AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Self-Supervised Representation Learning on Neural Network Weights for Model Characteristic Prediction

Konstantin Schürholt, Dimche Kostadinov, Damian Borth

NeurIPS 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Data-efficient and robust self-supervision · 7 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
add permutation augmentation (ablation)EcD · TETRIS-SEEDtest reconstruction R2 (%): 49.2 -> 72.9Improved
add permutation to erasing and noise augmentation (ablation)EcD · TETRIS-SEEDtest reconstruction R2 (%): 65.4 -> 84.1Improved
add erasing to permutation augmentation (ablation)EcD · TETRIS-SEEDtest reconstruction R2 (%): 72.9 -> 84.1Improved
add noise augmentation to permutation and erasing (ablation)EcD · TETRIS-SEEDtest reconstruction R2 (%): 84.1 -> 84.1No clear effect
combine reconstruction with contrastive loss (EcD vs ED) (ablation)hyper-representation · TETRIS-SEEDaccuracy prediction R2 (%): 68.9 -> 90.2Improved
add reconstruction to contrastive loss (EcD vs Ec) (ablation)hyper-representation · TETRIS-SEEDaccuracy prediction R2 (%): 90.8 -> 90.2No clear effect
use EcD instead of Ec+D (ablation)hyper-representation · TETRIS-SEEDaccuracy prediction R2 (%): 69.9 -> 90.2Improved
switch feed-forward to attention on neuron tokens with compression token (ablation)AttN+t · TETRIS-SEEDaccuracy prediction R2 (%): 85.3 -> 90.2Improved
tokenize each weight (AttW) instead of per neuron (AttN) (ablation)attention encoder · TETRIS-SEEDreconstruction R2 (%): 89.4 -> 6.8Hurt
add compression token to weight-token attention (ablation)AttW · TETRIS-SEEDreconstruction R2 (%): 6.8 -> 74.1Improved
add compression token to neuron-token attention (ablation)AttN · TETRIS-SEEDaccuracy prediction R2 (%): 88.4 -> 90.2Improved
add compression token to neuron-token attention (ablation)AttN · TETRIS-SEEDreconstruction R2 (%): 89.4 -> 84.1Hurt

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