Research map / Label-efficient and robust learning / Self-supervised and contrastive
Equivariance and Invariance Inductive Bias for Learning from Insufficient Data
Tan Wang, Qianru Sun, Sugiri Pranata, Karlekar Jayashree, Hanwang Zhang
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 EqInv (class-wise IRM) on top of MoCo-v2 | MoCo-v2 · VIPriors-50 test | accuracy: 46.98 -> 52.09 | Improved |
| add EqInv on top of MoCo-v2 | MoCo-v2 · VIPriors-10 test | accuracy: 16.97 -> 25.20 | Improved |
| add EqInv on top of SimSiam | SimSiam · VIPriors-10 test | accuracy: 6.68 -> 23.33 | Improved |
| add EqInv on top of IP-IRM | IP-IRM · NICO test | accuracy: 60.26 -> 64.14 | Improved |
| add EqInv on top of IP-IRM | IP-IRM · VIPriors-50 test | accuracy: 48.90 -> 52.27 | Improved |
| add EqInv on top of SimSiam | SimSiam · NICO test | accuracy: 42.68 -> 44.77 | Improved |
| add EqInv on top of MoCo-v2 | MoCo-v2 · ImageNet100 val | accuracy: 86.30 -> 88.38 | Improved |
| add SSL equivariance pretraining (Step 1) (ablation) | EqInv · VIPriors-20 test | accuracy: 12.41 -> 36.34 | Improved |
| add similarity adjustment (Step 2) to environment construction (ablation) | EqInv · VIPriors-20 test | accuracy: 36.34 -> 37.78 | Improved |
| add class-wise IRM (Step 3) after SSL (ablation) | EqInv · VIPriors-20 test | accuracy: 34.88 -> 36.34 | Improved |
| add class-wise IRM without SSL (ablation) | ResNet-50 from scratch · VIPriors-20 test | accuracy: 12.39 -> 12.41 | No clear effect |
| update feature backbone during IRM penalty with lambda above 10 (ablation) | EqInv · VIPriors-20 | accuracy: down to random guess (about 1%) | Hurt |
Closest papers by meaning
- Understanding the Role of Equivariance in Self-supervised Learning (NeurIPS 2024 · similarity 0.88)
- Relaxing Equivariance Constraints with Non-stationary Continuous Filters (NeurIPS 2022 · similarity 0.86)
- Balanced Data, Imbalanced Spectra: Unveiling Class Disparities with Spectral Imbalance (ICML 2024 · similarity 0.84)
- Learning (Approximately) Equivariant Networks via Constrained Optimization (NeurIPS 2025 · similarity 0.84)
- The Implicit Bias of Heterogeneity towards Invariance: A Study of Multi-Environment Matrix Sensing (NeurIPS 2024 · similarity 0.84)
- Learning Conditional Invariances through Non-Commutativity (ICLR 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.
