arXiv subjects

Wanli Li

Publications and source records attributed to Wanli Li.

13 recordsLinked to original sources

arXiv2026

AREX: Towards a Recursively Self-Improving Agent for Deep Research

Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constraint-wise checks. This discovery--verification asymmetry suggests that a research agent should do more than simply search longer: it should recursively improve its current answer by verifying intermediate results and using the partially verified state to guide subsequent refinement. We introduce AREX, a family of Recursively Self-Improving (RSI) deep research agents. AREX alternates between an inner research loop that gathers evidence and constructs a provisional answer, and an outer self-improvement loop that audits the answer constraint-wise, identifies unresolved claims, and launches targeted follow-up research. To sustain RSI over long horizons, AREX learns an autonomous context-update tool that compresses growing interaction history into a compact improvement state preserving verified evidence and unresolved constraints, without relying on an external model. We train AREX on verified synthetic tasks and high-quality trajectories through agentic mid-training and long-horizon reinforcement learning. To mitigate sparse final rewards during long horizon learning, we emphasize key steps where decisive evidence is acquired or erroneous research directions are corrected. We instantiate a dense 4B model and a 122B-A10B Mixture-of-Experts model. Across BrowseComp, WideSearch, DeepSearchQA, Humanity's Last Exam (HLE), and other reasoning and tool-use benchmarks, AREX substantially outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.

cs.AI↗

arXiv2026

WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces

Computer-use agents (CUAs) increasingly operate in runtimes that combine visual desktop control, command-line execution, code editing, browsers, and external tools. Existing benchmarks, however, often evaluate these interfaces as separable capabilities, leaving long-horizon cross-interface orchestration under-tested. Thus, we introduce WeaveBench, a long-horizon hybrid-interface benchmark with 114 tasks across 8 real-world work domains, grounded in real user requests and publicly verifiable artifacts. Each task requires agents to combine GUI observations/actions with CLI/code operations within a single trajectory. We evaluate these tasks on a real Ubuntu desktop inside deployed CLI-agent runtimes, augmented with a minimal desktop-control plugin. We also propose a companion trajectory-aware judge that inspects deliverables, files, screenshots, logs, and action traces, while detecting shortcut behaviors such as fabricated visual evidence or hard-coded metrics. Across frontier model-runtime pairings, the best PassRate reaches only 41.2%, showing the benchmark remains far from saturated. The trajectory-aware judge further reveals that outcome-only grading substantially overestimates agent performance. Overall, WeaveBench exposes a critical gap in CUA evaluation and provides an effective testbed to measure whether agents can orchestrate GUI, CLI, and code operations across long-horizon real-world tasks.

cs.AI↗

arXiv2026

LiteResearcher: A Scalable Agentic RL Training Framework for Deep Research Agent

Reinforcement Learning (RL) has emerged as a powerful training paradigm for LLM-based agents. However, scaling agentic RL for deep research remains constrained by two coupled challenges: hand-crafted synthetic data fails to elicit genuine real-world search capabilities, and real-world search dependency during RL training introduces instability and prohibitive cost, which limits the scalability of Agentic RL. LiteResearcher is a training framework that makes Agentic RL scalable: by constructing a lite virtual world that mirrors real-world search dynamics, we enable a continuously improving training recipe that empowers a tiny search agent to outperform large-scale open-source and commercial models (e.g., Tongyi DeepResearch and Claude-4.5 Sonnet). Specifically, on common benchmarks such as GAIA and Xbench, our LiteResearcher-4B achieves open-source state-of-the-art results of 71.3% and 78.0% respectively, demonstrating that scalable RL training is a key enabler for Deep Research Agents.

cs.AI↗

arXiv2026

ST-Lite: Training-Free KV Cache Compression with Spatio-Trajectory Guidance for Long-Horizon GUI Agents

Training-free KV cache compression is essential for deploying vision-language GUI agents under memory and latency constraints, yet existing methods are designed for generic language workloads and ignore the distinctive structure of GUI interaction traces. We characterize three GUI-specific workload properties--high inter-frame visual redundancy, extremely small UI-element spatial footprints, and near-uniform cross-layer attention sparsity--that cause existing schemes to retain as few as 39% of oracle-important KV pairs at the standard 20% budget. To address this, we propose ST-Lite, a training-free compression scheme whose three components each target one property: Trajectory-aware Semantic Gating (TSG) filters redundant historical frames, Component-centric Spatial Saliency (CSS) preserves fine-grained element boundaries, and a flat per-layer budget avoids hierarchical misallocation. Across seven GUI benchmarks and two backbones in the deployment-relevant 10%-40% window, ST-Lite consistently outperforms all existing compression baselines, matching or exceeding Full Cache task accuracy on the primary backbone at the 20% budget while delivering up to 2.35x decoding speedup at fivefold compression. The implementation is available at https://github.com/94wen94/ST-Lite.

cs.CV↗

arXiv2026

SAIL: Self-Amplified Iterative Learning for Diffusion Model Alignment with Minimal Human Feedback

Aligning diffusion models with human preferences remains challenging, particularly when reward models are unavailable or impractical to obtain, and collecting large-scale preference datasets is prohibitively expensive. \textit{This raises a fundamental question: can we achieve effective alignment using only minimal human feedback, without auxiliary reward models, by unlocking the latent capabilities within diffusion models themselves?} In this paper, we propose \textbf{SAIL} (\textbf{S}elf-\textbf{A}mplified \textbf{I}terative \textbf{L}earning), a novel framework that enables diffusion models to act as their own teachers through iterative self-improvement. Starting from a minimal seed set of human-annotated preference pairs, SAIL operates in a closed-loop manner where the model progressively generates diverse samples, self-annotates preferences based on its evolving understanding, and refines itself using this self-augmented dataset. To ensure robust learning and prevent catastrophic forgetting, we introduce a ranked preference mixup strategy that carefully balances exploration with adherence to initial human priors. Extensive experiments demonstrate that SAIL consistently outperforms state-of-the-art methods across multiple benchmarks while using merely 6\% of the preference data required by existing approaches, revealing that diffusion models possess remarkable self-improvement capabilities that, when properly harnessed, can effectively replace both large-scale human annotation and external reward models.

cs.CV↗

arXiv2026

MetaToolAgent: Towards Generalizable Tool Usage in LLMs through Meta-Learning

Tool learning is increasingly important for large language models (LLMs) to effectively coordinate and utilize a diverse set of tools in order to solve complex real-world tasks. By selecting and integrating appropriate tools, LLMs extend their capabilities beyond pure language understanding to perform specialized functions. However, existing methods for tool selection often focus on limited tool sets and struggle to generalize to novel tools encountered in practical deployments. To address these challenges, we introduce a comprehensive dataset spanning 7 domains, containing 155 tools and 9,377 question-answer pairs, which simulates realistic integration scenarios. Additionally, we propose MetaToolAgent (MTA), a meta-learning approach designed to improve cross-tool generalization. Experimental results show that MTA significantly outperforms baseline methods on unseen tools, demonstrating its promise for building flexible and scalable systems that require dynamic tool coordination.

cs.LG↗

arXiv2025

TempFlow-GRPO: When Timing Matters for GRPO in Flow Models

Recent flow matching models for text-to-image generation have achieved remarkable quality, yet their integration with reinforcement learning for human preference alignment remains suboptimal, hindering fine-grained reward-based optimization. We observe that the key impediment to effective GRPO training of flow models is the temporal uniformity assumption in existing approaches: sparse terminal rewards with uniform credit assignment fail to capture the varying criticality of decisions across generation timesteps, resulting in inefficient exploration and suboptimal convergence. To remedy this shortcoming, we introduce \textbf{TempFlow-GRPO} (Temporal Flow GRPO), a principled GRPO framework that captures and exploits the temporal structure inherent in flow-based generation. TempFlow-GRPO introduces three key innovations: (i) a trajectory branching mechanism that provides process rewards by concentrating stochasticity at designated branching points, enabling precise credit assignment without requiring specialized intermediate reward models; (ii) a noise-aware weighting scheme that modulates policy optimization according to the intrinsic exploration potential of each timestep, prioritizing learning during high-impact early stages while ensuring stable refinement in later phases; and (iii) a seed group strategy that controls for initialization effects to isolate exploration contributions. These innovations endow the model with temporally-aware optimization that respects the underlying generative dynamics, leading to state-of-the-art performance in human preference alignment and text-to-image benchmarks.

cs.CV↗

arXiv2025

OmniGen2: Towards Instruction-Aligned Multimodal Generation

In this work, we introduce OmniGen2, a versatile and open-source generative model designed to provide a unified solution for diverse generation tasks, including text-to-image, image editing, and in-context generation. Unlike OmniGen v1, OmniGen2 features two distinct decoding pathways for text and image modalities, utilizing unshared parameters and a decoupled image tokenizer. This design enables OmniGen2 to build upon existing multimodal understanding models without the need to re-adapt VAE inputs, thereby preserving the original text generation capabilities. To facilitate the training of OmniGen2, we developed comprehensive data construction pipelines, encompassing image editing and in-context generation data. Additionally, we introduce a reflection mechanism tailored for image generation tasks and curate a dedicated reflection dataset based on OmniGen2. Despite its relatively modest parameter size, OmniGen2 achieves competitive results on multiple task benchmarks, including text-to-image and image editing. To further evaluate in-context generation, also referred to as subject-driven tasks, we introduce a new benchmark named OmniContext. OmniGen2 achieves state-of-the-art performance among open-source models in terms of consistency. We will release our models, training code, datasets, and data construction pipeline to support future research in this field. Project Page: https://vectorspacelab.github.io/OmniGen2; GitHub Link: https://github.com/VectorSpaceLab/OmniGen2

cs.CV↗

arXiv2024

DropEdge not Foolproof: Effective Augmentation Method for Signed Graph Neural Networks

The paper discusses signed graphs, which model friendly or antagonistic relationships using edges marked with positive or negative signs, focusing on the task of link sign prediction. While Signed Graph Neural Networks (SGNNs) have advanced, they face challenges like graph sparsity and unbalanced triangles. The authors propose using data augmentation (DA) techniques to address these issues, although many existing methods are not suitable for signed graphs due to a lack of side information. They highlight that the random DropEdge method, a rare DA approach applicable to signed graphs, does not enhance link sign prediction performance. In response, they introduce the Signed Graph Augmentation (SGA) framework, which includes a structure augmentation module to identify candidate edges and a strategy for selecting beneficial candidates, ultimately improving SGNN training. Experimental results show that SGA significantly boosts the performance of SGNN models, with a notable 32.3% improvement in F1-micro for SGCN on the Slashdot dataset.

cs.LG↗

arXiv2023

Generative Meta-Learning for Zero-Shot Relation Triplet Extraction

Zero-shot Relation Triplet Extraction (ZeroRTE) aims to extract relation triplets from texts containing unseen relation types. This capability benefits various downstream information retrieval (IR) tasks. The primary challenge lies in enabling models to generalize effectively to unseen relation categories. Existing approaches typically leverage the knowledge embedded in pre-trained language models to accomplish the generalization process. However, these methods focus solely on fitting the training data during training, without specifically improving the model's generalization performance, resulting in limited generalization capability. For this reason, we explore the integration of bi-level optimization (BLO) with pre-trained language models for learning generalized knowledge directly from the training data, and propose a generative meta-learning framework which exploits the `learning-to-learn' ability of meta-learning to boost the generalization capability of generative models. Specifically, we introduce a BLO approach that simultaneously addresses data fitting and generalization. This is achieved by constructing an upper-level loss to focus on generalization and a lower-level loss to ensure accurate data fitting. Building on this, we subsequently develop three generative meta-learning methods, each tailored to a distinct category of meta-learning. Extensive experimental results demonstrate that our framework performs well on the ZeroRTE task. Our code is available at https://github.com/leeworry/TGM-MetaLearning.

cs.CL↗

arXiv2021

From Consensus to Disagreement: Multi-Teacher Distillation for Semi-Supervised Relation Extraction

Lack of labeled data is a main obstacle in relation extraction. Semi-supervised relation extraction (SSRE) has been proven to be a promising way for this problem through annotating unlabeled samples as additional training data. Almost all prior researches along this line adopt multiple models to make the annotations more reliable by taking the intersection set of predicted results from these models. However, the difference set, which contains rich information about unlabeled data, has been long neglected by prior studies. In this paper, we propose to learn not only from the consensus but also the disagreement among different models in SSRE. To this end, we develop a simple and general multi-teacher distillation (MTD) framework, which can be easily integrated into any existing SSRE methods. Specifically, we first let the teachers correspond to the multiple models and select the samples in the intersection set of the last iteration in SSRE methods to augment labeled data as usual. We then transfer the class distributions for samples in the difference set as soft labels to guide the student. We finally perform prediction using the trained student model. Experimental results on two public datasets demonstrate that our framework significantly promotes the performance of the base SSRE methods with pretty low computational cost.

cs.CL↗

arXiv2020

Exploit Multiple Reference Graphs for Semi-supervised Relation Extraction

Manual annotation of the labeled data for relation extraction is time-consuming and labor-intensive. Semi-supervised methods can offer helping hands for this problem and have aroused great research interests. Existing work focuses on mapping the unlabeled samples to the classes to augment the labeled dataset. However, it is hard to find an overall good mapping function, especially for the samples with complicated syntactic components in one sentence. To tackle this limitation, we propose to build the connection between the unlabeled data and the labeled ones rather than directly mapping the unlabeled samples to the classes. Specifically, we first use three kinds of information to construct reference graphs, including entity reference, verb reference, and semantics reference. The goal is to semantically or lexically connect the unlabeled sample(s) to the labeled one(s). Then, we develop a Multiple Reference Graph (MRefG) model to exploit the reference information for better recognizing high-quality unlabeled samples. The effectiveness of our method is demonstrated by extensive comparison experiments with the state-of-the-art baselines on two public datasets.

cs.CL↗

arXiv2009

Scaling in Plateau-to-Plateau Transition: A Direct Connection of Quantum Hall Systems with Anderson Localization Model

The quantum Hall plateau transition was studied at temperatures down to 1 mK in a random alloy disordered high mobility two-dimensional electron gas. A perfect power-law scaling with κ=0.42 was observed from 1.2K down to 12mK. This perfect scaling terminates sharply at a saturation temperature of T_s~10mK. The saturation is identified as a finite-size effect when the quantum phase coherence length (L_ϕ ~ T^{-p/2}) reaches the sample size (W) of millimeter scale. From a size dependent study, T_s \propto W^{-1} was observed and p=2 was obtained. The exponent of the localization length, determined directly from the measured κand p, is ν=2.38, and the dynamic critical exponent z = 1.

cond-mat.mes-hall↗