Μαθήματα

This crash-course on neural network compression provides a comprehensive overview of essential techniques for optimizing deep learning models. We will begin with a short introduction to the operation and training of (convolutional) neural networks. We will then delve into neural network quantization, explaining how reducing the precision of weights and activations can decrease model size and computational requirements. We will also explore pruning, highlighting methods for removing redundant neurons and connections to streamline the network without significantly impacting task performance. Finally, we will introduce the concept of knowledge distillation, demonstrating how knowledge can be transferred from a large, complex model (teacher) to a smaller, simpler model (student) to achieve comparable performance with reduced computational resources. All topics will be accompanied by short in-class programming assignments. This crash-course equips participants with the knowledge and hands-on experience to effectively compress neural networks, making them more suitable for deployment in resource-constrained embedded devices.

You can contact the lecturer at a.k.balatsoukas.stimming@tue.nl or through the course forum.

Our main topics will be:

- Chapter 2 of Shiryaev, Probability 1.

- Chapters 1-3 of Wong and Hajek.

The other material is supplementary and ranges from

- under-graduate in Real Analysis (Abbot, Tao) 

to

- post-graduate in Probability and/or Stochastic Processes (Doob, Billingsley), Real Analysis (Royden, Folland).

I also include two books on High Dimensional Probability.

Siegrist' s book is a very good introduction to serious Probability.

The seminar will be, mostly, a reading short course. I will "read" extensive parts of the books and try to explain the material and motivate the audience towards deeper study.

It will take place at 145Π58, daily, from next Monday (10:00-13:00) until ...

APL

Το μάθημα δεν διαθέτει περιγραφή

Το μάθημα δεν διαθέτει περιγραφή

Το μάθημα δεν διαθέτει περιγραφή