Research map / Label-efficient and robust learning / Recognition and long-tail
WeightCLIP: Aligning Datasets and Models for Weight Space Learning
Aron Asefaw, Konstantinos Tzevelekakis, Damian Falk, Léo Meynent, Damian Borth
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 dataset-model alignment loss (ablation) | WeightCLIP · In-distribution retrieval | R@1 (%): 10 -> 95 | Improved |
| add dataset-model alignment loss (ablation) | WeightCLIP · Linear-mapper generation Ep. 0 (first three OOD datasets) | accuracy (%): 14.5 -> 26.6 | Improved |
| align both model and dataset encoders instead of dataset encoder only (ablation) | WeightCLIP · Linear-mapper generation Ep. 0 | accuracy (%): 23.9 -> 26.6 | Improved |
| align dataset encoder only (D2NWG) instead of no alignment (ablation) | WeightCLIP · Linear-mapper generation Ep. 0 | accuracy (%): 14.5 -> 23.9 | Improved |
| replace nearest-neighbor selection with linear mapper | WeightCLIP · ResNet18 OOD Ep. 0 | accuracy (%): 26.3 -> 30.9 | Improved |
| use memory-bank mapper instead of linear mapper | WeightCLIP · ResNet18 OOD Ep. 0 | accuracy (%): 30.9 -> 31.5 | Improved |
| use latent-space refinement instead of weight-space fine-tuning | generated ResNet18 models · 6 OOD datasets average | accuracy (%): 38.8 -> 44.9 | Improved |
| add hyperspherical shell constraint to latent refinement (ablation) | WeightCLIP · refinement average | accuracy (%): 42.4 -> 44.9 | Improved |
| add hyperspherical shell constraint to unaligned latent refinement (ablation) | WeightCLIP w/o align. · refinement average | accuracy (%): 37.2 -> 40.7 | Improved |
| add alignment to shell-constrained latent refinement (ablation) | WeightCLIP · refinement average | accuracy (%): 40.7 -> 44.9 | Improved |
| increase dataset prompt size from 1 to 32 images (ablation) | WeightCLIP · OOD generation after 1 epoch | accuracy (%): 61.1 -> 64.3 | No clear effect |
| increase dataset prompt size from 32 to 100 images (ablation) | WeightCLIP · OOD generation after 1 epoch | accuracy (%): 64.3 -> 63.8 | No clear effect |
Closest papers by meaning
- Towards Scalable and Versatile Weight Space Learning (ICML 2024 · similarity 0.88)
- Improved Generalization of Weight Space Networks via Augmentations (ICML 2024 · similarity 0.87)
- Equivariant Deep Weight Space Alignment (ICML 2024 · similarity 0.84)
- Gradient-Weight Alignment as a Train-Time Proxy for Generalization in Classification Tasks (NeurIPS 2025 · similarity 0.84)
- Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning (NeurIPS 2025 · similarity 0.84)
- Latent Zoning Network: A Unified Principle for Generative Modeling, Representation Learning, and Classification (NeurIPS 2025 · 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.
