UCL x Deepmind Lecture Series

A brief review on the Deep Learning Lecture Series presented by DeepMind, at UCL.

On the time before the lecture starts, the Darwin building is free entry. People of different ages came one after another, crowded together, and chatting about AI. Anyone with tickets can enter the hall first, and the rest have to wait until the last.

In the lecture theater, two big cameras were setted on third row, which means lecture casts will be available in the near future.

Lecture 3: Convolutional Neural Networks

18:30-20:00 17 Feb, 2019

This lecture is rather introductory and the content is almost maths-free, but the audience still need some prior knowledge to understand fully.

Main contents

• Motivation and basic idea.

• Various convolution methods: strided, depthwise, pooling, etc.

• Introducing different kinds of CNNs by going through the ImageNet Challenge. The most impressive part is that he visualised models with sequences of colourful blocks:

Overall

★★★★☆
The oral presentation is not attractive, but the speaker illustrated how CNN works in a fresh way. This $1\frac{1}{2}$ hours lecture made me understand CNN better.