feat(segment): add segment-3 submission

[Utkarsh736]

Segment 3: Dataset Examples Submission

Summary

This submission implements dataset-based validation for activation maximization results by identifying the top 10 natural images that maximally activate neurons 0-9 in InceptionV1's mixed4a layer.

Dataset Used

ImageNette validation set (3,925 images)

• Lightweight alternative to full ImageNet for rapid iteration
• Contains 10 ImageNet classes: tench, English springer, cassette player, chain saw, church, French horn, garbage truck, gas pump, golf ball, parachute

Notes for Review

1. Dataset choice: This implementation uses ImageNette for computational efficiency..

2. Reproducibility:

   • Download ImageNette: wget https://s3.amazonaws.com/fast-ai-imageclas/imagenette2-320.tgz
   • Extract to ./imagenette2-320/
   • Run notebook: All cells execute sequentially without manual intervention

3. Output structure: Results saved to segment_3_outputs/ (not included in PR per .gitignore)

4. Dependencies: PyTorch, Lucent, torchvision, pandas, matplotlib