Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Deep Learning by Neighbourhood Discovery
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 |
|---|---|---|---|
| Use anchor-neighbourhood discovery rather than DeepCluster | AlexNet unsupervised representation learning · CIFAR10 | weighted kNN accuracy: 62.3 -> 74.8 | Helped |
| Use anchor-neighbourhood discovery rather than Instance learning | AlexNet unsupervised representation learning · CIFAR100 | weighted kNN accuracy: 32.7 -> 41.5 | Helped |
| Use anchor-neighbourhood discovery rather than DeepCluster | AlexNet unsupervised representation learning · SVHN | weighted kNN accuracy: 84.9 -> 90.9 | Helped |
| Use anchor-neighbourhood discovery rather than DeepCluster | AlexNet unsupervised representation learning · CIFAR10 | linear classifier accuracy: 77.9 -> 77.6 | Hurt |
| Use anchor-neighbourhood discovery rather than DeepCluster | AlexNet unsupervised representation learning · CIFAR100 | linear classifier accuracy: 41.9 -> 47.9 | Helped |
| Use anchor-neighbourhood discovery rather than Instance learning | ResNet18 unsupervised representation learning · CUB200 | accuracy: 11.6 -> 14.4 | Helped |
| Use anchor-neighbourhood discovery rather than Instance learning | ResNet18 unsupervised representation learning · Stanford Dogs | accuracy: 27.0 -> 32.3 | Helped |
| Replace AlexNet with a stronger ResNet backbone (ablation) | AND · CIFAR10 | weighted kNN accuracy: 74.8 -> 86.3; 74.8 -> 88.4 | Helped |
| Use Instance initialization instead of random initialization (ablation) | AND · CIFAR10 | accuracy: 85.7 -> 86.3 | Helped |
| Use curriculum rather than one-off neighbourhood discovery (ablation) | AND · CIFAR10 | accuracy: 84.2 -> 86.3 | Helped |
| Use 100 rather than 50 epochs per curriculum round (ablation) | AND · CIFAR10 | accuracy: 83.3% -> 84.8% | Helped |
| Use the smallest neighbourhood size (k=1) (ablation) | AND · CIFAR10 | accuracy | Helped |
Closest papers by meaning
- Unsupervised Deep Learning via Affinity Diffusion (AAAI 2020 · similarity 0.91)
- Unsupervised Learning of Discriminative Attributes and Visual Representations (CVPR 2016 · similarity 0.90)
- Unsupervised Representation Learning by Predicting Image Rotations (ICLR 2018 · similarity 0.88)
- Learning from Small Sample Sets by Combining Unsupervised Meta-Training with CNNs (NeurIPS 2016 · similarity 0.88)
- Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (AAAI 2019 · similarity 0.88)
- Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels (ICLR 2024 · 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.