This course aims to provide you a theoretical framework to understand deep learning algorithms. In particular, we focus on three aspects of deep learning.
(1) Approximation: How accurately can we approximate the ground truth function with a function space parameterized by some neural network architecture?
(2) Generalization: How well can we expect the model trained on the training data to perform on the test data?
(3) Optimization: How closely can we approximate the optimal function inside the neural network function space?
Recent breakthroughs suggest that, it is necessary to understand how these components interplay with each other to truly understand how deep learning works. Hope we all could arrive at that level of understanding.

Elementary probability and linear algebra.
Recommended: Real Analysis.

To be announced (will be modified based on co-teaching plans with Yonsei university).

Tentatively:
Homework: 80%
Final report: 20%

I would closely follow the notes of Matus Telgarsky: https://mjt.cs.illinois.edu/dlt/index.pdf
Recommended: "Understanding Machine Learning: From Theory to Algorithms" by Shalev-Shwartz and Ben-David.

To be announced.

Phase 1 (week 1~10): Lectures on basic materials.
Phase 2 (week 11~16): Lectures on recent papers, with student participations.