In modern software development, developers frequently need to understand code behavior at a glance—whether reviewing pull requests, debugging issues, or navigating unfamiliar codebases. This ability to reason about dynamic program behavior is fundamental to effective software engineering and increasingly supported by Large Language Models (LLMs). However, existing studies on code reasoning focus primarily on isolated code snippets, overlooking the complexity of real-world scenarios involving external API interactions and unfamiliar functions. This gap hinders our understanding of what truly makes code reasoning challenging for LLMs across diverse programming contexts.

Complex logic errors in LLM-generated code are challenging to diagnose and repair. While existing LLM-based self-repair approaches conduct intensive static semantic analysis or rely on superficial execution logs, they miss the in-depth runtime behaviors that often expose bug root causes—lacking the interactive dynamic analysis capabilities that make human debugging effective. We present InspectCoder, the first agentic program repair system that empowers LLMs to actively conduct dynamic analysis via interactive debugger control. Our dual-agent framework enables strategic breakpoint placement, targeted state inspection, and incremental runtime experimentation within stateful debugger sessions. Unlike existing methods that follow fixed log collection procedures, InspectCoder adaptively inspects and perturbs relevant intermediate states at runtime, and leverages immediate process rewards from debugger feedback to guide multi-step reasoning, transforming LLM debugging paradigm from blind trial-and-error into systematic root cause diagnosis. We conduct comprehensive experiments on two challenging self-repair benchmarks: BigCodeBench-R and LiveCodeBench-R. InspectCoder achieves 5.10%–60.37% relative improvements in repair accuracy over the strongest baseline, while delivering 1.67x-2.24x superior bug-fix efficiency respectively.We also contribute InspectWare, an open-source middleware that abstracts debugger complexities and maintains stateful debugging sessions across mainstream Python testing frameworks. Our work provides actionable insight into the interactive LLM-debugger systems, demonstrating the significant potential of LLM-driven dynamic analysis for automated software engineering.

Yunkun Wang, Yue Zhang, Zhen Qin, Chen Zhi, Binhua Li, Fei Huang, Yongbin Li, Shuiguang Deng. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Federated Learning (FL) enables collaborative model training across multiple edge devices without sharing raw data, yet existing FL frameworks often assume static data domains, limiting their applicability to real-world scenarios where data evolves over time. To address this, Federated Continual Learning (FCL) integrates continual learning into FL, but conventional strategies such as data replay are impractical due to privacy and storage constraints. In this paper, we propose PP-FCL, a privacy-preserving FCL framework that mitigates catastrophic forgetting without storing sensitive client data. PP-FCL employs a server-side generative model to synthesize representative samples of previously learned tasks, enhancing data diversity, preserving characteristic class features, and refining decision boundaries. On the client side, an improved contrastive incremental learning loss and a carefully designed feature distillation method decouple old and new knowledge, ensuring a balanced trade-off between plasticity and stability. As a result, PP-FCL not only enhances the model’s representational capabilities but also adapts effectively to non-stationary data distributions, maintaining robust performance in privacy-sensitive, evolving federated environments. Empirical results on CIFAR-10, CIFAR-100 and TinyImageNet demonstrate that PP-FCL outperforms state-of-the-art baselines by approximately 5–6% in average accuracy. This substantial improvement highlights PP-FCL’s effectiveness in preserving model performance under evolving conditions, ensuring robust and adaptive learning in dynamic federated environments.

A fully-integrated half-bridge driver with dual N-type power transistors is proposed in this paper, which adopts on-chip charge pumps with integrated capacitors based on charge sharing to avoid external bootstrap components. Without bootstrap-capacitor charging paths through low-side power transistors and high-frequency high-power switching nodes, overcharge and large electro-magnetic interference (EMI) can be avoided by eliminating parasitic interconnect inductance and body diode. Adaptive dead time is employed to enhance efficiency, enabling the system to achieve appropriate dead times in various applications. Besides, 100% duty-cycle control is realized by duty-cycle detection and additional continuous charge pump, which can charge the on-chip driver capacitor during on-time, the restriction of minimum off-time is eliminated and the proposed driver can work at full duty cycle range. The proposed charge-sharing-based (CSB) half-bridge driver with any duty-cycle capability embedded in a buck converter with 2A load is implemented in a 0.18 μm BCD technology, which occupies an active area of 0.151mm2. Verification results demonstrate that a 2.2 MHz switching frequency is realized with 2.4V–5.5V input voltage and a peak efficiency of 96.8% is achieved for 5 to 4V conversion. The minimum regulation voltage difference between input voltage and output voltage is 70 mV with the help of 100% duty-cycle circuit.

Ellipse detection is of great significance in the fields of image processing and computer vision. Accurate, stable and direct ellipse detection in real-world images has always been a key issue. Therefore, an ellipse detection method is proposed on the basis of the constructed three-intersection-chord-invariant. First, in the inflexion point detection, the PCA minimum bounding box considering the distribution characteristics of edge points is studied to achieve the more refined line segment screening. Second, a multi-scale inflexion point detection method is proposed to effectively avoid over-segmentation of small arc segments, providing assurance for more reasonable and reliable arc segment combinations. Then, the 20 precisely classified arc segment combinations are refined into 4 combinations. A number of non-homologous arc segment combinations can be quickly removed to reduce incorrect combinations by the constructed midpoint distance constraint and quadrant constraint. Moreover, in order to accurately reflect the strict arc segment combination constraints of geometric features of ellipses, a three-intersection-chord-invariant model of ellipses is established with strong constraint of relative distances among five constraint points, by which a more robust initial ellipse set of homologous arc segment combinations is further obtained. Finally, ellipse validation and clustering are performed on the initial set of ellipses to obtain the high-precision ellipses. The algorithm accuracy of the ellipse detection method is experimentally validated on 6 publicly available datasets and 2 established wheel rim datasets.

Weixiang Yan, Haitian Liu, Yunkun Wang, Yunzhe Li, Qian Chen, Wen Wang, Tingyu Lin, Weishan Zhao, Li Zhu, Hari Sundaram, Shuiguang Deng. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Deep Learning (DL) technologies have been widely adopted to tackle various tasks. In this process, through software dependencies, a multi-layer DL supply chain (SC) is formed, with DL frameworks acting as the root, DL packages acting as the bridge nodes, and downstream DL projects acting as the periphery. However, most Open Source Software (OSS) projects may fail. Considering the crucial position of DL packages in the DL SC, to foster the sustainable development of DL SCs and DL packages, we aim to forecast the long-term sustainability of DL packages. Here, sustained activity is adopted as the main proxy of sustainability, and the sustainability status is classified as “sus-tainable” or “dormant”. Relatedly, a DL package is considered as “sustainable” if it has sustained activity in its last 12 months. Otherwise, it is deemed as “dormant”. To this end, we propose an approach that begins with obtaining longitudinal features for each DL package in each month. Then, we develop a model to forecast the sustainability of DL packages by incorporating the longitudinal features, which can aptly predict sustainability with an accuracy of up to 0.81. Subsequently, an interpretable module is developed to interpret the determinants (i.e., important features) that impact the sustainability of DL packages. Finally, we generate sustainability trajectories for each DL package to better understand the monthly changes of their sustainability status. Our findings uncover that for most DL packages, fewer but more centralized developers and a balanced collaboration are more likely to help sustain the DL packages. Furthermore, although some DL packages are sustainable, their sustainability trajectories present statistically decreasing trends over time. Based on the findings, we shed light on the dynamic sustainability of DL packages, highlight future research directions, and provide practical suggestions to DL package maintainers, developers, users, and software engineering researchers.