ES 667 · Deep Learning

IIT Gandhinagar · Semester I, 2026–27 · Prof. Nipun Batra

This site contains the slides, notebooks, assignments, and interactive material for ES 667. The dated class record, grading, and announcements are on the 2026 course site.

Current class

Generalization & Regularization

• ▤ Slides · ▶ Presentation mode
• ★ Two-page cheatsheet
• ⌨ Learning curves, overfitting & early stopping
• ⌨ Dropout and regularization mechanisms

Attention and language

• Part 1: From characters to next-token prediction. Interactive lecture · Presentation · Two-page A4 cheat sheet
• Part 2: Self-attention, from first principles. Interactive lecture · Presentation · Two-page A4 cheat sheet

Convolutional neural networks

The complete ML-based CNN lecture, expanded with interactive calculations and an end-to-end MNIST walkthrough. Read the lecture · Present

Object detection and beyond boxes

• Part I: Object Detection. HTML lecture · Present · PDF · Detector lab
• Part II: Beyond Boxes. Segmentation, pose, depth and promptable vision. HTML lecture · Present · PDF · Beyond Boxes lab

Recently taught

• Valid Outputs & Learning-Rate Schedules — constrained parameters, warmup, decay, and cosine schedules. ▤ Valid outputs · ▤ LR schedules · ★ Cheatsheet
• Vector, Matrix & Affine Autograd — shape-safe reverse mode through dense layers and batches. ▤ Calculus · ▤ Autograd · ★ Cheatsheet · ⌨ Colab
• Backpropagation & Autograd from Scratch — computation graphs, reverse mode, and a scalar engine. ▤ Slides · ★ Cheatsheet · ⌨ Colab
• Optimization for Deep Learning — gradient estimates through momentum, RMSProp, and Adam. ▤ Slides · ★ Optimization I · ★ Optimization II

Course materials

• Lecture decks
• Notebooks
• Assignments
• Course guide
• References

Questions and announcements are handled in the course Slack workspace.