Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Geometry-Aware Adaptation for Pretrained Models
Nicholas Roberts, Xintong Li, Dyah Adila, Sonia Cromp, Tzu-Heng Huang, Jitian Zhao, Frederic Sala
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 |
|---|---|---|---|
| apply LOKI with internal text-embedding metric | CLIP-RN50 · CIFAR-100 | E[d2(y, y_hat)]: 0.2922 -> 0.2613 | Improved |
| apply LOKI with internal metric | CLIP-ViT-L-14 · CIFAR-100 | E[d2(y, y_hat)]: 0.1588 -> 0.1562 | Improved |
| apply LOKI with WordNet tree metric | CLIP-RN50 · CIFAR-100 | E[d2(y, y_hat)]: 24.3017 -> 19.5549 | Improved |
| apply LOKI with default CIFAR-100 tree | CLIP-RN50 · CIFAR-100 | E[d2(y, y_hat)]: 7.3528 -> 7.1888 | Improved |
| apply LOKI under the complete graph K100 | CLIP-RN50 · CIFAR-100 | E[d2(y, y_hat)]: 0.5941 -> 0.5941 | No clear effect |
| apply LOKI to 5-NN on partially observed labels | 5-NN · LSHTC K=50 | E[d2(y, y_hat)]: 1.2519 -> 0.2896 | Improved |
| apply LOKI to 5-NN on partially observed labels | 5-NN · PubMed K=250 | E[d2(y, y_hat)]: 1.52374 -> 1.47801 | Improved |
| apply LOKI to 5-NN on partially observed labels | 5-NN · PubMed K=100 | E[d2(y, y_hat)]: 1.68666 -> 1.42591 | Improved |
| apply LOKI to one-vs-rest classifier on ImageNet subsets | SimCLR logistic classifier · ImageNet, uniform and Gibbs class sampling | mean squared distance | Improved |
| select next class actively instead of randomly (ablation) | LOKI locus · synthetic tree N=100, K=3 | locus size | Improved |
| select 50 additional classes actively instead of randomly (ablation) | SimCLR + LOKI · ImageNet, 500 initial classes | mean squared distance | Improved |
| tune softmax temperature of CLIP (ablation) | LOKI · CIFAR-100 no external metric | zero-one error | Improved |
Closest papers by meaning
- Synthesized Classifiers for Zero-Shot Learning (CVPR 2016 · similarity 0.90)
- OTTER: Effortless Label Distribution Adaptation of Zero-shot Models (NeurIPS 2024 · similarity 0.88)
- Classifier and Exemplar Synthesis for Zero-Shot Learning (IJCV 2019 · similarity 0.88)
- SimZSL: Zero-Shot Learning Beyond a Pre-defined Semantic Embedding Space (IJCV 2025 · similarity 0.87)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 2024 · similarity 0.87)
- Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations (NeurIPS 2024 · similarity 0.86)
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.
