Searcharxiv⌕ Search

arXiv subjects

Boyi Liu

Publications and source records attributed to Boyi Liu.

At least 19 recordsLinked to original sources
arXiv2026

Benchmarking Hybrid Deep Research Across Database Querying and Web Search

While autonomous agents have made significant strides in "deep research" by iteratively navigating the open web to synthesize information, real-world problem-solving is rarely confined to a single environment. Complex analytical tasks inherently require agents to weave together evidence from both ambiguous unstructured text (e.g., the open web) and highly precise structured data (e.g., relational databases). However, existing benchmarks evaluate these modalities in isolation, failing to capture the critical "handoff" - the ability to preserve constraints when moving evidence between systems. We introduce HybridDeepResearch, to our knowledge the first deep-research benchmark that requires both web search and SQL to form a complete, verifiable answer. The benchmark contains 380 tool-dependent tasks grounded in LiveSQLBench-Base-Lite databases and public web corpora, validated through automated checks and human review, and covering three reasoning patterns: SQL2S, S2SQL, and Parallel. Evaluations across proprietary and open-weight models under various agentic scaffolds reveal that even state-of-the-art models like GLM-5.2, Claude-Sonnet-4.6 and GPT-5 achieve only about 50-54% Pass@8 on the hard subset. Notably, results show that directional reasoning is substantially more difficult than parallel intersection, highlighting that bridging structured and unstructured information spaces without losing constraints remains a major open challenge for agentic systems. Code and datasets are publicly available at GitHub (https://github.com/Snowflake-AI-Research/HybridDeepResearch) and Hugging Face (https://huggingface.co/datasets/Snowflake/HybridDeepResearch).

cs.CL↗
arXiv2026

ArcticSwarm: Deferring Early Consensus in Long-Horizon Multi-Agent Research

Multi-agent systems have shown strong performance in domains with reliable verifiers such as coding, where multi-parallel candidate generation selected by a verifier is effective. However, such pipelines would not generalize to open-ended, long-horizon research tasks without a verifier. While majority voting or self-consistency is often used to reach consensus as a proxy verifier, parallel agents repeatedly explore the same evidence, while access to peers' partial findings cause search to converge on an early candidate before alternatives are tested. We present ArcticSwarm, a multi-agent research architecture that separates evidence gathering from evidence integration. Subagents publish findings to a shared bulletin board, while gated isolation lets selected search tasks maintain their own prior, preventing early consensus. Structured review at three commitment boundaries enforce only confident candidates to be propagated. As a result, ArcticSwarm reaches 82.6% on the full BrowseComp-Plus set with the open-weight Qwen 3.5-27B model, compared with 78.8% without gated isolation and 74.5% additionally with structured review disabled, outperforming aligned baseline MiroFlow runs (70.6%). Extending to live-web BrowseComp, ArcticSwarm reaches 73.6% with GPT-5, which is well above the reported provider system (54.9%) and MiroFlow (63.4%). Overall, the results show that restricting peer reads during evidence gathering and strengthening commitment boundaries before a hypothesis is shared can broaden search and improve long-horizon multi-agent deep research.

cs.MA↗
arXiv2026

SkillAlchemy: Open-World Agent Skill Creation

Agent skills are reusable procedural artifacts that extend language agents with specialized workflows, tool conventions, and domain behaviors at inference time. However, creating reliable skills still depends largely on human authorship, model priors, or execution traces. These sources are often unavailable for unfamiliar tasks, suggesting the need to create skills from open-world materials. In this paper, we study open-world skill creation: given an underspecified skill brief and a source-access specification, a creator must discover behavior-relevant requirements omitted by the brief and determine how broadly each source-derived procedure is justified. We propose SkillAlchemy, an admission-centered framework for source-grounded skill creation. SkillAlchemy identifies implicit requirements through contrastive evidence, admits candidate procedures based on evidence-supported scope, and compiles the admitted content into a grammar-guided skill package. Extensive experiments across 87 SkillsBench v1.1 tasks demonstrate that SkillAlchemy improves pass rate over no-skill execution by 19.9 percentage points and the strongest automated baseline by 8.6 percentage points, while achieving performance comparable to human-curated skills.

cs.AI↗
arXiv2026

MidTool: Mid-training Data Synthesis for Agentic Tool Use

Mid-training is increasingly recognized as a critical stage for shaping the capabilities of large language models. Recent work has shown that targeted mid-training can strengthen reasoning-intensive abilities such as math and science, and can also improve agentic capabilities in software-engineering settings. In this work, we study the parallel but less explored agentic capability: general tool use. We present MidTool, an open corpus construction pipeline for agentic tool-use mid-training that combines large-scale web, PDF, and code data with synthesized supervision from real-world tool APIs, MCP skills, and document-grounded workflows. MidTool is designed to teach models how to recognize tool affordances, ground arguments from context, compose tool call workflow, and recover from incomplete information. We mid-train Qwen3-4B-Base and Qwen3-8B-Base on MidTool-Mix, and then apply follow-up post-training with both supervised fine-tuning and reinforcement learning. Compared with baselines, MidTool-Mix consistently improves downstream performance under both SFT and RL on BFCL, tau2-Bench, and MCP Universe. These results suggest that general tool use, like other important LLM capabilities, benefits from dedicated mid-training rather than being left entirely to post-training.

cs.AI↗
arXiv2026

SkillShapley: Boundary-Adaptive Shapley Valuation for Skill Step Attribution in LLM Agents

Agent skills are crucial external instructions that enable language agents to execute long procedural tasks such as coding or document processing. Existing agent skills are primarily created through human manual crafting or agent execution traces, with limited understanding of how each step contributes to overall skill performance on specific tasks; i.e., there remains an open problem in quantifying the contribution of individual steps within an agent skill. To address this issue, we first model skill-step attribution as a Shapley value-based contribution estimation problem, and then propose SkillShapley, a step-level attribution framework for agent skills. Notably, SkillShapley operates in two phases, motivated by key empirical insights, i.e., discretized benchmark rewards that create sharp performance cliffs, and step interactions that are largely additive rather than synergistic. Specifically, it first identifies informative coalitional regions, and then adaptively samples new coalitions that can yield reusable marginal evidence. Experiments on skills from the widely adopted SkillsBench demonstrate that our SkillShapley can effectively and efficiently identify high- or low-value skill steps, providing several key takeaways for agent skill creation.

cs.AI↗
arXiv2026

The Optimizer Is the Agent: Reasoning-Driven Search across Prompts, Programs, and ML Workflows

Recent systems for optimizing prompts, programs, and ML workflows typically rely on explicit outer-loop controllers such as evolutionary search, bandits, or textual-gradient methods. We ask a fundamentally different question: how much of this search policy can be internalized by a single tool-using agent? We present ReASearch, a unified framework for reasoning-driven optimization in which the agent autonomously decides what to evaluate, how to diagnose failures, which edits to make, and when to verify or restart. Rather than serving only as a proposal generator guided by hand-designed heuristics, the agent actively analyzes outcomes, allocates budget, and refines its strategy over long horizons through persistent memory. With a shared agent loop and domain-specific tools, ReASearch instantiates the exact same scaffold to optimize prompts, programs, and ML workflows. Across 14 diverse tasks, it is competitive with and mostly better than specialized optimization systems, achieving gains of 2% to 40% over strong domain-specific baselines, and in some cases discovering solutions that improve on prior human best-known results. Crucially, we observe that complex search behaviors, which are typically implemented by explicit controllers, emerge naturally from the agent's reasoning process.

cs.AI↗
arXiv2026

LooperMuscle: Fast and Stable Learning of Humanoid Whole-Body Tracking via Structured Mixture-of-Experts

FastSAC-style methods significantly reduce humanoid motion training time but often suffer from notable performance degradation compared with PPO in whole-body tracking tasks. We target this speed-performance gap by introducing LooperMuscle, a composed expert policy learning framework that restores tracking quality while preserving high training efficiency. LooperMuscle combines a semantically structured mixture-of-experts actor, an expert-aware distributional critic, and contribution-routed replay with deferred curriculum scheduling. These three components form a closed training loop in which expert contributions guide data routing, routed data shape value learning, and value gradients in turn refine expert specialization. Empirically, our approach substantially outperforms vanilla FastSAC in motion tracking accuracy while requiring far less wall-clock time than PPO: where FastSAC trains in about 15 minutes but underperforms, and PPO achieves stronger results but requires about 6 hours, LooperMuscle recovers a substantial fraction of the remaining gap to PPO in roughly 45 minutes of simulation training, delivering practical efficiency for rapid policy iteration. The code will be released to benefit the research community at https://loopermuscle.github.io/.

cs.RO↗
arXiv2026

Organize then Retrieve: Hierarchical Memory Navigation for Efficient Agents

Large language model (LLM) agents struggle with long-horizon tasks due to their inherent statelessness, requiring all task-relevant information to be encoded in growing input contexts. The resulting degraded reasoning quality, increased inference cost, and higher latency necessitate efficient working memory mechanisms. However, existing approaches either rely on lossy compression or similarity-based retrieval, which often fail to capture temporal structure and causal dependencies required for multi-step agentic tasks. In this work, we present HORMA, a Hierarchical Organize-and-Retrieve Memory Agent that organizes experience into a file-system-like hierarchical structure, where summarized entities are linked to the corresponding raw trajectories, enabling efficient access without losing detailed information. HORMA decomposes working memory into two stages: structured memory construction and navigation-based retrieval. The construction module iteratively refines how experiences are structured by distinguishing between failures caused by missing information and those caused by misleading or overloaded context. The navigation module retrieves task-relevant context by traversing the hierarchy using a lightweight agent trained with reinforcement learning to select minimal yet sufficient context, thereby reducing latency along the critical execution path. Across ALFWorld, LoCoMo, and LongMemEval, HORMA improves task performance under constrained context budgets while requiring at most 22.17% of the baseline token usage in long conversation tasks. Compared to existing methods, it consistently achieves better efficiency-performance trade-offs and generalizes effectively to unseen tasks.

cs.AI↗
arXiv2026

Learning to Self-Evolve

We introduce Learning to Self-Evolve (LSE), a reinforcement learning framework that trains large language models (LLMs) to improve their own contexts at test time. We situate LSE in the setting of test-time self-evolution, where a model iteratively refines its context from feedback on seen problems to perform better on new ones. Existing approaches rely entirely on the inherent reasoning ability of the model and never explicitly train it for this task. LSE reduces the multi-step evolution problem to a single-step RL objective, where each context edit is rewarded by the improvement in downstream performance. We pair this objective with a tree-guided evolution loop. On Text-to-SQL generation (BIRD) and general question answering (MMLU-Redux), a 4B-parameter model trained with LSE outperforms self-evolving policies powered by GPT-5 and Claude Sonnet 4.5, as well as prompt optimization methods including GEPA and TextGrad, and transfers to guide other models without additional training. Our results highlight the effectiveness of treating self-evolution as a learnable skill.

cs.CL↗
arXiv2026

DARE-bench: Evaluating Modeling and Instruction Fidelity of LLMs in Data Science

The fast-growing demands in using Large Language Models (LLMs) to tackle complex multi-step data science tasks create an emergent need for accurate benchmarking. There are two major gaps in existing benchmarks: (i) the lack of standardized, process-aware evaluation that captures instruction adherence and process fidelity, and (ii) the scarcity of accurately labeled training data. To bridge these gaps, we introduce DARE-bench, a benchmark designed for machine learning modeling and data science instruction following. Unlike many existing benchmarks that rely on human- or model-based judges, all tasks in DARE-bench have verifiable ground truth, ensuring objective and reproducible evaluation. To cover a broad range of tasks and support agentic tools, DARE-bench consists of 6,300 Kaggle-derived tasks and provides both large-scale training data and evaluation sets. Extensive evaluations show that even highly capable models such as gpt-o4-mini struggle to achieve good performance, especially in machine learning modeling tasks. Using DARE-bench training tasks for fine-tuning can substantially improve model performance. For example, supervised fine-tuning boosts Qwen3-32B's accuracy by 1.83x and reinforcement learning boosts Qwen3-4B's accuracy by more than 8x. These significant improvements verify the importance of DARE-bench both as an accurate evaluation benchmark and critical training data.

cs.AI↗
arXiv2026

Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning

Recent advances in large language model (LLM) have empowered autonomous agents to perform multi-turn interactions with tools and environments. However, scaling such agent training is limited by the lack of diverse and reliable environments. In this paper, we propose Agent World Model (AWM), a fully synthetic environment generation pipeline. Using this pipeline, we scale to 1,000 environments covering everyday scenarios, in which agents can interact with rich toolsets and obtain high-quality observations. Notably, these environments are code-driven and backed by databases, providing more reliable and consistent state transitions than environments simulated by LLMs. Moreover, they enable more efficient agent interaction compared with collecting trajectories from realistic environments. To demonstrate the effectiveness of this resource, we perform large-scale reinforcement learning for multi-turn tool-use agents. Thanks to the fully executable environments and accessible database states, we can also design reliable reward functions. Experiments on three benchmarks show that training exclusively in synthetic environments, rather than benchmark-specific ones, yields strong out-of-distribution generalization. The code is available at https://github.com/Snowflake-Labs/agent-world-model.

cs.AI↗
arXiv2026

FedMosaic: Federated Retrieval-Augmented Generation via Parametric Adapters

Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by grounding generation in external knowledge to improve factuality and reduce hallucinations. Yet most deployments assume a centralized corpus, which is infeasible in privacy aware domains where knowledge remains siloed. This motivates federated RAG (FedRAG), where a central LLM server collaborates with distributed silos without sharing raw documents. In context RAG violates this requirement by transmitting verbatim documents, whereas parametric RAG encodes documents into lightweight adapters that merge with a frozen LLM at inference, avoiding raw-text exchange. We adopt the parametric approach but face two unique challenges induced by FedRAG: high storage and communication from per-document adapters, and destructive aggregation caused by indiscriminately merging multiple adapters. We present FedMosaic, the first federated RAG framework built on parametric adapters. FedMosaic clusters semantically related documents into multi-document adapters with document-specific masks to reduce overhead while preserving specificity, and performs selective adapter aggregation to combine only relevance-aligned, nonconflicting adapters. Experiments show that FedMosaic achieves an average 10.9% higher accuracy than state-of-the-art methods in four categories, while lowering storage costs by 78.8% to 86.3% and communication costs by 91.4%, and never sharing raw documents.

cs.CL↗
arXiv2026

OpenNavMap: Multi-Session Appearance-Based Topometric Mapping for Scalable Visual Navigation

Scalable and maintainable maps are fundamental to large-scale navigation and the long-term deployment of robots in real-world environments. However, conventional maps that explicitly maintain dense geometry or 3D landmarks incur high storage and maintenance costs, while the core challenge of scaling to multi-session mapping is visual localization under sparse viewpoint overlap, temporal appearance shifts, and cross-device variance. To address this, we propose OpenNavMap, a lightweight, landmark-free topometric mapping system that organizes image nodes into covisibility, odometry, and traversability graphs and delegates local geometry recovery to 3D geometric foundation models (GFMs) on demand. For localization, dynamic-programming-based sequence matching narrows candidate correspondences for a GFM, reducing global estimation to a lightweight, pose-only optimization; for mapping, a lifelong pipeline fuses multi-session, multi-device data via cross-device merging and node culling. On a 19km dataset across four real-world environments, \methodname attains a state-of-the-art $0.62$m translation error on the Map-Free benchmark, bounds the absolute trajectory error below $3$m across 15.7km without depth sensors, and completes $12$ autonomous image-goal visual navigation tasks on both simulated and physical robots. Code and datasets will be made publicly available at https://rpl-cs-ucl.github.io/OpenNavMap_page.

cs.RO↗
arXiv2026

Federated Personalization of Early-Exit Networks

Personalized Federated Learning (PFL) excels at tailoring client-specific models, which is particularly critical for decentralized and heterogeneous data environments, yet existing methods produce static models with a fixed tradeoff between accuracy and efficiency. This inherent static nature limits their ability to adapt to inference demands that vary with context and resource availability, posing a challenge for real-world deployment. Early-exit networks (EENs), which enable adaptive inference via intermediate classifiers, offer a promising solution. However, integrating EENs into PFL introduces two intertwined conflicts: client-wise heterogeneity across clients and depth-wise interference arising from conflicting exit objectives. Prior studies fail to resolve both conflicts simultaneously, leading to suboptimal performance. In this paper, we propose X-FED, a novel Conflict-Aware Cross-Client Federated Exit Distillation framework that jointly addresses both client- and depth-wise conflicts while extending PFL to early-exit networks. At its core, X-FED employs a progressive, depth-prioritized student coordination mechanism that mitigates interference among shallow and deep exits while enabling effective personalized knowledge transfer across clients. Furthermore, we introduce a client-decoupled formulation that reduces communication overhead with theoretical soundness. Extensive evaluations on various datasets show that compared to state-of-the-art PFL and PFL-EE methods, X-FED achieves higher accuracy while reducing inference costs by 30.79%-46.86%.

cs.LG↗
arXiv2025

The Starlink Robot: A Platform and Dataset for Mobile Satellite Communication

The integration of satellite communication into mobile devices represents a paradigm shift in connectivity, yet the performance characteristics under motion and environmental occlusion remain poorly understood. We present the Starlink Robot, the first mobile robotic platform equipped with Starlink satellite internet, comprehensive sensor suite including upward-facing camera, LiDAR, and IMU, designed to systematically study satellite communication performance during movement. Our multi-modal dataset captures synchronized communication metrics, motion dynamics, sky visibility, and 3D environmental context across diverse scenarios including steady-state motion, variable speeds, and different occlusion conditions. This platform and dataset enable researchers to develop motion-aware communication protocols, predict connectivity disruptions, and optimize satellite communication for emerging mobile applications from smartphones to autonomous vehicles. In this work, we use LEOViz for real-time satellite tracking and data collection. The starlink robot project is available at https://github.com/StarlinkRobot.

cs.RO↗
arXiv2025

WiLLM: an Open Framework for LLM Services over Wireless Systems

Large Language Model (LLM) services fundamentally differ from traditional Deep Neural Network (DNN) applications in wireless networks. We identify three critical distinctions: (1) unlike traditional DNNs with unidirectional data flows, LLM's multimodal interactions create bidirectional heavy loads with contrasting bottlenecks, requiring direction-aware resource scheduling; (2) while traditional DNNs exhibit fixed computational patterns, LLM's highly variable inference times interact complexly with network slicing, causing dynamic bottleneck migration; and (3) in contrast to predictable DNN traffic, LLM's token streams demonstrate unprecedented burstiness and state dependencies. These insights motivate WiLLM, the first open-source framework, implemented as a wireless platform, for LLM service research. Built on OpenAirInterface, WiLLM introduces several technical innovations: dynamic slice compatibility, universal UE compatibility through application-layer tunneling, multi-UE multi-slice scheduling, dual-mode resource allocation, and cross-layer APIs. In addition, WiLLM eliminates the need for specialized wireless expertise, enabling researchers and developers to experiment with LLM services over realistic cellular networks. We demonstrate the platform's capabilities through a smart glasses case study and provide a comprehensive dataset of \~1.6 million synchronized measurements. The complete system, dataset, and appendix are available at https://openwillm.github.io.

cs.NI↗
arXiv2025

MedOrch: Medical Diagnosis with Tool-Augmented Reasoning Agents for Flexible Extensibility

Healthcare decision-making represents one of the most challenging domains for Artificial Intelligence (AI), requiring the integration of diverse knowledge sources, complex reasoning, and various external analytical tools. Current AI systems often rely on either task-specific models, which offer limited adaptability, or general language models without grounding with specialized external knowledge and tools. We introduce MedOrch, a novel framework that orchestrates multiple specialized tools and reasoning agents to provide comprehensive medical decision support. MedOrch employs a modular, agent-based architecture that facilitates the flexible integration of domain-specific tools without altering the core system. Furthermore, it ensures transparent and traceable reasoning processes, enabling clinicians to meticulously verify each intermediate step underlying the system's recommendations. We evaluate MedOrch across three distinct medical applications: Alzheimer's disease diagnosis, chest X-ray interpretation, and medical visual question answering, using authentic clinical datasets. The results demonstrate MedOrch's competitive performance across these diverse medical tasks. Notably, in Alzheimer's disease diagnosis, MedOrch achieves an accuracy of 93.26%, surpassing the state-of-the-art baseline by over four percentage points. For predicting Alzheimer's disease progression, it attains a 50.35% accuracy, marking a significant improvement. In chest X-ray analysis, MedOrch exhibits superior performance with a Macro AUC of 61.2% and a Macro F1-score of 25.5%. Moreover, in complex multimodal visual question answering (Image+Table), MedOrch achieves an accuracy of 54.47%. These findings underscore MedOrch's potential to advance healthcare AI by enabling reasoning-driven tool utilization for multimodal medical data processing and supporting intricate cognitive tasks in clinical decision-making.

cs.CL↗
arXiv2025

Graph-Reward-SQL: Execution-Free Reinforcement Learning for Text-to-SQL via Graph Matching and Stepwise Reward

Reinforcement learning (RL) has been widely adopted to enhance the performance of large language models (LLMs) on Text-to-SQL tasks. However, existing methods often rely on execution-based or LLM-based Bradley-Terry reward models. The former suffers from high execution latency caused by repeated database calls, whereas the latter imposes substantial GPU memory overhead, both of which significantly hinder the efficiency and scalability of RL pipelines. To this end, we propose a novel reward model framework for RL-based Text-to-SQL named Graph-Reward-SQL, which employs the GMNScore outcome reward model. We leverage SQL graph representations to provide accurate reward signals while significantly reducing time cost and GPU memory usage. Building on this foundation, we further introduce StepRTM, a stepwise reward model that provides intermediate supervision over Common Table Expression (CTE) subqueries. This encourages both functional correctness and readability of SQL. Extensive comparative and ablation experiments on standard benchmarks, including Spider and BIRD, demonstrate that our method consistently outperforms existing reward models.

cs.LG↗