Research map / Label-efficient and robust learning / Self-supervised and contrastive
IConE: Batch Independent Collapse Prevention for Self-Supervised Representation Learning
Konstantinos Almpanakis, Anna Kreshuk
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 diversity loss L_div on instance embedding table (ablation) | IConE · Synthetic 2D GMM | kNN accuracy (%): 57 -> 88 | Improved |
| add view-to-instance alignment loss L_vi (ablation) | IConE · Synthetic 2D GMM | kNN accuracy (%): 39 -> 88 | Improved |
| add view-view invariance loss L_vv (ablation) | IConE · Synthetic 2D GMM | kNN accuracy (%): 72 -> 88 | Improved |
| replace orthogonality loss with SIGReg (ablation) | IConE · 2D MedMNIST, B=1 | top-1 accuracy (%): 59.6 -> 57.3 | Hurt |
| replace orthogonality loss with VCReg (ablation) | IConE · 3D MedMNIST, B=32 | top-1 accuracy (%): 65.4 -> 67.9 | Mixed |
| increase number of views from 2 to 12 (ablation) | IConE · 2D MedMNIST average | balanced accuracy (%): 75.1 -> 74.8 | No clear effect |
| increase embedding init scale from 0.02 to 0.20 (ablation) | IConE · 2D MedMNIST, B=1 | balanced accuracy (%): 74.3 -> 70.9 | No clear effect |
| use IConE instead of SimCLR/VICReg-style batch-dependent SSL | 3D linear/kNN · 3D MedMNIST k-NN-1, B=4 | k-NN balanced accuracy (%): 55.8 -> 63.6 | Improved |
| use IConE-Class instead of Cross-Entropy | supervised learning · 2D MedMNIST, B=4 | top-1 balanced accuracy (%): 78.17 -> 82.42 | Improved |
| use IConE-Instance instead of SupCon | supervised learning · 3D MedMNIST, B=1 | top-1 balanced accuracy (%): 59.49 -> 68.77 | Improved |
| use IConE-VCReg or SIGReg instead of default | IConE · peak GPU memory N=100k, B=4 | GPU memory (GB): 637.2 -> 1.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VICReg — VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning (ICLR 2022) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Exploring Structural Degradation in Dense Representations for Self-supervised Learning (NeurIPS 2025 · similarity 0.85)
- Preventing Dimensional Collapse in Self-Supervised Learning via Orthogonality Regularization (NeurIPS 2024 · similarity 0.85)
- Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning (NeurIPS 2025 · similarity 0.85)
- Drive vs. Decay: On the Training Dynamics of Joint-Embedding Predictive Architectures (NeurIPS 2026 · similarity 0.85)
- Symmetric Neural-Collapse Representations with Supervised Contrastive Loss: The Impact of ReLU and Batching (ICLR 2024 · similarity 0.84)
- MoME: Mixture-of-Memory Embeddings for Context-Aware Sparse Lookup (NeurIPS 2026 · similarity 0.84)
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.
