AI Paper Atlas

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

NeurIPS 2023 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging · 2 citations (OpenAlex)

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 changedWhereMeasuredEffect
apply the adapter to intermediate layers instead of only the last layer (ablation)InCA (ViT-L/16) · 11 fine-grained datasetsmean top-1 test error: 11.5 -> 9.8Improved
apply adapter to intermediate layers instead of last layer (ablation)InCA (ViT-L/16) · Aircraftstop-1 test error: 21.9 -> 15.6Improved
apply adapter to intermediate layers instead of last layer (ablation)InCA (ViT-L/16) · Stanf. Carstop-1 test error: 10.2 -> 7.7Improved
intermediate layers instead of last layer on SWIN-L (ablation)InCA (SWIN-L) · max gap to full FTgap: -9.5 -> -3.6Improved
replace linear probing with InCAViT-L/16 · 11 fine-grained datasetsmean top-1 test error: 21.9 -> 9.8Improved
replace 3-layer MLP head with InCAViT-L/16 · 11 fine-grained datasetsmean top-1 test error: 17.7 -> 9.8Improved
replace LoRA with InCAViT-L/16 · 11 fine-grained datasetsmean top-1 test error: 11.3 -> 9.8Improved
replace VPT with InCAViT-L/16 · 11 fine-grained datasetsmean top-1 test error: 12.3 -> 9.8Improved
replace full fine-tuning with InCAViT-L/16 · 11 fine-grained datasetsmean top-1 test error: 10.0 -> 9.8Improved
InCA vs full fine-tuning on SWIN-LSWIN-L · 11 fine-grained datasetsmean top-1 test error: 9.3 -> 10.0Hurt
use adapter on intermediate activations of ViT-B/16 In21K backboneViT-B/16 In21K · 11 fine-grained datasetsmean top-1 test error: 13.0 -> 15.9Hurt
InCA vs full FT on ViT-B/16 ALBEFViT-B/16 ALBEF · 11 fine-grained datasetsmean top-1 test error: 13.8 -> 13.5Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy