Part IV: Deep Learning for Sensor Time Series
Chapter 17  [A][R]

Self-Supervised and Contrastive Sensor Learning

Placeholder chapter page. Sections are produced by the book-skills pipeline.

Sections

  1. 17.1 Why labels are scarce in sensing
  2. 17.2 Contrastive learning with sensor augmentations
  3. 17.3 Masked reconstruction (MAE for time series)
  4. 17.4 Temporal predictive coding and TS2Vec
  5. 17.5 Relative and relational objectives (RelCon)
  6. 17.6 Cross-device, cross-user, cross-modal pretraining
  7. 17.7 Probing and evaluating learned representations

Lab 17

pretrain a contrastive/masked encoder on unlabeled sensor windows; fine-tune with few labels and measure the label-efficiency curve.