arXiv subjects

Jiyan He

Publications and source records attributed to Jiyan He.

At least 19 recordsLinked to original sources

arXiv2026

OrchBench: Evaluating Multi-Agent Orchestration Plans in Isolation via Deterministic Simulation

Complex tasks often decompose into parallelizable yet interdependent subtasks, making orchestration critical to the performance of multi-agent systems (MAS). Existing evaluations typically rely on end-to-end execution, which conflates orchestration-plan quality with worker capabilities, tool reliability, and environmental noise. Moreover, the time and token costs of real execution grow rapidly with workflow scale, making systematic evaluation expensive. We present OrchBench, a simulation-based benchmark for evaluating multi-agent orchestration plans in isolation. Starting from real-world tasks, OrchBench constructs directed acyclic graphs (DAGs) that encode task dependencies, with controlled sizes and degrees of parallelism. Given a DAG, a per-agent context limit, and an agent budget, the evaluated planner assigns subtasks to agents and specifies cross-agent information transfers and their retention ratios. A deterministic simulator evaluates the resulting plan without invoking worker agents and returns interpretable measures of result quality, makespan, and token cost. The simulated scores produced by OrchBench correlate strongly with quality scores from Claude Code executions, achieving a Pearson correlation of \(r=0.816\), while requiring only \(1.3\%\) of the tokens and \(10.3\%\) of the wall-clock time. Across diverse planners and workflow scales, we find that preserving task-critical information is more important than simply increasing the number of agents, and the benefits of parallelism diminish as coordination failures accumulate. These results establish OrchBench as an efficient and interpretable benchmark for comparing and diagnosing multi-agent orchestration plans.

cs.AI↗

arXiv2026

Learning behavior accounts for background-related advantage in AI-assisted education

Generative AI has been found, and will likely be found increasingly, useful in education. However, existing AI-for-education studies provide inconsistent evidence on its average effects. More broadly, research on prior educational technologies shows that average effects often mask substantial heterogeneity across student populations. Motivated by this evidence, this study examines heterogeneity in students' learning behavior with AI, which students benefit from AI assistance, and how learner profiles and learning behavior shape these patterns. To this end, we recruited 318 university students to participate in structured learning experiments lasting up to 125 minutes. Our findings indicate that students' learning behavior is strongly associated with learning outcomes, with behaviors characterized by proactive and critical engagement, rather than limited engagement, associated with significantly better performance. These behavioral differences are related to learner profiles, with students from higher-ranking universities and those with greater prior knowledge tending to benefit more, consistent with their greater likelihood of adopting proactive interaction strategies. Accounting for learning behavior substantially weakens or eliminates the associations between learner profiles and learning outcomes, suggesting that how students use AI is a key pathway through which background differences are linked to learning gains. Overall, this work provides a deeper understanding of AI assistance in education by showing how differences in learner profiles and learning behavior shape who benefits from AI-supported learning. These insights can help educators and students better navigate and integrate AI into educational practices.

cs.HC↗

arXiv2026

DEEPRUBRIC: Evidence-Tree Rubric Supervision for Efficient Reinforcement Learning of Deep Research Agents

Deep research agents synthesize long-form reports by searching and reasoning over retrieved evidence. Reinforcement learning with rubric-based rewards improves these agents by optimizing them against checkable criteria that translate report quality into reward signals, but its efficiency depends on whether those criteria reliably capture the task scope and evidence needs. Most existing studies ask an LLM to generate rubrics for a given query, but when the model fails to infer the underlying information needs, the generated rubrics may be incomplete and reduce RL efficiency. To obtain more reliable query--rubric supervision, we introduce DeepRubric, a data construction framework that reverses this process: instead of inferring evaluation criteria for a given query, it first determines what an evidence-backed report should be evaluated on and then synthesizes aligned query--rubric pairs from those evaluation targets. Starting from a sampled seed topic, DeepRubric builds an evidence tree by recursively expanding evidence-backed sub-questions, whose leaves serve as atomic and verifiable evaluation targets. It then uses the evidence tree to synthesize the training query and rubrics, ensuring that the reward evaluates exactly the information requested by the query. Using DeepRubric, we construct 9K query--rubric supervision examples and train DeepRubric-8B with rubric-based GRPO, achieving comparable performance to prior open state-of-the-art deep research models across three benchmarks with roughly 13x fewer RL GPU-hours.

cs.CL↗

arXiv2026

FutureWorld: A Live Reinforcement Learning Environment for Predictive Agents with Real-World Outcome Rewards

Live future prediction refers to the task of making predictions about real-world events before they unfold. This task is increasingly studied using large language model-based agent systems, and it is important for building agents that can continually learn from the real world. It can provide a large number of prediction questions grounded in diverse real-world events, while preventing answer leakage. To leverage the advantages of future prediction, we present FutureWorld, a live agentic reinforcement learning environment that closes the training loop between prediction, outcome realization, and parameter updates. Specifically, we modify and extend verl-tool, resulting in a new framework that we call verl-tool-future. Unlike standard reinforcement learning training frameworks that rely on immediate rewards, verl-tool-future stores prediction-time rollouts, backfills rewards after real-world outcomes become available, and then replays the completed trajectories for policy update. Across three open-source agents, successive FutureWorld training rounds lead to consistent improvements in prediction accuracy, probabilistic scoring, and calibration, demonstrating that delayed real-world outcome feedback can serve as an effective reinforcement learning signal.

cs.AI↗

arXiv2026

Harnessing Pre-Resolution Signals for Future Prediction Agents

Many high-stakes decisions depend on forecasts made before outcomes are known. In this future prediction setting, the central challenge is that public evidence evolves over time, while the main supervision signal arrives only after resolution: the realized outcome mainly assesses final correctness, offering only coarse guidance on what to track, what to verify, and which judgments to leave uncertain along the way. Our key observation is that revisiting the same unresolved question over time creates informative temporal contrasts across evolving evidence and repeated forecasts, exposing what earlier attempts missed before resolution and yielding a diagnostic signal we call the pre-resolution signal. We instantiate this idea in Milkyway, a future prediction agent with a persistent future prediction harness, an editable external state that stores reusable procedural guidance across revisits to the same unresolved question. As the same unresolved question is revisited, Milkyway extracts pre-resolution signals from evolving evidence and repeated forecasts, uses them to update the harness, and improves later forecasts on that question before resolution. After resolution, the realized outcome serves as a post-resolution check of provisional updates. On the FutureX and FutureWorld benchmarks, Milkyway achieves strong performance against competitive baselines, and a mechanism study suggests that the gains stem from harness evolution driven by pre-resolution signals rather than repeated prediction alone.

cs.AI↗

arXiv2026

Can a Lightweight Automated AI Pipeline Solve Research-Level Mathematical Problems?

Large language models (LLMs) have recently achieved remarkable success in generating rigorous mathematical proofs, with "AI for Math" emerging as a vibrant field of research (Ju et al., 2026). While these models have mastered competition-level benchmarks like the International Mathematical Olympiad (Huang et al., 2025; Duan et al., 2025) and show promise in research applications through auto-formalization (Wang et al., 2025), their deployment via lightweight, natural-language pipelines for research problems remains underexplored. In this work, we demonstrate that next-generation models (e.g., Gemini 3 Pro, GPT-5.2 Pro), when integrated into a streamlined automated pipeline optimized for citation-based verification, can solve sophisticated research-grade problems. We evaluate our pipeline on two novel datasets: (1) the ICCM (2025) problem sets (comparable to the S.-T. Yau College Student Mathematics Contest) proposed by leading mathematicians (Shanghai Math Challenge, 2026), and (2) the "First Proof" problem set (Abouzaid et al., 2026), consisting of previously unpublished research questions. Our pipeline generated candidate proofs for all problems in the first two ICCM sets and the "First Proof" set. The solutions for the first two ICCM sets and Problem 4 of the "First Proof" set have been fully verified by our team. All generated proofs have been submitted to the official organization, and our generated results are publicly available at https://github.com/ml1301215/question_sets-test_results. We have open-sourced the code and developed a user-friendly UI for this workflow, accessible at https://github.com/ml1301215/research-math-assistant.

cs.AI↗

arXiv2026

GUIGuard-Bench: Toward a General Evaluation for Privacy-Preserving GUI Agents

As GUI agents increasingly rely on screenshots to perceive and operate digital environments, they may inadvertently expose sensitive information such as identities, accounts, locations, and behavioral traces. While existing benchmarks primarily focus on task completion, grounding, or defenses against third-party attacks, current visual privacy datasets remain largely restricted to static natural images, limiting their ability to capture the contextual dependence and task relevance of privacy risks in GUI task trajectories. To bridge this gap, we introduce \textbf{GUIGuard-Bench}, a first-step benchmark for studying privacy-preserving GUI agents in trajectory-based GUI workflows. GUIGuard-Bench contains 241 real GUI-agent trajectories with 4,080 screenshots across Android and PC environments. Each screenshot is annotated at the region level with privacy bounding boxes, semantic privacy categories, risk levels, and whether the private information is necessary for completing the task. Built on these annotations, GUIGuard-Bench supports three complementary evaluations: privacy recognition, offline planning fidelity under protected screenshots, and the utility impact of different protection strategies. Our results show that current models can often detect whether a screenshot contains private information, but they struggle with fine-grained localization, category recognition, risk assessment, and task-necessity judgment. We also find that closed-source models, exemplified by Claude Sonnet 4.6, can maintain largely consistent planner semantics in Android environments after privacy protection is applied. Our results highlight privacy recognition as a critical bottleneck for practical GUI agents. Project: https://futuresis.github.io/GUIGuard-page/

cs.CR↗

arXiv2025

One Tool Is Enough: Reinforcement Learning for Repository-Level LLM Agents

Locating files and functions requiring modification in large software repositories is challenging due to their scale and structural complexity. Existing LLM-based methods typically treat this as a repository-level retrieval task and rely on multiple auxiliary tools, which often overlook code execution logic and complicate model control. We propose RepoNavigator, an LLM agent equipped with a single execution-aware tool: jumping to the definition of an invoked symbol. This unified design reflects the actual flow of code execution while simplifying tool manipulation. RepoNavigator is trained end-to-end via Reinforcement Learning (RL) directly from a base pretrained model, without relying on closed-source distillation. Experiments demonstrate that RL-trained RepoNavigator achieves state-of-the-art performance, with the 7B model outperforming 14B baselines, the 14B model surpassing 32B competitors, and the 32B model exceeding closed-source models such as GPT-5 on most metrics. These results confirm that integrating a single, structurally grounded tool with RL training provides an efficient and scalable solution for repository-level issue localization.

cs.SE↗

arXiv2025

Population-Evolve: a Parallel Sampling and Evolutionary Method for LLM Math Reasoning

Test-time scaling has emerged as a promising direction for enhancing the reasoning capabilities of Large Language Models in last few years. In this work, we propose Population-Evolve, a training-free method inspired by Genetic Algorithms to optimize LLM reasoning. Our approach maintains a dynamic population of candidate solutions for each problem via parallel reasoning. By incorporating an evolve prompt, the LLM self-evolves its population in all iterations. Upon convergence, the final answer is derived via majority voting. Furthermore, we establish a unification framework that interprets existing test-time scaling strategies through the lens of genetic algorithms. Empirical results demonstrate that Population-Evolve achieves superior accuracy with low performance variance and computational efficiency. Our findings highlight the potential of evolutionary strategies to unlock the reasoning power of LLMs during inference.

cs.AI↗

arXiv2025

PhysBrain: Human Egocentric Data as a Bridge from Vision Language Models to Physical Intelligence

Robotic generalization relies on physical intelligence: the ability to reason about state changes, contact-rich interactions, and long-horizon planning under egocentric perception and action. Vision Language Models (VLMs) are essential to Vision-Language-Action (VLA) systems, but the reliance on third-person training data creates a viewpoint gap for humanoid robots. Collecting massive robot-centric data is an ideal but impractical solution due to cost and diversity constraints. Conversely, human egocentric videos offer a highly scalable data source with rich interaction context, yet the embodiment mismatch prevents the direct application. To bridge this gap, we propose an Egocentric2Embodiment Translation Pipeline that transforms raw human egocentric videos into multi-level, schema-driven embodiment supervision with enforced evidence grounding and temporal consistency, enabling the construction of the Egocentric2Embodiment dataset (E2E-3M) at scale. An egocentric-aware embodied brain, termed PhysBrain, is obtained by training on the E2E-3M dataset. PhysBrain exhibits substantially improved egocentric understanding, particularly for planning. It provides an egocentric-aware initialization that enables more sample-efficient VLA fine-tuning and higher success rates, demonstrating effective transfer from human egocentric supervision to downstream robot control.

cs.RO↗

arXiv2025

GTM: Simulating the World of Tools for AI Agents

The integration of external tools is pivotal for empowering Large Language Model (LLM) agents with real-world capabilities. However, training these agents through direct, continuous interaction with diverse tools is often prohibitively expensive, slow, and introduces additional development and maintenance overhead. To address this challenge, we introduce the Generalist Tool Model (GTM), a 1.5-billion-parameter model that learns to act as a universal tool simulator. With only prompt-level configuration, GTM accesses tool functionalities along with input arguments and generates outputs that faithfully mimic real tool execution, providing a fast and cost-effective solution that eliminates development overhead. To build GTM, we propose the Context-Aware Response Generation (CARG) pipeline, which synthesizes comprehensive training data covering over 20,000 tools across 300 domains including physics, medicine, robotics, and finance. Through this pipeline, GTM learns to produce not only syntactically correct outputs but also logically coherent and contextually appropriate responses. Experiments demonstrate that GTM produces high-quality outputs with strong consistency and reliability. Besides when used in real reinforcement learning scenarios for agent training, GTM exhibits significantly faster simulation speed compared to real tools while maintaining comparable output quality, along with remarkable generalization and domain adaptability. Our results establish GTM as a foundational component for developing future AI agents, enabling efficient and scalable training of tool-augmented systems.

cs.AI↗

arXiv2025

Sparse Autoencoders Reveal Interpretable Structure in Small Gene Language Models

Sparse autoencoders (SAEs) have recently emerged as a powerful tool for interpreting the internal representations of large language models (LLMs), revealing latent latent features with semantical meaning. This interpretability has also proven valuable in biological domains: applying SAEs to protein language models uncovered meaningful features related to protein structure and function. More recently, SAEs have been used to analyze genomics-focused models such as Evo 2, identifying interpretable features in gene sequences. However, it remains unclear whether SAEs can extract meaningful representations from small gene language models, which have fewer parameters and potentially less expressive capacity. To address it, we propose applying SAEs to the activations of a small gene language model. We demonstrate that even small-scale models encode biologically relevant genomic features, such as transcription factor binding motifs, that SAEs can effectively uncover. Our findings suggest that compact gene language models are capable of learning structured genomic representations, and that SAEs offer a scalable approach for interpreting gene models across various model sizes.

q-bio.OT↗

arXiv2025

No Free Lunch: Rethinking Internal Feedback for LLM Reasoning

Reinforcement learning has emerged as a powerful paradigm for post-training large language models (LLMs) to improve reasoning. Approaches like Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) have shown strong results, but they require extensive external supervision. We investigate an alternative class of methods, Reinforcement Learning from Internal Feedback (RLIF), which relies solely on intrinsic model-derived signals instead of external rewards. In particular, we leverage unsupervised reward proxies such as token-level entropy, trajectory-level entropy, and self-certainty. Our theoretical analysis shows these internal objectives are partially equivalent, and we empirically evaluate various RLIF strategies on challenging math reasoning benchmarks. Experimental results demonstrate that RLIF can boost the reasoning performance of base LLMs at the beginning phase of the training, matching or surpassing RLVR techniques on these tasks. However, when training progresses, performance degrades even below the model before training. Moreover, we find that RLIF yields little improvement for instruction-tuned models, indicating diminishing returns of intrinsic feedback once an LLM is already instruction-tuned. We further analyze this limitation by mixing model weights and explain the reason of RLIF's training behaviors, providing practical guidelines for integrating internal feedback signals into LLM training. We hope our analysis of internal feedback will inform more principled and effective strategies for LLM post-training.

cs.LG↗

arXiv2025

Modeling Earth-Scale Human-Like Societies with One Billion Agents

Understanding the dynamic evolution of complex social phenomena requires both high-fidelity modeling of human behavior and large-scale simulations. Traditional agent-based models (ABMs) have been employed to study these dynamics, but are constrained by simplified agent behaviors. Recent advances in large language models (LLMs) enable agents to exhibit sophisticated social behaviors, yet face significant scaling challenges. We present Light Society, an agent-based simulation framework that advances both fronts. Light Society formalizes social processes as structured transitions of agent and environment states, governed by a set of LLM-powered simulation operations. Joint algorithmic and system optimizations, particularly a mixture-of-models engine that combines full LLMs with distilled surrogates, enable Light Society to efficiently simulate societies with over one billion agents. Grounded in real-world demographic profiles from the World Values Survey, simulations of Trust Games and opinion diffusion at up to one billion agents demonstrate Light Society's high fidelity and efficiency in modeling diverse social phenomena, providing researchers with a practical foundation for hypothesis testing and the study of emergent collective behaviors at planetary scale.

cs.MA↗

arXiv2025

RedChronos: A Large Language Model-Based Log Analysis System for Insider Threat Detection in Enterprises

Internal threat detection (IDT) aims to address security threats within organizations or enterprises by identifying potential or already occurring malicious threats within vast amounts of logs. Although organizations or enterprises have dedicated personnel responsible for reviewing these logs, it is impossible to manually examine all logs entirely.In response to the vast number of logs, we propose a system called RedChronos, which is a Large Language Model-Based Log Analysis System. This system incorporates innovative improvements over previous research by employing Query-Aware Weighted Voting and a Semantic Expansion-based Genetic Algorithm with LLM-driven Mutations. On the public datasets CERT 4.2 and 5.2, RedChronos outperforms or matches existing approaches in terms of accuracy, precision, and detection rate. Moreover, RedChronos reduces the need for manual intervention in security log reviews by approximately 90% in the Xiaohongshu Security Operation Center. Therefore, our RedChronos system demonstrates exceptional performance in handling IDT tasks, providing innovative solutions for these challenges. We believe that future research can continue to enhance the system's performance in IDT tasks while also reducing the response time to internal risk events.

cs.CR↗

arXiv2024

Physical Consistency Bridges Heterogeneous Data in Molecular Multi-Task Learning

In recent years, machine learning has demonstrated impressive capability in handling molecular science tasks. To support various molecular properties at scale, machine learning models are trained in the multi-task learning paradigm. Nevertheless, data of different molecular properties are often not aligned: some quantities, e.g. equilibrium structure, demand more cost to compute than others, e.g. energy, so their data are often generated by cheaper computational methods at the cost of lower accuracy, which cannot be directly overcome through multi-task learning. Moreover, it is not straightforward to leverage abundant data of other tasks to benefit a particular task. To handle such data heterogeneity challenges, we exploit the specialty of molecular tasks that there are physical laws connecting them, and design consistency training approaches that allow different tasks to exchange information directly so as to improve one another. Particularly, we demonstrate that the more accurate energy data can improve the accuracy of structure prediction. We also find that consistency training can directly leverage force and off-equilibrium structure data to improve structure prediction, demonstrating a broad capability for integrating heterogeneous data.

cs.LG↗

arXiv2024

AquaLoRA: Toward White-box Protection for Customized Stable Diffusion Models via Watermark LoRA

Diffusion models have achieved remarkable success in generating high-quality images. Recently, the open-source models represented by Stable Diffusion (SD) are thriving and are accessible for customization, giving rise to a vibrant community of creators and enthusiasts. However, the widespread availability of customized SD models has led to copyright concerns, like unauthorized model distribution and unconsented commercial use. To address it, recent works aim to let SD models output watermarked content for post-hoc forensics. Unfortunately, none of them can achieve the challenging white-box protection, wherein the malicious user can easily remove or replace the watermarking module to fail the subsequent verification. For this, we propose \texttt{\method} as the first implementation under this scenario. Briefly, we merge watermark information into the U-Net of Stable Diffusion Models via a watermark Low-Rank Adaptation (LoRA) module in a two-stage manner. For watermark LoRA module, we devise a scaling matrix to achieve flexible message updates without retraining. To guarantee fidelity, we design Prior Preserving Fine-Tuning (PPFT) to ensure watermark learning with minimal impacts on model distribution, validated by proofs. Finally, we conduct extensive experiments and ablation studies to verify our design.

cs.CR↗

arXiv2024

Towards Generalist Prompting for Large Language Models by Mental Models

Large language models (LLMs) have demonstrated impressive performance on many tasks. However, to achieve optimal performance, specially designed prompting methods are still needed. These methods either rely on task-specific few-shot examples that require a certain level of domain knowledge, or are designed to be simple but only perform well on a few types of tasks. In this work, we attempt to introduce the concept of generalist prompting, which operates on the design principle of achieving optimal or near-optimal performance on a wide range of tasks while eliminating the need for manual selection and customization of prompts tailored to specific problems. Furthermore, we propose MeMo (Mental Models), an innovative prompting method that is simple-designed yet effectively fulfills the criteria of generalist prompting. MeMo distills the cores of various prompting methods into individual mental models and allows LLMs to autonomously select the most suitable mental models for the problem, achieving or being near to the state-of-the-art results on diverse tasks such as STEM, logical reasoning, and commonsense reasoning in zero-shot settings. We hope that the insights presented herein will stimulate further exploration of generalist prompting methods for LLMs.

cs.CL↗