Research map / Label-efficient and robust learning / Self-supervised and contrastive
Beyond Learning Features: Training a Fully-Functional Classifier with ZERO Instance-Level Labels
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 self-supervised feature learning and feature clustering (ablation) | ZERO-label classifier · CIFAR-10 | accuracy: 9.9 ± 0.5 -> 42.1 ± 5.5 | Helped |
| add feature clustering and cluster-guided distribution matching (ablation) | ZERO-label classifier · CIFAR-10 | accuracy: 9.8 ± 0.2 -> 42.1 ± 5.5 | Helped |
| initialize the classifier from feature-clustering centers rather than randomly (ablation) | ZERO-label classifier · CIFAR-10 | accuracy: 39.4 ± 14.3 -> 42.1 ± 5.5 | Helped |
| add self-supervised feature learning and feature clustering (ablation) | ZERO-label classifier · CIFAR-10 | accuracy: 10.5 ± 1.9 -> 42.1 ± 5.5 | Helped |
| train the classifier using the complete ZERO-label pipeline rather than a random classifier layer | ZERO-label classifier · CIFAR-10 | accuracy: 11.2 -> 42.1 ± 5.5 | Helped |
| train the classifier using the complete ZERO-label pipeline rather than a random classifier layer | ZERO-label classifier · STL-10 | accuracy: 12.8 -> 36.5 ± 3.3 | Helped |
| train the classifier using the complete ZERO-label pipeline rather than a random classifier layer | ZERO-label classifier · CIFAR100-20 | accuracy: 4.5 -> 13.6 ± 3.2 | Helped |
| train the classifier using the complete ZERO-label pipeline rather than a random classifier layer | ZERO-label classifier · ImageNet-10 | accuracy: 6.6 -> 47.3 ± 8.6 | Helped |
| train the classifier using the complete ZERO-label pipeline rather than a random classifier layer | ZERO-label classifier · Adience | accuracy: 13.2 -> 32.5 ± 8.1 | Helped |
| train the classifier using the complete ZERO-label pipeline rather than a random classifier layer | ZERO-label classifier · Aesthetic | accuracy: 10.73 -> 57.93 | Helped |
| train the classifier using the complete ZERO-label pipeline rather than a random classifier layer | ZERO-label classifier · DR | accuracy: 39.8 -> 57.7 | Helped |
Closest papers by meaning
- A critical analysis of self-supervision, or what we can learn from a single image (ICLR 2020 · similarity 0.90)
- Theoretical Analysis of Self-Training with Deep Networks on Unlabeled Data (ICLR 2021 · similarity 0.88)
- Masked Unsupervised Self-training for Label-free Image Classification (ICLR 2023 · similarity 0.87)
- Self-supervised Label Augmentation via Input Transformations (ICML 2020 · similarity 0.87)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.85)
- Deep Probabilistic Supervision for Image Classification (NeurIPS 2026 · 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.