Yang Yu

Zehan Qi, Xiao Liu, Iat Long Iong et al. · tsinghua

Large language models (LLMs) have shown remarkable potential as autonomous agents, particularly in web-based tasks. However, existing LLM web agents heavily rely on expensive proprietary LLM APIs, while open LLMs lack the necessary decision-making capabilities. This paper introduces WebRL, a self-evolving online curriculum reinforcement learning framework designed to train high-performance web agents using open LLMs. WebRL addresses three key challenges in building LLM web agents, including the scarcity of training tasks, sparse feedback signals, and policy distribution drift in online learning. Specifically, WebRL incorporates 1) a self-evolving curriculum that generates new tasks from unsuccessful attempts, 2) a robust outcome-supervised reward model (ORM), and 3) adaptive reinforcement learning strategies to ensure consistent improvements. We apply WebRL to transform open Llama-3.1 and GLM-4 models into proficient web agents. On WebArena-Lite, WebRL improves the success rate of Llama-3.1-8B from 4.8% to 42.4%, and from 6.1% to 43% for GLM-4-9B. These open models significantly surpass the performance of GPT-4-Turbo (17.6%) and GPT-4o (13.9%) and outperform previous state-of-the-art web agents trained on open LLMs (AutoWebGLM, 18.2%). Our findings demonstrate WebRL's effectiveness in bridging the gap between open and proprietary LLM-based web agents, paving the way for more accessible and powerful autonomous web interaction systems.

Fan-Ming Luo, Zuolin Tu, Zefang Huang et al.

Real-world decision-making tasks are usually partially observable Markov decision processes (POMDPs), where the state is not fully observable. Recent progress has demonstrated that recurrent reinforcement learning (RL), which consists of a context encoder based on recurrent neural networks (RNNs) for unobservable state prediction and a multilayer perceptron (MLP) policy for decision making, can mitigate partial observability and serve as a robust baseline for POMDP tasks. However, previous recurrent RL methods face training stability issues due to the gradient instability of RNNs. In this paper, we propose Recurrent Off-policy RL with Context-Encoder-Specific Learning Rate (RESeL) to tackle this issue. Specifically, RESeL uses a lower learning rate for context encoder than other MLP layers to ensure the stability of the former while maintaining the training efficiency of the latter. We integrate this technique into existing off-policy RL methods, resulting in the RESeL algorithm. We evaluated RESeL in 18 POMDP tasks, including classic, meta-RL, and credit assignment scenarios, as well as five MDP locomotion tasks. The experiments demonstrate significant improvements in training stability with RESeL. Comparative results show that RESeL achieves notable performance improvements over previous recurrent RL baselines in POMDP tasks, and is competitive with or even surpasses state-of-the-art methods in MDP tasks. Further ablation studies highlight the necessity of applying a distinct learning rate for the context encoder.

Ye Yu, Yaoning Yu, Haohan Wang

Large reasoning models (LRMs) such as Claude 3.7 Sonnet and OpenAI o1 achieve strong performance on mathematical benchmarks using lengthy chain-of-thought (CoT) reasoning, but the resulting traces are often unnecessarily verbose. This inflates token usage and cost, limiting deployment in latency-sensitive or API-constrained settings. We introduce PREMISE (PRompt-based Efficient Mathematical Inference with Strategic Evaluation), a prompt-only framework that reduces reasoning overhead without modifying model weights. PREMISE combines trace-level diagnostics with gradient-inspired prompt optimization to minimize redundant computation while preserving answer accuracy. The approach jointly optimizes brevity and correctness through a multi-objective textual search that balances token length and answer validity. Unlike prior work, PREMISE runs in a single-pass black-box interface, so it can be applied directly to commercial LLMs. On GSM8K, SVAMP, and Math500 we match or exceed baseline accuracy ($96\%\rightarrow96\%$ with Claude, $91\%\rightarrow92\%$ with Gemini) while reducing reasoning tokens by up to $87.5\%$ and cutting dollar cost by $69$--$82\%$. These results show that prompt-level optimization is a practical and scalable path to efficient LRM inference without compromising reasoning quality.

Heng Tan, Hua Yan, Yu Yang

While reinforcement learning (RL) has achieved notable success in various domains, training effective policies for complex tasks remains challenging. Agents often converge to local optima and fail to maximize long-term rewards. Existing approaches to mitigate training bottlenecks typically fall into two categories: (i) Automated policy refinement, which identifies critical states from past trajectories to guide policy updates, but suffers from costly and uncertain model training; and (ii) Human-in-the-loop refinement, where human feedback is used to correct agent behavior, but this does not scale well to environments with large or continuous action spaces. In this work, we design a large language model-guided policy modulation framework that leverages LLMs to improve RL training without additional model training or human intervention. We first prompt an LLM to identify critical states from a sub-optimal agent's trajectories. Based on these states, the LLM then provides action suggestions and assigns implicit rewards to guide policy refinement. Experiments across standard RL benchmarks demonstrate that our method outperforms state-of-the-art baselines, highlighting the effectiveness of LLM-based explanations in addressing RL training bottlenecks.

Yile Yu, Anzhi Xu, Yi Wang

Causal inference in observational panel data has become a central concern in economics,policy analysis,and the broader social sciences.To address the core contradiction where traditional difference-in-differences (DID) struggles with high-dimensional confounding variables in observational panel data,while machine learning (ML) lacks causal structure interpretability,this paper proposes an innovative framework called S-DIDML that integrates structural identification with high-dimensional estimation.Building upon the structure of traditional DID methods,S-DIDML employs structured residual orthogonalization techniques (Neyman orthogonality+cross-fitting) to retain the group-time treatment effect (ATT) identification structure while resolving high-dimensional covariate interference issues.It designs a dynamic heterogeneity estimation module combining causal forests and semi-parametric models to capture spatiotemporal heterogeneity effects.The framework establishes a complete modular application process with standardized Stata implementation paths.The introduction of S-DIDML enriches methodological research on DID and DDML innovations, shifting causal inference from method stacking to architecture integration.This advancement enables social sciences to precisely identify policy-sensitive groups and optimize resource allocation.The framework provides replicable evaluation tools, decision optimization references,and methodological paradigms for complex intervention scenarios such as digital transformation policies and environmental regulations.

Jing Yang, Chun Ouyang, Wil M. P. van der Aalst et al.

Providing appropriate structures around human resources can streamline operations and thus facilitate the competitiveness of an organization. To achieve this goal, modern organizations need to acquire an accurate and timely understanding of human resource grouping while faced with an ever-changing environment. The use of process mining offers a promising way to help address the need through utilizing event log data stored in information systems. By extracting knowledge about the actual behavior of resources participating in business processes from event logs, organizational models can be constructed, which facilitate the analysis of the de facto grouping of human resources relevant to process execution. Nevertheless, open research gaps remain to be addressed when applying the state-of-the-art process mining to analyze resource grouping. For one, the discovery of organizational models has only limited connections with the context of process execution. For another, a rigorous solution that evaluates organizational models against event log data is yet to be proposed. In this paper, we aim to tackle these research challenges by developing a novel framework built upon a richer definition of organizational models coupling resource grouping with process execution knowledge. By introducing notions of conformance checking for organizational models, the framework allows effective evaluation of organizational models, and therefore provides a foundation for analyzing and improving resource grouping based on event logs. We demonstrate the feasibility of this framework by proposing an approach underpinned by the framework for organizational model discovery, and also conduct experiments on real-life event logs to discover and evaluate organizational models.