arXiv subjects

Gefei Zhang

Publications and source records attributed to Gefei Zhang.

5 recordsLinked to original sources

arXiv2026

ClassAid: A Real-time Instructor-AI-Student Orchestration System for Classroom Programming Activities

Generative AI is reshaping education, but it also raises concerns about instability and overreliance. In programming classrooms, we aim to leverage its feedback capabilities while reinforcing the educator's role in guiding student-AI interactions. We developed ClassAid, a real-time orchestration system that integrates TA Agents to provide personalized support and an AI-driven dashboard that visualizes student-AI interactions, enabling instructors to dynamically adjust TA Agent modes. Instructors can configure the Agent to provide technical feedback (direct coding solutions), heuristic feedback (hint-based guidance), automatic feedback (autonomously selecting technical or heuristic support), or silent operation (no AI support). We evaluated ClassAid through three aspects: (1) the TA Agents' performance, (2) feedback from 54 students and one instructor during a classroom deployment, and (3) interviews with eight educators. Results demonstrate that dynamic instructor control over AI supports effective real-time personalized feedback and provides design implications for integrating AI into authentic educational settings.

cs.HC↗

arXiv2025

ConflictLens: LLM-Based Conflict Resolution Training in Romantic Relationship

Our poster presents ConflictLens, a three-stage simulation system powered by large language models (LLMs) and grounded in psychological theory, designed to help users reflect on and practice conflict resolution in romantic relationships. Users can upload real conflict scenarios to receive evaluation of behavioral patterns, reflect on conflicts by annotating their negative behaviors, and practice different conflict resolution strategies in AI-simulated duologues. Initial evaluation by three domain experts suggests that ConflictLens offers a realistic experience and effectively supports self-guided reflection and communication practice in romantic relationships.

cs.HC↗

arXiv2025

Towards Extreme Pruning of LLMs with Plug-and-Play Mixed Sparsity

N:M structured pruning is essential for large language models (LLMs) because it can remove less important network weights and reduce the memory and computation requirements. Existing pruning methods mainly focus on designing metrics to measure the importance of network components to guide pruning. Apart from the impact of these metrics, we observe that different layers have different sensitivities over the network performance. Thus, we propose an efficient method based on the trace of Fisher Information Matrix (FIM) to quantitatively measure and verify the different sensitivities across layers. Based on this, we propose Mixed Sparsity Pruning (MSP) which uses a pruning-oriented evolutionary algorithm (EA) to determine the optimal sparsity levels for different layers. To guarantee fast convergence and achieve promising performance, we utilize efficient FIM-inspired layer-wise sensitivity to initialize the population of EA. In addition, our MSP can work as a plug-and-play module, ready to be integrated into existing pruning methods. Extensive experiments on LLaMA and LLaMA-2 on language modeling and zero-shot tasks demonstrate our superior performance. In particular, in extreme pruning ratio (e.g. 75%), our method significantly outperforms existing methods in terms of perplexity (PPL) by orders of magnitude (Figure 1).

cs.CL↗

arXiv2025

CPVis: Evidence-based Multimodal Learning Analytics for Evaluation in Collaborative Programming

As programming education becomes more widespread, many college students from non-computer science backgrounds begin learning programming. Collaborative programming emerges as an effective method for instructors to support novice students in developing coding and teamwork abilities. However, due to limited class time and attention, instructors face challenges in monitoring and evaluating the progress and performance of groups or individuals. To address this issue, we collect multimodal data from real-world settings and develop CPVis, an interactive visual analytics system designed to assess student collaboration dynamically. Specifically, CPVis enables instructors to evaluate both group and individual performance efficiently. CPVis employs a novel flower-based visual encoding to represent performance and provides time-based views to capture the evolution of collaborative behaviors. A within-subject experiment (N=22), comparing CPVis with two baseline systems, reveals that users gain more insights, find the visualization more intuitive, and report increased confidence in their assessments of collaboration.

cs.HC↗

arXiv2020

Specifying and Model Checking Workflows of Single Page Applications with TLA+

Single Page Applications (SPAs) are different than hypertext-based web applications in that their workflow is not defined by explicit links, but rather implicitly by changes of their widgets' states. The workflow may hence be hard to track. We present an approach to specifying and model checking SPAs with TLA+. Our approach makes it easier to document and to track the workflow of SPAs and to find potential design flaws.

cs.SE↗