temporal decision-making, incorporates demographic characteristics, and provides interpretability. Our method utilizes a deep generative model to simulate heterogeneous individual life patterns, a reliable labeler for inferring individual demographic characteristics, and a Bayesian approach for generating spatial choices. Through our method, we have achieved the generation of heterogeneous individual human mobility data without accessing individual-level personal information, with good quality - we evaluated the proposed method based on physical features, activity patterns, and spatial-temporal characteristics, demonstrating its good performance, compared to mechanism-based modeling and black-box deep learning approaches. Furthermore, this method maintains extensibility for broader applications, making it a promising paradigm for generating human mobility data.

cs.CY↗

arXiv2023

Heavy Neutral Leptons at Muon Colliders

The future high-energy muon colliders, featuring both high energy and low background, could play a critical role in our searches for new physics. The smallness of neutrino mass is a puzzle of particle physics. Broad classes of solutions to the neutrino puzzles can be best tested by seeking the partners of SM light neutrinos, dubbed as heavy neutral leptons (HNLs), at muon colliders. We can parametrize HNLs in terms of the mass $m_N$ and the mixing angle with $\ell$-flavor $U_\ell$. In this work, we focus on the regime $m_N > O(100)$ GeV and study the projected sensitivities on the $|U_\ell|^2 - m_N$ plane with the full-reconstructable HNL decay into a hadronic $W$ and a charged lepton. The projected reach in $|U_\ell|^2$ leads to the best sensitivities in the TeV realm.

hep-ph↗

arXiv2022

Small World Model for scaling up prediction result based on SEIR model

Data-driven epidemic simulation helps better policymaking. Compared with macro-scale simulations driven by statistical data, individual-level GPS data can afford finer and spatialized results. However, the big GPS data, usually collected from mobile phone users, cannot cover all populations. Therefore, this study proposes a Small World Model, to map the results from the "small world" (simulation with partially sampled data) to the real world. Based on the basic principles of disease transmission, this study derives two parameters: a time scaling factor to map the simulated period to the real period, and an amount scaling factor to map the simulated infected number to the real infected number. It is believed that this model could convert the simulation of the "small world" into the state of the real world, and analyze the effectiveness of different mobility restriction policies.

cs.CY↗

arXiv2021

Effective Metagraph-based Life Pattern Clustering with Big Human Mobility Data

Life pattern clustering is essential for abstracting the groups' characteristics of daily mobility patterns and activity regularity. Based on millions of GPS records, this paper proposed a framework on the life pattern clustering which can efficiently identify the groups have similar life pattern. The proposed method can retain original features of individual life pattern data without aggregation. Metagraph-based data structure is proposed for presenting the diverse life pattern. Spatial-temporal similarity includes significant places semantics, time sequential properties and frequency are integrated into this data structure, which captures the uncertainty of an individual and the diversities between individuals. Non-negative-factorization-based method was utilized for reducing the dimension. The results show that our proposed method can effectively identify the groups have similar life pattern and takes advantages in computation efficiency and robustness comparing with the traditional method. We revealed the representative life pattern groups and analyzed the group characteristics of human life patterns during different periods and different regions. We believe our work will help in future infrastructure planning, services improvement and policies making related to urban and transportation, thus promoting a humanized and sustainable city.

cs.CY↗