qiuqiangkong
Research Scientist
Summary
🤩
RockstarResearch Scientist with a decade of hands-on experience in machine learning and audio signal processing, combining academic rigor with practical engineering. Known for improving model training, evaluation pipelines, and visualization tools in prominent open-source audio projects such as music source separation and AudioSet tagging. Skilled in augmenting data pipelines, spectral feature extraction, and decision-level modeling to boost reproducibility and interpretability of audio ML experiments. Balances research depth with production-minded contributions—fixing checkpointing, setup scripts, and inference pathways to make models easier to train and analyze. Also holds an academic role as an Assistant Professor at CUHK, highlighting a blend of teaching, research, and open-source impact.
10 years of coding experience
Github Skills (14)
mask-rcnn10
faster-rcnn10
pytorch10
machine-learning10
eval10
audio-processing10
deep-learning10
trainings10
data-augmentation10
python10
evaluation10
modeling10
st9
tensorflow5
Programming languages (3)
Github contributions (5)
qiuqiangkong/audioset_tagging_cnn
Jun 2019 - Jul 2021
Role in this project:
ML Engineer
ML EngineerContributions:34 commits, 1 PR, 72 pushes in 2 years 1 month
Contributions summary:qiuqiangkong primarily worked on updating and improving the model training and evaluation scripts for an audio tagging CNN. They modified `utils/plot_statistics.py` to plot model performance, specifically focusing on comparing different CNN architectures (e.g., Cnn13, ResNet, DenseNet) and training configurations. Furthermore, the user updated `pytorch/models.py`, `pytorch/stft.py`, `pytorch/inference.py`, `pytorch/main.py`, and `pytorch/finetune_template.py` with various changes including decision level implementations and adjustments to the spectral features extraction.
bytedance/music_source_separation
Sep 2021 - Oct 2021
Role in this project:ML Engineer
ML EngineerContributions:1 review, 17 commits, 2 PRs in 1 month
Contributions summary:qiuqiangkong primarily focused on enhancing the model training and evaluation pipelines for music source separation. Their contributions include modifications to data augmentation techniques, path fixes for model checkpoints, and the addition of new plotting functionalities to analyze model performance. The user also updated the project's setup configuration and download scripts, and added more plotting options to visualize model performance.
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.