arXiv subjects

Xiaocong Zhou

Publications and source records attributed to Xiaocong Zhou.

5 recordsLinked to original sources

arXiv2026

AlphaEval: Evaluating Agents in Production

The rapid deployment of AI agents in commercial settings has outpaced the development of evaluation methodologies that reflect production realities. Existing benchmarks measure agent capabilities through retrospectively curated tasks with well-specified requirements and deterministic metrics -- conditions that diverge fundamentally from production environments where requirements contain implicit constraints, inputs are heterogeneous multi-modal documents with information fragmented across sources, tasks demand undeclared domain expertise, outputs are long-horizon professional deliverables, and success is judged by domain experts whose standards evolve over time. We present AlphaEval, a production-grounded benchmark of 94 tasks sourced from seven companies deploying AI agents in their core business, spanning six O*NET (Occupational Information Network) domains. Unlike model-centric benchmarks, AlphaEval evaluates complete agent products -- Claude Code, Codex, etc. -- as commercial systems, capturing performance variations invisible to model-level evaluation. Our evaluation framework covers multiple paradigms (LLM-as-a-Judge, reference-driven metrics, formal verification, rubric-based assessment, automated UI testing, etc.), with individual domains composing multiple paradigms. Beyond the benchmark itself, we contribute a requirement-to-benchmark construction framework -- a systematic methodology that transforms authentic production requirements into executable evaluation tasks in minimal time. This framework standardizes the entire pipeline from requirement to evaluation, providing a reproducible, modular process that any organization can adopt to construct production-grounded benchmarks for their own domains.

cs.CL↗

arXiv2024

RemoteCLIP: A Vision Language Foundation Model for Remote Sensing

General-purpose foundation models have led to recent breakthroughs in artificial intelligence. In remote sensing, self-supervised learning (SSL) and Masked Image Modeling (MIM) have been adopted to build foundation models. However, these models primarily learn low-level features and require annotated data for fine-tuning. Moreover, they are inapplicable for retrieval and zero-shot applications due to the lack of language understanding. To address these limitations, we propose RemoteCLIP, the first vision-language foundation model for remote sensing that aims to learn robust visual features with rich semantics and aligned text embeddings for seamless downstream application. To address the scarcity of pre-training data, we leverage data scaling which converts heterogeneous annotations into a unified image-caption data format based on Box-to-Caption (B2C) and Mask-to-Box (M2B) conversion. By further incorporating UAV imagery, we produce a 12 $\times$ larger pretraining dataset than the combination of all available datasets. RemoteCLIP can be applied to a variety of downstream tasks, including zero-shot image classification, linear probing, $\textit{k}$-NN classification, few-shot classification, image-text retrieval, and object counting in remote sensing images. Evaluation on 16 datasets, including a newly introduced RemoteCount benchmark to test the object counting ability, shows that RemoteCLIP consistently outperforms baseline foundation models across different model scales. Impressively, RemoteCLIP beats the state-of-the-art method by 9.14% mean recall on the RSITMD dataset and 8.92% on the RSICD dataset. For zero-shot classification, our RemoteCLIP outperforms the CLIP baseline by up to 6.39% average accuracy on 12 downstream datasets. Project website: https://github.com/ChenDelong1999/RemoteCLIP

cs.CV↗

arXiv2024

Are your comments outdated? Towards automatically detecting code-comment consistency

In software development and maintenance, code comments can help developers understand source code, and improve communication among developers. However, developers sometimes neglect to update the corresponding comment when changing the code, resulting in outdated comments (i.e., inconsistent codes and comments). Outdated comments are dangerous and harmful and may mislead subsequent developers. More seriously, the outdated comments may lead to a fatal flaw sometime in the future. To automatically identify the outdated comments in source code, we proposed a learning-based method, called CoCC, to detect the consistency between code and comment. To efficiently identify outdated comments, we extract multiple features from both codes and comments before and after they change. Besides, we also consider the relation between code and comment in our model. Experiment results show that CoCC can effectively detect outdated comments with precision over 90%. In addition, we have identified the 15 most important factors that cause outdated comments, and verified the applicability of CoCC in different programming languages. We also used CoCC to find outdated comments in the latest commits of open source projects, which further proves the effectiveness of the proposed method.

cs.SE↗

arXiv2024

Few-shot Adaptation of Multi-modal Foundation Models: A Survey

Multi-modal (vision-language) models, such as CLIP, are replacing traditional supervised pre-training models (e.g., ImageNet-based pre-training) as the new generation of visual foundation models. These models with robust and aligned semantic representations learned from billions of internet image-text pairs and can be applied to various downstream tasks in a zero-shot manner. However, in some fine-grained domains like medical imaging and remote sensing, the performance of multi-modal foundation models often leaves much to be desired. Consequently, many researchers have begun to explore few-shot adaptation methods for these models, gradually deriving three main technical approaches: 1) prompt-based methods, 2) adapter-based methods, and 3) external knowledge-based methods. Nevertheless, this rapidly developing field has produced numerous results without a comprehensive survey to systematically organize the research progress. Therefore, in this survey, we introduce and analyze the research advancements in few-shot adaptation methods for multi-modal models, summarizing commonly used datasets and experimental setups, and comparing the results of different methods. In addition, due to the lack of reliable theoretical support for existing methods, we derive the few-shot adaptation generalization error bound for multi-modal models. The theorem reveals that the generalization error of multi-modal foundation models is constrained by three factors: domain gap, model capacity, and sample size. Based on this, we propose three possible solutions from the following aspects: 1) adaptive domain generalization, 2) adaptive model selection, and 3) adaptive knowledge utilization.

cs.CV↗

arXiv2022

Towards Exploring the Code Reuse from Stack Overflow during Software Development

As one of the most well-known programmer Q&A websites, Stack Overflow (i.e., SO) is serving tens of thousands of developers every day. Previous work has shown that many developers reuse the code snippets on SO when they find an answer (from SO) that functionally matches the programming problem they encounter in their development activities. To study how programmers reuse code on SO during project development, we conduct a comprehensive empirical study. First, to capture the development activities of programmers, we collect 342,148 modified code snippets in commits from 793 open-source Java projects, and these modified code can reflect the programming problems encountered during development. We also collect the code snippets from 1,355,617 posts on SO. Then, we employ CCFinder to detect the code clone between the modified code from commits and the code from SO, and further analyze the code reuse when programmer solves a programming problem during development. We count the code reuse ratios of the modified code snippets in the commits of each project in different years, the results show that the average code reuse ratio is 6.32%, and the maximum is 8.38%. The code reuse ratio in project commits has increased year by year, and the proportion of code reuse in the newly established project is higher than that of old projects. We also find that some projects reuse the code snippets from many years ago. Additionally, we find that experienced developers seem to be more likely to reuse the knowledge on SO. Moreover, we find that the code reuse ratio in bug-related commits (6.67%) is slightly higher than that of in non-bug-related commits (6.59%). Furthermore, we also find that the code reuse ratio (14.44%) in Java class files that have undergone multiple modifications is more than double the overall code reuse ratio (6.32%).

cs.SE↗