Completed
Conventional Paradigm: Supervised Learning
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
Secure Self-supervised Learning: Challenges and Solutions
Automatically move to the next video in the Classroom when playback concludes
- 1 Intro
- 2 Conventional Paradigm: Supervised Learning
- 3 Key Challenge of Supervised Learning
- 4 Road Map
- 5 Background on Self-supervised Learning
- 6 Data Augmentation
- 7 Pre-training an Encoder - SimCLR [ICML'20]
- 8 Building a Downstream Classifier
- 9 Backdoor Attack
- 10 Key Idea of Our Attack
- 11 Quantifying Effectiveness Goal
- 12 Quantifying Condition I
- 13 Quantifying Utility Goal
- 14 Optimization Problem
- 15 Attack Setting
- 16 Attack Success Rate
- 17 Clean Accuracy and Backdoored Accuracy
- 18 Evaluation on Real-world Pre-trained Encoders
- 19 Existing Defenses are Insufficient
- 20 Summary
- 21 Motivation on Data Auditing
- 22 Auditing Unauthorized Data Use
- 23 Examples of Real-world Unauthorized Data Use
- 24 Our EncoderMI: Membership Inference based Data Auditing for Pre-trained Encoders
- 25 Revisiting Encoder Pre-training
- 26 Shadow Training Setup
- 27 Pre-training a Shadow Encoder
- 28 Constructing a Training Set for Inference Classifier
- 29 Building an Inference Classifier
- 30 Experimental Setup
- 31 Evaluation on CLIP
- 32 Conclusion
