Episode 13: Jonathan Frankle, MIT, on the lottery ticket hypothesis and the science of deep learning

1

Introduction

00:00 • 2min

2

Machine Learning - What's the Story of Your Research Interests?

02:04 • 4min

3

Did You Have Research Interests?

06:31 • 3min

4

Can You Be a Great Researcher Without Being the World's Greatest Mathematician?

09:28 • 3min

5

Why You Can't Train a Prude Network?

12:24 • 4min

6

Is There a Way to Start Small?

16:47 • 3min

7

How Do You Preserve Your Performance With Network Pruning?

19:25 • 4min

8

Is Early in Training Really Early?

23:38 • 4min

9

Prue Networks of Initialization - Is This a Good Technique?

28:00 • 3min

10

Deep Learning - What Are Some of the Most Important Open Questions?

30:55 • 3min

11

Machine Learning

33:38 • 2min

12

Those Roses After You Found an Idea That Worked

36:07 • 5min

13

Deep Learning

40:56 • 2min

14

I'm Not Trying to Be the First Person to Have the Next Advance

43:01 • 2min

15

How Do You Find More Papers Like This?

44:43 • 3min

16

I Think We Need More Than That in the Field of Science

47:47 • 4min

17

Manifesto for Papers

51:34 • 2min

18

The Original Lottery Ticket Paper

53:10 • 2min

19

Grat Lt, Bu Wa O Yotot?

54:43 • 4min

20

Should We Use Facial Recognition?

58:48 • 2min

21

Why Do We Write Assembles?

01:00:22 • 3min

22

I'm a Scientist, and I Build on What I See in Front of Me

01:02:53 • 2min

23

Object Oriented Programming

01:04:29 • 2min

24

The Hardest Thing for a Junior Researcher

01:06:48 • 2min

25

Publication Doesn't Have to Be Significant

01:08:42 • 2min

26

The Story Behind the Bachnorm Paper

01:10:28 • 2min

27

Why Is Sparsity Everywhere?

01:12:20 • 2min

28

I'm Not a Lawyer, but I Want to Know.

01:13:56 • 2min

29

Is There Any Scale Set for Machine Learning?

01:15:30 • 2min

30

How to Train 700 Networks on Image Net?

01:17:23 • 4min