arXiv subjects

Yunfei Ma

Publications and source records attributed to Yunfei Ma.

12 recordsLinked to original sources

arXiv2026

Step-Level Preference Learning for Generative Agents in Social Simulations

Large language model (LLM)-based generative agents simulate human behavior through long-horizon decision-making processes that comprise intermediate steps such as planning, memory retrieval, reflection, and action selection. However, fine-grained human annotations of these intermediate steps remain scarce, and existing agents are not grounded in human preferences over such intermediate decisions. To address this gap, we introduce \method, an interactive simulation interface that enables us to collect step-level human preference supervision over agent decision trajectories, leading to a dataset of 57K fine-grained annotations. We conduct step-level preference learning on open-weight language models using supervised finetuning and direct preference optimization on this data, consistently improving simulation fidelity, coordination, and interaction quality, and inducing more socially effective agent behavior. Our results show that step-level human supervision is an effective training signal for improving both local decision quality and long-horizon agent behavior.

cs.AI↗

arXiv2026

JT-SAFE-V2: Safety-by-Design Foundation Model with World-Context Data

We introduce JT-Safe-V2, a large language model designed to advance the safety and trustworthiness of foundation models, extending our previous JT-Safe model toward a more comprehensive safety-by-design paradigm. JT-Safe-V2 emphasizes the joint optimization of general intelligence and safety-by-design through several key innovations: enriching pre-training data with contextual world knowledge, high-certainty pre-training procedures, and safety strengthening post-training mechanisms for enterprise-oriented agentic capabilities. Building on these safety-enhanced foundation models, we propose Safe-MoMA (Safe Mixture of Models and Agents), a framework that enables traceable and efficient inference through the orchestrated deployment of multiple models and agents. Extensive evaluations demonstrate that JT-Safe-V2 achieves state-of-the-art performance across both general intelligence and safety benchmarks. Moreover, Safe-MoMA reduces inference costs by more than 30\% compared to using the largest standalone model baseline while maintaining comparable performance. To facilitate future research on safety-by-design foundation models, we publicly release the post-trained JT-Safe-V2-35B model checkpoint.

cs.AI↗

arXiv2026

Strategy-Aware Optimization Modeling with Reasoning LLMs

Large language models (LLMs) can generate syntactically valid optimization programs, yet often struggle to reliably choose an effective modeling strategy, leading to incorrect formulations and inefficient solver behavior. We propose SAGE, a strategy-aware framework that makes Modeling Strategy explicit in both data construction and post-training. SAGE builds a solver-verified multi-strategy dataset and trains a student model with supervised fine-tuning followed by Segment-Weighted GRPO using a composite reward over format compliance, correctness, and solver efficiency. Across eight benchmarks spanning synthetic and real-world settings, SAGE improves average pass@1 from 72.7 to 80.3 over the strongest open-source baseline. With multiple generations, SAGE discovers more distinct correct formulations and improves component-level diversity at pass@16 by 19-29%. At the largest scale, SAGE produces more compact constraint systems with 14.2% fewer constraints than the baseline, consistent with solver-efficient modeling. Overall, these results show that making Modeling Strategy explicit improves automated optimization modeling. Code is available at https://github.com/rachhhhing/SAGE.

cs.AI↗

arXiv2026

CCR-Bench: A Comprehensive Benchmark for Evaluating LLMs on Complex Constraints, Control Flows, and Real-World Cases

Enhancing the ability of large language models (LLMs) to follow complex instructions is critical for their deployment in real-world applications. However, existing evaluation methods often oversimplify instruction complexity as a mere additive combination of atomic constraints, failing to adequately capture the high-dimensional complexity arising from the intricate interplay of content and format, logical workflow control, and real-world applications. This leads to a significant gap between current evaluation practices and practical demands. To bridge this gap, we introduce CCR-Bench, a novel benchmark designed to assess LLMs' adherence to complex instructions. CCR-Bench is characterized by: (1) deep entanglement of content and formatting requirements in task specifications; (2) instructions that involve intricate task decomposition, conditional reasoning, and procedural planning; and (3) evaluation samples derived entirely from real-world industrial scenarios. Extensive experiments on CCR-Bench demonstrate that even state-of-the-art models exhibit substantial performance deficiencies, clearly quantifying the gap between current LLM capabilities and the demands of realworld instruction understanding. We believe that CCR-Bench offers a more rigorous and realistic evaluation framework, advancing the development of LLMs toward the next generation of models capable of understanding and executing complex tasks in industrial applications.

cs.CL↗

arXiv2026

Beyond One-Size-Fits-All: Adaptive Subgraph Denoising for Zero-Shot Graph Learning with Large Language Models

Graph-based tasks in the zero-shot setting remain a significant challenge due to data scarcity and the inability of traditional Graph Neural Networks (GNNs) to generalize to unseen domains or label spaces. While recent advancements have transitioned toward leveraging Large Language Models (LLMs) as predictors to enhance GNNs, these methods often suffer from cross-modal alignment issues. A recent paradigm (i.e., Graph-R1) overcomes the aforementioned architectural dependencies by adopting a purely text-based format and utilizing LLM-based graph reasoning, showing improved zero-shot generalization. However, it employs a task-agnostic, one-size-fits-all subgraph extraction strategy, which inevitably introduces significant structural noise--irrelevant neighbors and edges--that distorts the LLMs' receptive field and leads to suboptimal predictions. To address this limitation, we introduce GraphSSR, a novel framework designed for adaptive subgraph extraction and denoising in zero-shot LLM-based graph reasoning. Specifically, we propose the SSR pipeline, which dynamically tailors subgraph extraction to specific contexts through a "Sample-Select-Reason" process, enabling the model to autonomously filter out task-irrelevant neighbors and overcome the one-size-fits-all issue. To internalize this capability, we develop SSR-SFT, a data synthesis strategy that generates high-quality SSR-style graph reasoning traces for supervised fine-tuning of LLMs. Furthermore, we propose SSR-RL, a two-stage reinforcement learning framework that explicitly regulates sampling and selection operations within the proposed SSR pipeline designed for adaptive subgraph denoising. By incorporating Authenticity-Reinforced and Denoising-Reinforced RL, we guide the model to achieve accurate predictions using parsimonious, denoised subgraphs for reasoning.

cs.LG↗

arXiv2026

VirtualCrime: Evaluating the Criminal Potential and Agentic Behaviors of Large Language Models via Sandbox Simulation

Large language models (LLMs) have shown strong capabilities in multi-step decision-making, and are increasingly integrated into real-world agentic applications. While existing safety evaluations focus on static interactions or refusal rates, it remains largely underexplored what level of threat these models pose once an adversary successfully assigns them a malicious role. To evaluate this agentic misuse risk, we propose VirtualCrime, a dynamic multi-agent sandbox framework that elicits and measures multi-step criminal behavior from LLMs. We design 40 diverse crime tasks within this framework, covering 11 realistic maps and 14 objectives such as theft, robbery, and kidnapping. We evaluate 8 strong LLMs and find that: (1) even with robust safety alignment, all evaluated models exhibit alarming agentic capabilities to generate and execute detailed multi-step criminal plans when placed in an adversarial red-teaming setting; (2) models frequently suffer from behavioral alignment failures, exhibiting a high propensity for over-harm by escalating to violent means far beyond what the task requires; (3) their most sophisticated criminal behaviors concentrate in social deception and team coordination rather than technical exploitation. Our work highlights a critical gap in current safety paradigms, emphasizing the urgent need for behavioral alignment in autonomous agent deployments.

cs.CR↗

arXiv2025

JT-Safe: Intrinsically Enhancing the Safety and Trustworthiness of LLMs

The hallucination and credibility concerns of large language models (LLMs) are global challenges that the industry is collectively addressing. Recently, a significant amount of advances have been made on post-training and inference techniques to mitigate these challenges. However, it is widely agreed that unsafe and hallucinations of LLMs intrinsically originate from pre-training, involving pre-training data and the next-token prediction learning mechanism. In this paper, we focus on enhancing pre-training data to improve the trustworthiness and safety of LLMs. Since the data is vast, it's almost impossible to entirely purge the data of factual errors, logical inconsistencies, or distributional biases. Moreover, the pre-training data lack grounding in real-world knowledge. Each piece of data is treated as a sequence of tokens rather than as a representation of a part of the world. To overcome these issues, we propose approaches to enhancing our pre-training data with its context in the world and increasing a substantial amount of data reflecting industrial scenarios. We argue that most source data are created by the authors for specific purposes in a certain spatial-temporal context. They have played a role in the real world. By incorporating related world context information, we aim to better anchor pre-training data within real-world scenarios, thereby reducing uncertainty in model training and enhancing the model's safety and trustworthiness. We refer to our Data with World Context as DWC. We continue pre-training an earlier checkpoint of JT-35B-Base with 1.5 trillion of DWC tokens. We introduce our post-training procedures to activate the potentials of DWC. Compared with the Qwen model of a similar scale, JT-Safe-35B achieves an average performance improvement of 1.79% on the Safety and Trustworthy evaluation benchmarks, while being pretrained with only 6.2 trillion tokens.

cs.CL↗

arXiv2023

CloudEval-YAML: A Practical Benchmark for Cloud Configuration Generation

Among the thriving ecosystem of cloud computing and the proliferation of Large Language Model (LLM)-based code generation tools, there is a lack of benchmarking for code generation in cloud-native applications. In response to this need, we present CloudEval-YAML, a practical benchmark for cloud configuration generation. CloudEval-YAML tackles the diversity challenge by focusing on YAML, the de facto standard of numerous cloud-native tools. We develop the CloudEval-YAML benchmark with practicality in mind: the dataset consists of hand-written problems with unit tests targeting practical scenarios. We further enhanced the dataset to meet practical needs by rephrasing questions in a concise, abbreviated, and bilingual manner. The dataset consists of 1011 problems that take more than 1200 human hours to complete. To improve practicality during evaluation, we build a scalable evaluation platform for CloudEval-YAML that achieves a 20 times speedup over a single machine. To the best of our knowledge, the CloudEval-YAML dataset is the first hand-written dataset targeting cloud-native applications. We present an in-depth evaluation of 12 LLMs, leading to a deeper understanding of the problems and LLMs, as well as effective methods to improve task performance and reduce cost.

cs.DC↗

arXiv2022

Technical Report: Implementation of Single Packet Number Space in Multi-Path QUIC

Over the past few of years, we have witnessed increasing interests in the use cases of multi-path QUIC from both industry and academia. For example, Alibaba deployed XLINK, a QoE-driven multi-path QUIC solution, in Taobao short video and showed benefits in both reduced tail latency and video re-buffering. For the time being, the multi-path QUIC protocol is in the process of standardization at the IETF QUIC working group, with the draft recently updated to version 02. The focus of the draft is to provide basic guidance on the implementation so that we can encourage more exploration, testing, and finally, an accelerated adoption of this technology. However, draft-02 has brought up an open issue on whether the multi-path QUIC should be implemented using single packet number space (SPNS) or multiple packet number space (MPNS), as in the current draft, both options co-exist. Knowing that one cannot draw a solid conclusion without experiments, we implement both SPNS and MPNS at Alibaba and measured their performance. The goal is to help the community better understand the implication, and we hope this report can be a useful resource for engineers and researchers who are interested in deploying multi-path QUIC.

cs.NI↗

arXiv2020

Towards Flexible Wireless Charging for Medical Implants Using Distributed Antenna System

This paper presents the design, implementation and evaluation of In-N-Out, a software-hardware solution for far-field wireless power transfer. In-N-Out can continuously charge a medical implant residing in deep tissues at near-optimal beamforming power, even when the implant moves around inside the human body. To accomplish this, we exploit the unique energy ball pattern of distributed antenna array and devise a backscatter-assisted beamforming algorithm that can concentrate RF energy on a tiny spot surrounding the medical implant. Meanwhile, the power levels on other body parts stay in low level, reducing the risk of overheating. We prototype In-N-Out on 21 software-defined radios and a printed circuit board (PCB). Extensive experiments demonstrate that In-N-Out achieves 0.37~mW average charging power inside a 10~cm-thick pork belly, which is sufficient to wirelessly power a range of commercial medical devices. Our head-to-head comparison with the state-of-the-art approach shows that In-N-Out achieves 5.4$\times$--18.1$\times$ power gain when the implant is stationary, and 5.3$\times$--7.4$\times$ power gain when the implant is in motion.

cs.NI↗

arXiv2014

A FETI-DP preconditioner of discontinuous Galerkin method for multiscale problems in high constrast media

In this paper we consider second order elliptic partial differential equations with highly varying (heterogeneous) coefficients on a two-dimensional region. The problems are discretized by a composite finite element (FE) and discontinuous Galerkin (DG) Method. The fine grids are in general nonmatching across the subdomain boundaries, and the subdomain partitioning does not need to resolve the jumps in the coefficient. A FETI-DP preconditioner is proposed and analyzed to solve the resulting linear system. Numerical results are presented to support our theory.

math.NA↗

arXiv2012

Domain Decomposition Preconditioners for a Discontinuous Galerkin Formulation of a Multiscale Elliptic Problem

In this paper, we propose a domain decomposition method for multiscale second order elliptic partial differential equations with highly varying coefficients. The method is based on a discontinuous Galerkin formulation. We present both a nonoverlapping and an overlapping version of the method. We prove that the condition number bound of the preconditioned algebraic system in either case can be made independent of the coefficients under certain assumptions. Also, in our analysis, we do not need to assume that the coefficients are continuous across the coarse grid boundaries. The analysis and the condition number bounds are new, and contribute towards further extension of the theory for the discontinuous Galerkin discretization for multiscale problems.

math.NA↗