Research map / Video / Video segmentation and tracking
Federated Self-Supervised Learning for Video Understanding
Yasar Abbas Ur Rehman, Yan Gao, Jiajun Shen, Pedro Porto Buarque de Gusmão, Nicholas Lane
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 |
|---|---|---|---|
| pretrain Speed with federated learning instead of centralized | Speed · UCF-101 clip retrieval | R@1: 16.84 -> 21.97 | Improved |
| pretrain CtP with federated learning instead of centralized | CtP · UCF-101 action recognition fine-tune | top-1 accuracy: 86.20 -> 81.95 | Hurt |
| pretrain CtP with federated learning instead of centralized | CtP · HMDB-51 clip retrieval | R@1: 11.80 -> 13.66 | Improved |
| use Non-IID (8 classes per client) partition instead of IID | CtP federated · UCF-101 fine-tune | top-1 accuracy: 81.92 -> 81.95 | No clear effect |
| use Non-IID (4 classes per client) partition instead of IID | CtP federated · UCF-101 fine-tune | top-1 accuracy: 81.92 -> 81.15 | No clear effect |
| use FedU instead of FedAvg aggregation | CtP federated · UCF-101 clip retrieval | R@1: 32.62 -> 34.07 | Improved |
| use loss-based instead of FedAvg aggregation | CtP federated · HMDB-51 clip retrieval | R@1: 16.54 -> 14.44 | Hurt |
| aggregate only backbone weights (FedVSSL alpha=0, beta=0) instead of full model (ablation) | FedAvg · UCF-101 clip retrieval | R@1: 32.62 -> 34.34 | Improved |
| add stochastic weight averaging (beta=1) to loss-weighted aggregation alpha=0 (ablation) | FedVSSL · UCF-101 clip retrieval | R@1: 34.34 -> 35.61 | Improved |
| mix loss and FedAvg aggregation with alpha=0.9 (beta=0) instead of alpha=0 (ablation) | FedVSSL · UCF-101 linear probe | top-1 accuracy: 47.95 -> 50.36 | Improved |
| add alpha=0.9, beta=0 FedVSSL over FedAvg baseline on UCF fine-tuning | FedAvg · UCF-101 fine-tune | top-1 accuracy: 79.91 -> 80.62 | Improved |
| add SWA (beta=1) to alpha=0.9 FedVSSL (ablation) | FedVSSL · UCF-101 fine-tune | top-1 accuracy: 80.62 -> 79.41 | Hurt |
Closest papers by meaning
- Divergence-aware Federated Self-Supervised Learning (ICLR 2022 · similarity 0.89)
- FedSC: Provable Federated Self-supervised Learning with Spectral Contrastive Objective over Non-i.i.d. Data (ICML 2024 · similarity 0.85)
- Fed-QSSL: A Framework for Personalized Federated Learning under Bitwidth and Data Heterogeneity (AAAI 2024 · similarity 0.84)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.84)
- FedMSGL: A Self-Expressive Hypergraph Based Federated Multi-View Learning (AAAI 2025 · similarity 0.83)
- Towards Diverse Device Heterogeneous Federated Learning via Task Arithmetic Knowledge Integration (NeurIPS 2024 · similarity 0.83)
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.
