AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Uncovering Neural Scaling Laws in Molecular Representation Learning

Dingshuo Chen, Yanqiao Zhu, Jieyu Zhang, Yuanqi Du, Zhixun Li, Qiang Liu, Shu Wu, Liang Wang

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Scalable approximation and equivariant compute · 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
add GraphMAE pre-training on PCQM4Mv2 before fine-tuningGIN · small downstream datasetsROC-AUCImproved
add GraphMAE pre-trainingGIN · PCBA, above 40K training samplesROC-AUCHurt
pre-training vs from scratch: scaling exponentGraphMAE-pretrained GIN · downstream scaling curvespower-law exponentHurt
use random split instead of scaffold or imbalanced splitGIN · scaling curvespower-law exponentImproved
use graph modality instead of SMILES or fingerprintGIN vs Transformer · 3 tasksdata efficiencyImproved
use SMILES strings instead of graph1-layer Transformer · MUVROC-AUC at large dataHurt
increase GIN capacity from 128 to 1.5K parameters (ablation)GIN · HIV and PCBAROC-AUCMixed
use smallest GIN (128 parameters) (ablation)GIN · MUVROC-AUCImproved
increase Transformer layers for SMILES (RoBERTa, 1-layer, 3-layer) (ablation)Transformer · SMILES modalityROC-AUCMixed
replace 1-layer Transformer by MLP for fingerprints (ablation)fingerprint encoder · scaling curvespower-law adherenceNo clear effect
data pruning with k-means instead of randomGIN · PCBA uniform, 1% subsetROC-AUC: 5.2 -> 4.1Hurt
data pruning with Herding instead of randomGIN · PCBA uniform, 1% subsetROC-AUC: 5.2 -> 3.7Hurt

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