Improving Developer Code Understanding with GitHub Issues and Retrieval-Augmented Generation
Built a RAG-based doc tool with GitHub issues; raised README clarity by 116% and code clarity by 16%.
🎧Benjamin Chou
3rd-year ECE PhD Candidate at Purdue University specializing in AI/ML research for audio, music, and multimodal systems
Professional Summary
PhD candidate at Purdue University researching AI/ML for audio, music, and multimodal systems. I build generative and transformer-based models for error detection and deepfake detection.
Education
Ph.D. in Electrical & Computer Engineering
2023-01-01
Purdue University
B.Sc. in Electrical Engineering
2020-01-01
2023-12-31
National Cheng Kung University
Interests
Audio and music machine learning
Multimodal systems
Generative models
Transformers
Deepfake detection
Research Focus
My work centers on audio and music intelligence, multimodal modeling, and
reliable detection of synthetic and manipulated media. I combine Audio DSP,
deep learning, and high-performance computing to build systems that scale
from real-world datasets to production settings.
Publications
(2026).
AdaPerceiver: Transformers with Adaptive Width, Depth, and Tokens.
Under Submission.
(2026).
ICLAD: In-Context Learning with Comparison-Guidance for Audio Deepfake Detection.
Under Submission.
(2026).
Inference-Time Alignment of Diffusion Models with Evolutionary Algorithms.
Under Submission.
(2026).
LadderSym: A Multimodal Interleaved Transformer for Music Practice Error Detection.
Under Submission.
(2025).
Detecting Performance Errors with Transformers.
Accepted to AAAI 2025.
Selected Projects
Multi-Agent Self-Play for Beating Atari Games
Used DQN/PPO self-play to improve Atari win rates by 11–32%.
Spectral Image Inpainting with Deep Learning
Recovered 98% corrupted hyperspectral data at 0.02 RMSE using convex optimization and deep learning.
Contact