Research map / Architectures and efficiency / Pruning and efficient tuning
Your representations are in the network: composable and parallel adaptation for large scale models
Yonatan Dukler, Alessandro Achille, Hao Yang, Varsha Vivek, Luca Zancato, Benjamin Bowman, Avinash Ravichandran, Charless Fowlkes, Ashwin Swaminathan, Stefano Soatto
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 the adapter to intermediate layers instead of only the last layer (ablation) | InCA (ViT-L/16) · 11 fine-grained datasets | mean top-1 test error: 11.5 -> 9.8 | Improved |
| apply adapter to intermediate layers instead of last layer (ablation) | InCA (ViT-L/16) · Aircrafts | top-1 test error: 21.9 -> 15.6 | Improved |
| apply adapter to intermediate layers instead of last layer (ablation) | InCA (ViT-L/16) · Stanf. Cars | top-1 test error: 10.2 -> 7.7 | Improved |
| intermediate layers instead of last layer on SWIN-L (ablation) | InCA (SWIN-L) · max gap to full FT | gap: -9.5 -> -3.6 | Improved |
| replace linear probing with InCA | ViT-L/16 · 11 fine-grained datasets | mean top-1 test error: 21.9 -> 9.8 | Improved |
| replace 3-layer MLP head with InCA | ViT-L/16 · 11 fine-grained datasets | mean top-1 test error: 17.7 -> 9.8 | Improved |
| replace LoRA with InCA | ViT-L/16 · 11 fine-grained datasets | mean top-1 test error: 11.3 -> 9.8 | Improved |
| replace VPT with InCA | ViT-L/16 · 11 fine-grained datasets | mean top-1 test error: 12.3 -> 9.8 | Improved |
| replace full fine-tuning with InCA | ViT-L/16 · 11 fine-grained datasets | mean top-1 test error: 10.0 -> 9.8 | Improved |
| InCA vs full fine-tuning on SWIN-L | SWIN-L · 11 fine-grained datasets | mean top-1 test error: 9.3 -> 10.0 | Hurt |
| use adapter on intermediate activations of ViT-B/16 In21K backbone | ViT-B/16 In21K · 11 fine-grained datasets | mean top-1 test error: 13.0 -> 15.9 | Hurt |
| InCA vs full FT on ViT-B/16 ALBEF | ViT-B/16 ALBEF · 11 fine-grained datasets | mean top-1 test error: 13.8 -> 13.5 | Improved |
Closest papers by meaning
- Time- Memory- and Parameter-Efficient Visual Adaptation (CVPR 2024 · similarity 0.88)
- Minimal Interaction Seperated Tuning: A New Paradigm for Visual Adaptation (CVPR 2025 · similarity 0.86)
- Unleashing Foundation Vision Models: Adaptive Transfer for Diverse Data-Limited Scientific Domains (NeurIPS 2025 · similarity 0.85)
- Transferring Knowledge From Large Foundation Models to Small Downstream Models (ICML 2024 · similarity 0.85)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.85)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (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.
