Research map / Label-efficient and robust learning / Recognition and long-tail
A Stationary (and Therefore Compatible) Representation is All You Need
Niccolò Biondi, Federico Pernici, Simone Ricci, Alberto Del Bimbo
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 higher-order contrastive term to the d-Simplex cross-entropy (HOC vs FD) | d-Simplex fixed classifier · CIFAR100/10 (CL2R, 31 tasks) | ACA: 15.37 -> 23.44 | Improved |
| use d-Simplex-HOC instead of d-Simplex-FD | d-Simplex · CUB180/20 (CL2R, 31 tasks) | ACA: 21.97 -> 25.46 | Improved |
| use d-Simplex-HOC instead of d-Simplex-FD | d-Simplex · TinyImageNet200/20 (CL2R, 31 tasks) | ACA: 0.44 -> 25.14 | Improved |
| use d-Simplex-HOC instead of CVS | fine-tuning with replay · CIFAR100/10 (CL2R, 31 tasks) | ACA: 2.928 -> 23.44 | Improved |
| use d-Simplex-HOC instead of d-Simplex-FD under two model replacements | d-Simplex · CIFAR100R/10 (IAM-CL2R, 7 tasks, two model replacements) | AA: 56.581 -> 68.13 | Improved |
| use d-Simplex-HOC instead of d-Simplex-FD under two model replacements | d-Simplex · CIFAR100R/10 (IAM-CL2R, 31 tasks, two model replacements) | AA: 56.267 -> 67.40 | Improved |
| use d-Simplex-HOC instead of d-Simplex-FD with changing architectures | d-Simplex · CIFAR100R/10 (IAM-CL2R, 31 tasks, architecture change, AA at task 21) | AA: 56.88 -> 67.28 | Improved |
| set lambda to 0.1 rather than 0 (contrastive only) (ablation) | d-Simplex-HOC · CIFAR100/10 (CL2R, 16 tasks) | ACA: 40.706 -> 43.20 | Improved |
| set lambda to 0.1 rather than 1 (cross-entropy only) (ablation) | d-Simplex-HOC · CIFAR100/10 (CL2R, 16 tasks) | ACA: 0 -> 43.20 | Improved |
| raise rho from 0.1 to 5 (ablation) | d-Simplex-HOC · CIFAR100/10 (CL2R, 16 tasks) | ACA: 8.367 -> 43.20 | Improved |
| raise rho from 5 to 25 (ablation) | d-Simplex-HOC · CIFAR100/10 (CL2R, 16 tasks) | ACA: 43.20 -> 26.195 | Hurt |
| vary number of pre-allocated classes K from 100 to 10,000 (ablation) | d-Simplex-HOC · CIFAR100/10 (CL2R, 16 tasks) | ACA: 43.199 -> 45.46 | No clear effect |
Closest papers by meaning
- Stationary Representations: Optimally Approximating Compatibility and Implications for Improved Model Replacements (CVPR 2024 · similarity 0.93)
- Test-time Adaptation in Non-stationary Environments via Adaptive Representation Alignment (NeurIPS 2024 · similarity 0.85)
- Instance-Conditional Timescales of Decay for Non-Stationary Learning (AAAI 2024 · similarity 0.84)
- Provably Reliable Classifier Guidance via Cross-Entropy Control (NeurIPS 2026 · similarity 0.84)
- Guarantees for Nonlinear Representation Learning: Non-identical Covariates, Dependent Data, Fewer Samples (ICML 2024 · similarity 0.83)
- Learning Along the Arrow of Time: Hyperbolic Geometry for Backward-Compatible Representation Learning (ICML 2025 · 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.
