Deep Learning — Advanced Course Notebooks

About

This repository contains my notebooks for an advanced Deep Learning course, organized into five themed sets. The sets progress from foundations and classic vision tasks to sequence modeling/LLMs, generative modeling, and self-supervised/text–image methods.
Each set has its own README inside the set folder (Set1.md, Set2.md, …) with details, instructions, and notebook notes.

---

Repository Structure

• Set 1 — Foundations
  Basics of tensors & autograd, a NumPy neural network from scratch (manual forward/backward + gradient checking), and an optimization playground comparing first/second-order methods on standard test functions.
  Details: Set 1/Set1.md

• Set 2 — Vision Tasks
  Three practical pipelines: image classification, semantic segmentation, and object detection—including data loading, training loops, metrics (e.g., accuracy, mIoU, mAP), and qualitative visualizations.
  Details: Set 2/Set2.md

• Set 3 — Sequence Modeling & LLMs
  RNN/LSTM/GRU sequence models, a GPT-2 causal LM (training + sampling), PEFT/LoRA for parameter-efficient fine-tuning, and reasoning at inference time (CoT, self-consistency, best-of-n).
  Details: Set 3/Set3.md

• Set 4 — Generative Models
  Variational Autoencoders (VAE) with ELBO/β-VAE options and diagnostics; Diffusion (DDPM) with noise schedules, UNet denoiser, DDPM/DDIM sampling, and optional classifier-free guidance.
  Details: Set 4/Set4.md

• Set 5 — Self-Supervision & Text–Image Generation
  DINO self-distillation (EMA teacher, multi-crop), CLIP-guided image optimization from prompts, and Stable Diffusion (text2img/img2img/inpainting) with practical controls for quality/speed.
  Details: Set 5/Set5.md

🔎 Each SetX.md explains the notebooks inside that set (what’s implemented, how to run, metrics, and key takeaways).