Lifelong Learning Accounts: Creating a Partnership in Lifelong Learning
Lifelong Learning Hw14 - speech.ee.ntu.edu.tw
Transcript of Lifelong Learning Hw14 - speech.ee.ntu.edu.tw
![Page 2: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/2.jpg)
Outline● Introduction● Dataset● Sample Code● COOL Quiz● Grading● Submission
![Page 3: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/3.jpg)
Introduction - LifeLong Learning Goal: A model can beat all task!
![Page 4: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/4.jpg)
Introduction - LifeLong LearningCondition: Model Sequentially Learn Different Task! (In Training Time)
![Page 5: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/5.jpg)
Introduction - LifeLong LearningKeyPoint: Avoid Catastrophic Forgetting
Catastrophic Forgetting
Avoid Catastrophic Forgetting
![Page 6: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/6.jpg)
Introduction
In Sample Code:
![Page 7: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/7.jpg)
DatasetPermuted MNIST
![Page 8: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/8.jpg)
Sample Code - Training Detail● 5 task / Each task has 10 epoches for training.● Each method cost ~20 minutes for training model.● CoLab Link (copy to your drive first! Don’t simply run colab.)
![Page 9: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/9.jpg)
Sample Code - Methods● Baseline● EWC● MAS● SI● RWalk● SCP
![Page 10: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/10.jpg)
Sample Code - Guideline● Utility● Visualization● Methods
![Page 11: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/11.jpg)
Sample Code● Utility
○ Permutation○ Dataloader and Training Argument○ Model○ train○ evaluate○ evaluate metric
● Visualization● Methods
![Page 12: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/12.jpg)
Sample Code● Utility
○ Permutation○ Dataloader and Training Argument○ Model○ train○ evaluate○ evaluate metric
● Visualization● Methods
5 PERMUTATION = 5 TASK
![Page 13: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/13.jpg)
Sample Code● Utility
○ Permutation○ Dataloader and Training Argument○ Model○ train○ evaluate○ evaluate metric
● Visualization● Methods
Fixed model size!
![Page 14: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/14.jpg)
Sample Code● Utility
○ Permutation○ Dataloader and Training Argument○ Model○ train○ evaluate○ evaluate metric
● Visualization● Methods
In k th task:
![Page 15: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/15.jpg)
Sample Code● Utility● Visualization● Methods
![Page 16: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/16.jpg)
Sample Code - Guideline● Utility● Visualization● Methods
○ Baseline○ EWC○ MAS○ SI○ RWalk○ SCP
![Page 17: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/17.jpg)
BaselineSequentially Train
Training Pipeline:
![Page 18: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/18.jpg)
Lifelong learning class
![Page 19: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/19.jpg)
Lifelong learning class
![Page 20: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/20.jpg)
EWC - Elastic Weight Consolidation1. You need to know how to generate Guardiance weight from EWC!2. Do this method need to use label ?3. Hint: (trace the class ewc and its calculate_importance function)
Paper Link: https://arxiv.org/pdf/1612.00796.pdf
![Page 21: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/21.jpg)
MAS - Memory Aware Synapse1. You need to know how to generate Guardiance weight from MAS!2. Do this method need to use label ?3. Hint: (trace the class mas and its calculate_importance function)
Paper Link: https://arxiv.org/abs/1711.09601
![Page 22: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/22.jpg)
SI - Synaptic Intelligence1. You need to know how to generate Guardiance weight from SI!2. Do this method need to use label ?3. Hint: (Accumulated loss change in each update step)
Paper Link: https://arxiv.org/abs/1703.04200, Talk Slide
![Page 23: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/23.jpg)
SI - Main Idea
Picture comes from: Talk Slide
Parameter importance on-line from learning trajectory!
![Page 24: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/24.jpg)
SI - Abstract
Picture comes from: Talk Slide
![Page 25: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/25.jpg)
SI - Method
Picture comes from: Talk Slide
![Page 26: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/26.jpg)
RWalk - Remanian Walk1. Trace class rwalk and its update function!2. Do this method need to use label ?3. Hint:(The code is similar to two method which mentioned in sample code)
Paper Link: https://arxiv.org/abs/1801.10112
![Page 27: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/27.jpg)
SCP - Sliced Cramer Preservation1.Paper Link: https://openreview.net/pdf?id=BJge3TNKwH
2.Do this method need to use label ?
![Page 28: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/28.jpg)
SCP - Main Idea● Propose Distributed-based Distance to prevent fast intransigence and
avoid overestimate the importance of parameter.
intransigence
Model do not want to learn new task, and it just keep old task performance
![Page 29: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/29.jpg)
SCP - Sliced Cramer Preservation (Hint)1.
Paper Link: https://openreview.net/pdf?id=BJge3TNKwH
![Page 30: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/30.jpg)
COOL Quiz● 25 multiple choice questions● Basic Concept & Dataset : 4 Questions● Sample Code: 15 Questions
○ EWC: 3 Questions○ MAS: 3 Questions○ SI: 3 Questions○ Remanian Walk: 3 Questions○ Sliced Cramer Preservation: 3 Questions
● Other Methods & scenario: 6 Questions○ ICaRL, LwF, GEM, DGR○ Three Scenario
![Page 31: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/31.jpg)
Grading● All Questions (0.4pt) ● You have to choose ALL the correct answers for each question● Warning: The answers in NTU Cool are not correct, NTU grading score
before deadline is not the final grading score. ● Please do not select the choice depends on the grading score.
![Page 32: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/32.jpg)
Submission● No late submission!
● We can only pick the last submission!
● Deadline: 2021/07/02 23:59
● Warning: The answers in NTU Cool are not correct, NTU grading score before deadline is not the final grading score.
● Please do not select the choice depends on the grading score.
![Page 33: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/33.jpg)
Links● CoLab Link● NTU Cool Multiple Choice Question● Warning: The answers in NTU Cool are not correct, NTU grading score
before deadline is not the real grading score. ● Please do not select the choice depends on the grading score.
![Page 34: Lifelong Learning Hw14 - speech.ee.ntu.edu.tw](https://reader031.fdocuments.net/reader031/viewer/2022012023/6169cf4411a7b741a34b9c10/html5/thumbnails/34.jpg)
If any questions, you can ask us via...● NTU COOL (recommended)
○ https://cool.ntu.edu.tw/courses/4793● Email
○ [email protected]○ The title must begin with “[HW14]”
● TA hours ○ Each Monday 19:00~21:00 線上
○ Each Friday 13:30~14:20 線上 ○ Each Friday During Class