Deep Learning Coursework — Portfolio Highlights

Projects

Audio Scene Classification (CNN)

Environmental sound classification using mel-spectrogram and MFCC features, with a custom CNN architecture and training pipeline. Result: 90.00% server accuracy (621,807 params, within budget) → See lab6-audio-classification/

CIFAR-10 Image Classification (CNN)

Image classification on CIFAR-10 using a custom convolutional neural network designed under a strict parameter budget. Result: 82.84% accuracy on hidden server evaluation (~3.2M params) → See lab5-cifar10/

---

Notes

• Further details and results in each folder's README.
• This repository highlights the components I implemented without the coursework scaffolding.