arXiv subjects

Junbao Hu

Publications and source records attributed to Junbao Hu.

3 recordsLinked to original sources

arXiv2026

Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs

Training effective software engineering agents requires large volumes of task-specific trajectories, incurring substantial data construction costs. Inspired by the "Less-Is-More" hypothesis in mathematical reasoning, we investigate its extension to agentic scenarios and propose an end-to-end training framework that achieves superior agentic capabilities with fewer but higher-quality training trajectories. This is achieved via STITCH (Sliding-memory Trajectory Inference and Task Chunking Heuristic), a coarse-to-fine mechanism that filters low-value noise and retains decision-critical tokens to maximize training signal quality. We conduct experiments across multiple agent frameworks (e.g., mini-SWE-agent, MSWE-agent), model scales (30B to 355B), and multilingual settings (Python, Java, and ArkTS). On SWE-bench Verified, models trained with STITCH achieve up to 63.16% relative improvement over base models. On Multi-SWE-bench (Java), MiniMax-M2.5-STITCH achieves 43.75% with our CodeArts Agent scaffold (+16.67%). On HarmonyOS (ArkTS), GLM-4.7-STITCH improves the compilation pass rate to 61.31% (+43.34%) with less than 1K training trajectories. Our results confirm that the "Less-Is-More" paradigm generalizes effectively to complex agentic tasks across diverse languages and model scales.

cs.SE↗

arXiv2025

Physical embedding machine learning force fields for organic systems

Machine learning force fields possess unprecedented potential in achieving both accuracy and efficiency in molecular simulations. Nevertheless, their application in organic systems is often hindered by structural collapse during simulation and significant deviations in the prediction of macroscopic properties. Here, two physics-embedded strategies are introduced to overcome these limitations. First, a physics-inspired self-adaptive bond-length sampling method achieves long-timescale stable simulations by requiring only several tens of single-molecule data sets, and has been validated across molecular systems, including engineering fluids, polypeptides, and pharmaceuticals. Second, a top-down intermolecular correction strategy based on a physical equation is introduced. This strategy requires only a small amount of simulation data and completes the optimization of tunable parameters within a few hours on a single RTX 4090 GPU, significantly reducing errors in density and viscosity, as validated in systems including ethylene carbonate, ethyl acetate, and dimethyl carbonate. Together, these approaches directly integrate physical insights into the machine learning models, thereby enhancing robustness and generalizability, and providing a scalable pathway for physics-embedded machine learning force fields.

physics.comp-ph↗

arXiv2023

Efficient Machine Learning Force Field for Large-Scale Molecular Simulations of Organic Systems

To address the computational challenges of ab initio molecular dynamics and the accuracy limitations of empirical force fields, the introduction of machine learning force fields has proven effective in various systems including metals and inorganic materials. However, in large-scale organic systems, the application of machine learning force fields is often hindered by impediments such as the complexity of long-range intermolecular interactions and molecular conformations, as well as the instability in long-time molecular simulations. Therefore, we propose a universal multiscale higher-order equivariant model combined with active learning techniques, efficiently capturing the complex long-range intermolecular interactions and molecular conformations. Compared to existing equivariant models, our model achieves the highest predictive accuracy, and magnitude-level improvements in computational speed and memory efficiency. In addition, a bond length stretching method is designed to improve the stability of long-time molecular simulations. Utilizing only 901 samples from a dataset with 120 atoms, our model successfully extends high precision to systems with hundreds of thousands of atoms. These achievements guarantee high predictive accuracy, fast simulation speed, minimal memory consumption, and robust simulation stability, satisfying the requirements for high-precision and long-time molecular simulations in large-scale organic systems.

cond-mat.soft↗