Tutorial Reading Group 2024 #3

The third tutorial reading group 2024 will be held on 16th May 2024, Thursday 13:00-16:00 online via Teams. Mr. Xiangyu Yin will give us a talk on “Deep Model Complexity and Its Applications in Adversarial Learning”.

Teams Link

Meeting ID: 357 738 191 673

Passcode: fkpyUW

Reference:

• [1] Deep Double Descent: Where Bigger Models and More Data Hurt.

• [2] Model Complexity of Deep Learning: A Survey.

• [3] The Curious Case of Adversarial Robust Models: More Data Can Help, Double Descend, or Hurt Generalization.

• [4] Reconciling modern machine learning practice and the bias-variance trade-off.

• [5] Rademacher Complexity for Adversarial Robust Generalization.

Related Codes:

• [1] https://github.com/dongyin92/adversarially-robust-generalization

• [2] https://github.com/lightonai/double-descent-curve

• [3] https://github.com/vasusingla/low_curvature_activations

Slides:

• Slides is available!

Potential Interesting Topics:

• [1] The lower bound of expressive capacity that is sufficient for a given task.

• [2] Bottleneck of expressive capacity

• [3] Measuring Effective Model Complexity across different models

Contact:

If you have any ideas or questions related to this tutorial/presentation, you are most welcome to contact Mr. Xiangyu Yin (X.Yin22@liverpool.ac.uk) for further discussion and collaboration.