Research map / Label-efficient and robust learning / Self-supervised and contrastive
The Challenges of Continuous Self-Supervised Learning
Senthil Purushwalkam, Pedro Morgado, Abhinav Gupta
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add a replay buffer (K=10, 64K) to SimSiam under single-pass streaming | SimSiam · Flickr 20M stream (single pass) | ImageNet top-1: 32.3 -> 41.8 | Improved |
| add a replay buffer (K=40, 16K) on a 5M stream | SimSiam · Flickr 5M stream (single pass) | ImageNet top-1: 14.5 -> 39.9 | Improved |
| grow replay buffer from 16K to 256K at hyper-sampling K=200 (ablation) | Buffered SSL · Flickr 1M stream (single pass) | ImageNet top-1: 30.5 -> 38.8 | Improved |
| grow replay buffer from 16K to 256K at K=10 (ablation) | Buffered SSL · Flickr 20M stream (single pass) | ImageNet top-1: 41.4 -> 41.5 | No clear effect |
| replace FIFO buffer with MinRed buffer (64K) (ablation) | Buffered SSL · Kinetics stream (Nseq=64) | ImageNet top-1: 11.7 -> 31.2 | Improved |
| replace FIFO buffer with MinRed buffer (64K) (ablation) | Buffered SSL · KrishnaCAM stream | ImageNet top-1: 1.7 -> 17.9 | Improved |
| replace FIFO buffer with MinRed buffer (16K) (ablation) | Buffered SSL · KrishnaCAM stream | ImageNet top-1: 0.5 -> 15.2 | Improved |
| replace FIFO buffer with MinRed buffer (64K) (ablation) | Buffered SSL · Kinetics stream (Nseq=16) | ImageNet top-1: 25.9 -> 26.2 | No clear effect |
| replace FIFO buffer with MinRed buffer (64K) (ablation) | Buffered SSL · Kinetics stream (Nseq=16) | iNaturalist top-1: 8.4 -> 7.9 | Hurt |
| add a replay buffer to SimSiam on correlated Kinetics stream | SimSiam · Kinetics stream (Nseq=64) | ImageNet top-1: 7.6 -> 11.7 | Improved |
| add a replay buffer to SimSiam on KrishnaCAM stream | SimSiam · KrishnaCAM stream | ImageNet top-1: 0.4 -> 1.7 | No clear effect |
| add MinRed buffer to SimSiam on KrishnaCAM stream (64K) | SimSiam · KrishnaCAM stream | ImageNet top-1: 0.4 -> 17.9 | Improved |
Closest papers by meaning
- Self-Supervised Models Are Continual Learners (CVPR 2022 · similarity 0.90)
- Continual Learning on Noisy Data Streams via Self-Purified Replay (ICCV 2021 · similarity 0.89)
- How Well Does Self-Supervised Pre-Training Perform with Streaming Data? (ICLR 2022 · similarity 0.89)
- Summarizing Stream Data for Memory-Constrained Online Continual Learning (AAAI 2024 · similarity 0.86)
- Learning from Memory: Non-Parametric Memory Augmented Self-Supervised Learning of Visual Features (ICML 2024 · similarity 0.86)
- Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
