Advice for Getting Models Work Mahdi Roozbahani Lecturer, - - PowerPoint PPT Presentation

▶
advice for getting models work
SMART_READER_LITE
LIVE PREVIEW

Advice for Getting Models Work Mahdi Roozbahani Lecturer, - - PowerPoint PPT Presentation

Mar 14, 2023 383 likes •672 views

CX4242: Advice for Getting Models Work Mahdi Roozbahani Lecturer, Computational Science and Engineering, Georgia Tech These slides are adopted from Polo, Andrew w. Moore, and Vivek Srikumar, and Chao Zhang Outline Model Diagnostics

slide-1
SLIDE 1

CX4242:

Advice for Getting Models Work

Mahdi Roozbahani Lecturer, Computational Science and Engineering, Georgia Tech

These slides are adopted from Polo, Andrew w. Moore, and Vivek Srikumar, and Chao Zhang

slide-2
SLIDE 2

Outline

  • Model Diagnostics
  • Error Analysis
  • Practical Advice

2

slide-3
SLIDE 3

Debugging Machine Learning

3

slide-4
SLIDE 4

Different Ways to Improve Your Model

4

slide-5
SLIDE 5

First Step: Diagnose Your Model

5

slide-6
SLIDE 6

Overfitting v.s. Underfitting

6

slide-7
SLIDE 7

Overfitting (High Variance)

7

slide-8
SLIDE 8

Underfitting (High Bias)

8

slide-9
SLIDE 9

Different Ways to Improve Your Model

9

slide-10
SLIDE 10

Diagnostics

10

slide-11
SLIDE 11

Have Your Model Converged?

11

slide-12
SLIDE 12

Have Your Model Converged?

12

slide-13
SLIDE 13

Have Your Model Converged?

13

slide-14
SLIDE 14

Different Ways to Improve Your Model

14

slide-15
SLIDE 15

Diagnostics

15

slide-16
SLIDE 16

What to Measure

16

slide-17
SLIDE 17

Outline

  • Model Diagnostics
  • Error Analysis
  • Practical Advice

17

slide-18
SLIDE 18

Machine Learning in Context

18

slide-19
SLIDE 19

Error Analysis

19

slide-20
SLIDE 20

Example: Text Processing System

20

slide-21
SLIDE 21

Example: Text Processing System

21

slide-22
SLIDE 22

Error Analysis

22

slide-23
SLIDE 23

Outline

  • Model Diagnostics
  • Error Analysis
  • Practical Advice

23

slide-24
SLIDE 24

Advice for ML Workflow in Practice

24

slide-25
SLIDE 25

Advice for ML Workflow in Practice

25

slide-26
SLIDE 26

What to Watch Out For?

26

slide-27
SLIDE 27

What to Watch Out For?

27

slide-28
SLIDE 28

What to Watch Out For?

28