Self-supervised Learning for Reading Activity Classification
summary
Claims a self-supervised pipeline—self-supervised pretraining and fine-tuning—that uses large unlabeled sensor recordings and small labeled sets to classify reading activity and answer-confidence. Implements training on electrooculography (EOG) glasses plus accelerometer/gyro for a four-class reading detection task and on eye-tracker gaze for a two-class confidence task. Results show the SSL model consistently outperforms fully-supervised deep learning and SVM baselines, with the advantage growing as labeled samples per class shrink.
AI-written summary · about
chapters (8)
- 0:00Intro
- 0:18Reading Activity Classification
- 1:19Motivation
- 2:32State of the Art
- 3:15Proposed SSL Method
- 4:03Experiment Reading Detection
- 4:22Experiment Confidence Estimation
- 5:24Conclusion and Future Work
related talks
53mSelf-Supervised Learning in Transformer Models for Birdsong
17mLearning Cooperative Personalized Policies from Gaze Data
22mA Classification Model for Sensing Human Trust in Machines Using EEG and GSR
39mDeep Learning vs. Conventional Machine Learning
47mHow I Helped My Daughter Learn to Read with Machine Learning
41mLarge Scale Representation Learning In-the-wild
45mActive learning and weak supervision in NLP projects
28mDeep Learning-Based Human Activity Recognition for Physical Skills...