arXiv subjects

Xiao Liu

Publications and source records attributed to Xiao Liu.

At least 127 records · Page 7Linked to original sources

arXiv2026

VisionReward: Fine-Grained Multi-Dimensional Human Preference Learning for Image and Video Generation

Visual generative models have achieved remarkable progress in synthesizing photorealistic images and videos, yet aligning their outputs with human preferences across critical dimensions remains a persistent challenge. Though reinforcement learning from human feedback offers promise for preference alignment, existing reward models for visual generation face limitations, including black-box scoring without interpretability and potentially resultant unexpected biases. We present VisionReward, a general framework for learning human visual preferences in both image and video generation. Specifically, we employ a hierarchical visual assessment framework to capture fine-grained human preferences, and leverages linear weighting to enable interpretable preference learning. Furthermore, we propose a multi-dimensional consistent strategy when using VisionReward as a reward model during preference optimization for visual generation. Experiments show that VisionReward can significantly outperform existing image and video reward models on both machine metrics and human evaluation. Notably, VisionReward surpasses VideoScore by 17.2% in preference prediction accuracy, and text-to-video models with VisionReward achieve a 31.6% higher pairwise win rate compared to the same models using VideoScore. All code and datasets are provided at https://github.com/THUDM/VisionReward.

cs.CV↗

arXiv2026

GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning

We present GLM-4.1V-Thinking, GLM-4.5V, and GLM-4.6V, a family of vision-language models (VLMs) designed to advance general-purpose multimodal understanding and reasoning. In this report, we share our key findings in the development of the reasoning-centric training framework. We first develop a capable vision foundation model with significant potential through large-scale pre-training, which arguably sets the upper bound for the final performance. We then propose Reinforcement Learning with Curriculum Sampling (RLCS) to unlock the full potential of the model, leading to comprehensive capability enhancement across a diverse range of tasks, including STEM problem solving, video understanding, content recognition, coding, grounding, GUI-based agents, and long document interpretation. In a comprehensive evaluation across 42 public benchmarks, GLM-4.5V achieves state-of-the-art performance on nearly all tasks among open-source models of similar size, and demonstrates competitive or even superior results compared to closed-source models such as Gemini-2.5-Flash on challenging tasks including Coding and GUI Agents. Meanwhile, the smaller GLM-4.1V-9B-Thinking remains highly competitive-achieving superior results to the much larger Qwen2.5-VL-72B on 29 benchmarks. We open-source both GLM-4.1V-9B-Thinking and GLM-4.5V. We further introduce the GLM-4.6V series, open-source multimodal models with native tool use and a 128K context window. A brief overview is available at https://z.ai/blog/glm-4.6v. Code, models and more information are released at https://github.com/zai-org/GLM-V.

cs.CV↗

arXiv2025

Sigma-MoE-Tiny Technical Report

Mixture-of-Experts (MoE) has emerged as a promising paradigm for foundation models due to its efficient and powerful scalability. In this work, we present Sigma-MoE-Tiny, an MoE language model that achieves the highest sparsity compared to existing open-source models. Sigma-MoE-Tiny employs fine-grained expert segmentation with up to 96 experts per layer, while activating only one expert for each token, resulting in 20B total parameters with just 0.5B activated. The major challenge introduced by such extreme sparsity lies in expert load balancing. We find that the widely-used load balancing loss tends to become ineffective in the lower layers under this setting. To address this issue, we propose a progressive sparsification schedule aiming to balance expert utilization and training stability. Sigma-MoE-Tiny is pre-trained on a diverse and high-quality corpus, followed by post-training to further unlock its capabilities. The entire training process remains remarkably stable, with no occurrence of irrecoverable loss spikes. Comprehensive evaluations reveal that, despite activating only 0.5B parameters, Sigma-MoE-Tiny achieves top-tier performance among counterparts of comparable or significantly larger scale. In addition, we provide an in-depth discussion of load balancing in highly sparse MoE models, offering insights for advancing sparsity in future MoE architectures. Project page: https://qghuxmu.github.io/Sigma-MoE-Tiny Code: https://github.com/microsoft/ltp-megatron-lm

cs.CL↗

arXiv2025

CFLight: Enhancing Safety with Traffic Signal Control through Counterfactual Learning

Traffic accidents result in millions of injuries and fatalities globally, with a significant number occurring at intersections each year. Traffic Signal Control (TSC) is an effective strategy for enhancing safety at these urban junctures. Despite the growing popularity of Reinforcement Learning (RL) methods in optimizing TSC, these methods often prioritize driving efficiency over safety, thus failing to address the critical balance between these two aspects. Additionally, these methods usually need more interpretability. CounterFactual (CF) learning is a promising approach for various causal analysis fields. In this study, we introduce a novel framework to improve RL for safety aspects in TSC. This framework introduces a novel method based on CF learning to address the question: ``What if, when an unsafe event occurs, we backtrack to perform alternative actions, and will this unsafe event still occur in the subsequent period?'' To answer this question, we propose a new structure causal model to predict the result after executing different actions, and we propose a new CF module that integrates with additional ``X'' modules to promote safe RL practices. Our new algorithm, CFLight, which is derived from this framework, effectively tackles challenging safety events and significantly improves safety at intersections through a near-zero collision control strategy. Through extensive numerical experiments on both real-world and synthetic datasets, we demonstrate that CFLight reduces collisions and improves overall traffic performance compared to conventional RL methods and the recent safe RL model. Moreover, our method represents a generalized and safe framework for RL methods, opening possibilities for applications in other domains. The data and code are available in the github https://github.com/AdvancedAI-ComplexSystem/SmartCity/tree/main/CFLight.

cs.LG↗

arXiv2025

SIGMA: An AI-Empowered Training Stack on Early-Life Hardware

An increasing variety of AI accelerators is being considered for large-scale training. However, enabling large-scale training on early-life AI accelerators faces three core challenges: frequent system disruptions and undefined failure modes that undermine reliability; numerical errors and training instabilities that threaten correctness and convergence; and the complexity of parallelism optimization combined with unpredictable local noise that degrades efficiency. To address these challenges, SIGMA is an open-source training stack designed to improve the reliability, stability, and efficiency of large-scale distributed training on early-life AI hardware. The core of this initiative is the LUCIA TRAINING PLATFORM (LTP), the system optimized for clusters with early-life AI accelerators. Since its launch in March 2025, LTP has significantly enhanced training reliability and operational productivity. Over the past five months, it has achieved an impressive 94.45% effective cluster accelerator utilization, while also substantially reducing node recycling and job-recovery times. Building on the foundation of LTP, the LUCIA TRAINING FRAMEWORK (LTF) successfully trained SIGMA-MOE, a 200B MoE model, using 2,048 AI accelerators. This effort delivered remarkable stability and efficiency outcomes, achieving 21.08% MFU, state-of-the-art downstream accuracy, and encountering only one stability incident over a 75-day period. Together, these advances establish SIGMA, which not only tackles the critical challenges of large-scale training but also establishes a new benchmark for AI infrastructure and platform innovation, offering a robust, cost-effective alternative to prevailing established accelerator stacks and significantly advancing AI capabilities and scalability. The source code of SIGMA is available at https://github.com/microsoft/LuciaTrainingPlatform.

cs.DC↗

arXiv2025

Attacking All Tasks at Once Using Adversarial Examples in Multi-Task Learning

Visual content understanding frequently relies on multi-task models to extract robust representations of a single visual input for multiple downstream tasks. However, in comparison to extensively studied single-task models, the adversarial robustness of multi-task models has received significantly less attention and many questions remain unclear: 1) How robust are multi-task models to single task adversarial attacks, 2) Can adversarial attacks be designed to simultaneously attack all tasks in a multi-task model, and 3) How does parameter sharing across tasks affect multi-task model robustness to adversarial attacks? This paper aims to answer these questions through careful analysis and rigorous experimentation. First, we analyze the inherent drawbacks of two commonly-used adaptations of single-task white-box attacks in attacking multi-task models. We then propose a novel attack framework, Dynamic Gradient Balancing Attack (DGBA). Our framework poses the problem of attacking all tasks in a multi-task model as an optimization problem that can be efficiently solved through integer linear programming. Extensive evaluation on two popular MTL benchmarks, NYUv2 and Tiny-Taxonomy, demonstrates the effectiveness of DGBA compared to baselines in attacking both clean and adversarially trained multi-task models. Our results also reveal a fundamental trade-off between improving task accuracy via parameter sharing across tasks and undermining model robustness due to increased attack transferability from parameter sharing.

cs.LG↗

arXiv2025

Tunable dual-band atomic mirror based on subwavelength atomic arrays under electromagnetically induced transparency

Subwavelength atomic arrays offer a powerful platform for engineering cooperative light-matter interactions and enabling quantum metasurfaces. We demonstrate that a two-dimensional array of three-level atoms operating under electromagnetically induced transparency can function as a tunable dual-band atomic mirror, where two independently controllable reflection bands emerge from the collective optical responses mediated by dipole-dipole interactions. These resonances yield dual reflection bands with asymmetric linewidths, whose spectral positions and bandwidths can be tuned through the control-field parameters, dipole orientation, incident geometry, and lattice constant. We further identify the conditions under which additional diffraction orders emerge, which delineate the operational and tunable range of the atomic mirror via its collective-mode structure. This scheme provides a fully tunable dual-band atomic mirror operating across broad frequency and angular ranges, offering a practical and experimentally accessible pathway toward reconfigurable photonic elements in atomic-array platforms at low energy levels.

quant-ph↗

arXiv2025

BMGQ: A Bottom-up Method for Generating Complex Multi-hop Reasoning Questions from Semi-structured Data

Building training-ready multi-hop question answering (QA) datasets that truly stress a model's retrieval and reasoning abilities remains highly challenging recently. While there have been a few recent evaluation datasets that capture the characteristics of hard-to-search but easy-to-verify problems -- requiring the integration of ambiguous, indirect, and cross-domain cues -- these data resources remain scarce and are mostly designed for evaluation, making them unsuitable for supervised fine-tuning (SFT) or reinforcement learning (RL). Meanwhile, manually curating non-trivially retrievable questions -- where answers cannot be found through a single direct query but instead require multi-hop reasoning over oblique and loosely connected evidence -- incurs prohibitive human costs and fails to scale, creating a critical data bottleneck for training high-capability retrieval-and-reasoning agents. To address this, we present BMGQ, a bottom-up automated method for generating high-difficulty, training-ready multi-hop questions from semi-structured knowledge sources. The BMGQ system (i) grows diverse, logically labeled evidence clusters through Natural Language Inference (NLI)-based relation typing and diversity-aware expansion; (ii) applies reverse question construction to compose oblique cues so that isolated signals are underinformative but their combination uniquely identifies the target entity; and (iii) enforces quality with a two-step evaluation pipeline that combines multi-model consensus filtering with structured constraint decomposition and evidence-based matching. The result is a scalable process that yields complex, retrieval-resistant yet verifiable questions suitable for SFT/RL training as well as challenging evaluation, substantially reducing human curation effort while preserving the difficulty profile of strong evaluation benchmarks.

cs.AI↗

arXiv2025

Parallel Unlearning in Inherited Model Networks

Unlearning is challenging in generic learning frameworks with the continuous growth and updates of models exhibiting complex inheritance relationships. This paper presents a novel unlearning framework that enables fully parallel unlearning among models exhibiting inheritance. We use a chronologically Directed Acyclic Graph (DAG) to capture various unlearning scenarios occurring in model inheritance networks. Central to our framework is the Fisher Inheritance Unlearning (FIUn) method, designed to enable efficient parallel unlearning within the DAG. FIUn utilizes the Fisher Information Matrix (FIM) to assess the significance of model parameters for unlearning tasks and adjusts them accordingly. To handle multiple unlearning requests simultaneously, we propose the Merging-FIM (MFIM) function, which consolidates FIMs from multiple upstream models into a unified matrix. This design supports all unlearning scenarios captured by the DAG, enabling one-shot removal of inherited knowledge while significantly reducing computational overhead. Experiments confirm the effectiveness of our unlearning framework. For single-class tasks, it achieves complete unlearning with 0% accuracy for unlearned labels while maintaining 94.53% accuracy for retained labels. For multi-class tasks, the accuracy is 1.07% for unlearned labels and 84.77% for retained labels. Our framework accelerates unlearning by 99% compared to alternative methods. Code is in https://github.com/MJLee00/Parallel-Unlearning-in-Inherited-Model-Networks.

cs.LG↗

arXiv2025

Classification of autoimmune diseases from Peripheral blood TCR repertoires by multimodal multi-instance learning

T cell receptor (TCR) repertoires encode critical immunological signatures for autoimmune diseases, yet their clinical application remains limited by sequence sparsity and low witness rates. We developed EAMil, a multi-instance deep learning framework that leverages TCR sequencing data to diagnose systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA) with exceptional accuracy. By integrating PrimeSeq feature extraction with ESMonehot encoding and enhanced gate attention mechanisms, our model achieved state-of-the-art performance with AUCs of 98.95% for SLE and 97.76% for RA. EAMil successfully identified disease-associated genes with over 90% concordance with established differential analyses and effectively distinguished disease-specific TCR genes. The model demonstrated robustness in classifying multiple disease categories, utilizing the SLEDAI score to stratify SLE patients by disease severity as well as to diagnose the site of damage in SLE patients, and effectively controlling for confounding factors such as age and gender. This interpretable framework for immune receptor analysis provides new insights for autoimmune disease detection and classification with broad potential clinical applications across immune-mediated conditions.

cs.LG↗

arXiv2025

High-Dimensional Linear Bandits under Stochastic Latent Heterogeneity

This paper addresses the critical challenge of stochastic latent heterogeneity in online decision-making, where individuals' responses to actions vary not only with observable contexts but also with unobserved, randomly realized subgroups. Existing data-driven approaches largely capture observable heterogeneity through contextual features but fail when the sources of variation are latent and stochastic. We propose a latent heterogeneous bandit framework that explicitly models probabilistic subgroup membership and group-specific reward functions, using promotion targeting as a motivating example. Our phased EM-greedy algorithm jointly learns latent group probabilities and reward parameters in high dimensions, achieving optimal estimation and classification guarantees. Our analysis reveals a new phenomenon unique to decision-making with stochastic latent subgroups: randomness in group realizations creates irreducible classification uncertainty, making sub-linear regret against a fully informed strong oracle fundamentally impossible. We establish matching upper and minimax lower bounds for both the strong and regular regrets, corresponding, respectively, to oracles with and without access to realized group memberships. The strong regret necessarily grows linearly, while the regular regret achieves a minimax-optimal sublinear rate. These findings uncover a fundamental stochastic barrier in online decision-making and point to potential remedies through simple strategic interventions and mechanism-design-based elicitation of latent information.

cs.LG↗

arXiv2025

From Single to Societal: Analyzing Persona-Induced Bias in Multi-Agent Interactions

Large Language Model (LLM)-based multi-agent systems are increasingly used to simulate human interactions and solve collaborative tasks. A common practice is to assign agents with personas to encourage behavioral diversity. However, this raises a critical yet underexplored question: do personas introduce biases into multi-agent interactions? This paper presents a systematic investigation into persona-induced biases in multi-agent interactions, with a focus on social traits like trustworthiness (how an agent's opinion is received by others) and insistence (how strongly an agent advocates for its opinion). Through a series of controlled experiments in collaborative problem-solving and persuasion tasks, we reveal that (1) LLM-based agents exhibit biases in both trustworthiness and insistence, with personas from historically advantaged groups (e.g., men and White individuals) perceived as less trustworthy and demonstrating less insistence; and (2) agents exhibit significant in-group favoritism, showing a higher tendency to conform to others who share the same persona. These biases persist across various LLMs, group sizes, and numbers of interaction rounds, highlighting an urgent need for awareness and mitigation to ensure the fairness and reliability of multi-agent systems.

cs.MA↗

arXiv2025

Efficient integrated quantum memory for light

Scalable implementation of quantum networks and photonic processors require integrated photonic memories with high efficiency, yet current integrated systems have been limited to storage efficiencies below 27.8%. Here, we demonstrate highly efficient integrated quantum memories based on rare-earth-iondoped crystals coupled with impedance-matched microcavities, realized in two novel architectures: 200-micrometer-thin membranes of Eu3+:Y2SiO5 integrated with fiber-based microcavities, and waveguide-based cavities fabricated using femtosecond lasers. Our approach achieves reliable integrated quantum storage with record efficiencies of 80.3(7)% for weak coherent pulses and 69.8(1.6)% for telecom-heralded single photons, alongside the storage of 20 temporal modes with an average efficiency of 51.3(2)%. Moreover, the thin-membrane Eu3+:Y2SiO5 architecture enables spectrally tunable efficient quantum storage via variable strain, providing a flexible interface for quantum networks. By combining high efficiency, large multimode capacity, and tunability, our devices establish a versatile hardware foundation for scalable quantum repeaters and chip-scale photonic processors.

quant-ph↗

arXiv2025

Introducing LongCat-Flash-Thinking: A Technical Report

We present LongCat-Flash-Thinking, an efficient 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model. Its advanced capabilities are cultivated through a meticulously crafted training process, beginning with long Chain-of-Thought (CoT) data cold-start and culminating in large-scale Reinforcement Learning (RL). We first employ a well-designed cold-start training strategy, which significantly enhances the reasoning potential and equips the model with specialized skills in both formal and agentic reasoning. Then, a core innovation is our domain-parallel training scheme, which decouples optimization across distinct domains (e.g., STEM, Code, Agentic) and subsequently fuses the resulting expert models into a single, nearly Pareto-optimal model. This entire process is powered by our Dynamic ORchestration for Asynchronous rollout (DORA) system, a large-scale RL framework that delivers a greater than threefold training speedup over synchronous methods on tens of thousands of accelerators. As a result, LongCat-Flash-Thinking achieves state-of-the-art performance among open-source models on a suite of complex reasoning tasks. The model exhibits exceptional efficiency in agentic reasoning, reducing average token consumption by 64.5% (from 19, 653 to 6, 965) on AIME-25, without degrading task accuracy. We release LongCat-Flash-Thinking to promote further advances in reasoning systems and agentic AI research.

cs.AI↗

arXiv2025

Understanding and Optimizing Agentic Workflows via Shapley value

Agentic workflows have become the dominant paradigm for building complex AI systems, orchestrating specialized components, such as planning, reasoning, action execution, and reflection, to tackle sophisticated real-world tasks. However, systematically analyzing and optimizing these workflows remains challenging due to intricate component interdependencies and the lack of principled attribution methods. In this work, we introduce ShapleyFlow, the first framework that employs cooperative game theory to analyze and optimize agentic workflows. By applying the Shapley value to evaluate all possible component configurations, ShapleyFlow enables fine-grained attribution of each component's contribution and facilitates the identification of task-specific optimal configurations. Through a constructed dataset evaluated across 7 scenarios, such as navigation, math and OS, we demonstrate 3 key contributions: (1) Theoretical Framework: a principled game-theoretic approach for the attribution of contributions in agentic workflows. (2) Optimal Workflow Discovery: ShapleyFlow identifies task-specific component configurations that consistently outperform workflows relying on a single LLM across all tested tasks. (3) Comprehensive Analysis: we construct and analyze over 1,500 tasks, providing actionable insights and design guidelines for optimizing workflows across multiple domains.

cs.AI↗

arXiv2025

Beyond Length: Quantifying Long-Range Information for Long-Context LLM Pretraining Data

Long-context language models unlock advanced capabilities in reasoning, code generation, and document summarization by leveraging dependencies across extended spans of text. However, a significant portion of readily available long-text data lacks meaningful long-distance dependencies; most spans can be predicted using only local context. Training on such data is inefficient, making careful data selection crucial. Therefore, we introduce LongFilter, a framework for curating training data tailored to long-context pretraining. LongFilter measures the information gain provided by extended context by contrasting model predictions under long-context versus short-context settings, thereby identifying samples where long-range dependencies are essential. Experiments with LLaMA-3-8B, extending its context length from 8K to 64K, show that LongFilter efficiently selects high-quality data and yields substantial improvements on benchmarks such as HELMET, LongBench, and RULER.

cs.CL↗

arXiv2025

Self-Calibrated Consistency can Fight Back for Adversarial Robustness in Vision-Language Models

Pre-trained vision-language models (VLMs) such as CLIP have demonstrated strong zero-shot capabilities across diverse domains, yet remain highly vulnerable to adversarial perturbations that disrupt image-text alignment and compromise reliability. Existing defenses typically rely on adversarial fine-tuning with labeled data, limiting their applicability in zero-shot settings. In this work, we identify two key weaknesses of current CLIP adversarial attacks -- lack of semantic guidance and vulnerability to view variations -- collectively termed semantic and viewpoint fragility. To address these challenges, we propose Self-Calibrated Consistency (SCC), an effective test-time defense. SCC consists of two complementary modules: Semantic consistency, which leverages soft pseudo-labels from counterattack warm-up and multi-view predictions to regularize cross-modal alignment and separate the target embedding from confusable negatives; and Spatial consistency, aligning perturbed visual predictions via augmented views to stabilize inference under adversarial perturbations. Together, these modules form a plug-and-play inference strategy. Extensive experiments on 22 benchmarks under diverse attack settings show that SCC consistently improves the zero-shot robustness of CLIP while maintaining accuracy, and can be seamlessly integrated with other VLMs for further gains. These findings highlight the great potential of establishing an adversarially robust paradigm from CLIP, with implications extending to broader vision-language domains such as BioMedCLIP.

cs.CV↗

arXiv2025

PaperAsk: A Benchmark for Reliability Evaluation of LLMs in Paper Search and Reading

Large Language Models (LLMs) increasingly serve as research assistants, yet their reliability in scholarly tasks remains under-evaluated. In this work, we introduce PaperAsk, a benchmark that systematically evaluates LLMs across four key research tasks: citation retrieval, content extraction, paper discovery, and claim verification. We evaluate GPT-4o, GPT-5, and Gemini-2.5-Flash under realistic usage conditions-via web interfaces where search operations are opaque to the user. Through controlled experiments, we find consistent reliability failures: citation retrieval fails in 48-98% of multi-reference queries, section-specific content extraction fails in 72-91% of cases, and topical paper discovery yields F1 scores below 0.32, missing over 60% of relevant literature. Further human analysis attributes these failures to the uncontrolled expansion of retrieved context and the tendency of LLMs to prioritize semantically relevant text over task instructions. Across basic tasks, the LLMs display distinct failure behaviors: ChatGPT often withholds responses rather than risk errors, whereas Gemini produces fluent but fabricated answers. To address these issues, we develop lightweight reliability classifiers trained on PaperAsk data to identify unreliable outputs. PaperAsk provides a reproducible and diagnostic framework for advancing the reliability evaluation of LLM-based scholarly assistance systems.

cs.IR↗