AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Direct Feedback Alignment Scales to Modern Deep Learning Tasks and Architectures

Julien Launay, Iacopo Poli, François Boniface, Florent Krzakala

NeurIPS 2020 · Architectures and efficiency › RNNs and normalization · approach: Backprop-free credit assignment · 5 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
train with DFA instead of BPNeRF · NeRF-SyntheticPSNR: 31.01 -> 25.41Hurt
train with DFA instead of BPNeRF · LLFF-RealPSNR: 26.50 -> 20.77Hurt
train with DFA instead of BPDeepFM · CriteoAUC: 0.7954 -> 0.7956No clear effect
train with DFA instead of BPDeep&Cross · CriteoAUC: 0.8104 -> 0.8009Hurt
train with DFA instead of BPAFN · CriteoAUC: 0.7933 -> 0.7924No clear effect
train with DFA instead of BPGraphConv · Coraaccuracy (%): 80.1 -> 79.9No clear effect
train with DFA instead of BPChebConv · Coraaccuracy (%): 79.2 -> 75.4Hurt
train with DFA instead of BPDNAConv · PubMedaccuracy (%): 87.2 -> 79.9Hurt
DFA instead of shallow training (top layer only)Deep&Cross · CriteoAUC: 0.7324 -> 0.8009Improved
use Adam instead of baseline optimizer (DFA, Macro) (ablation)Transformer · WikiText-103validation perplexity: 95.0 -> 77.1Improved
increase beta2 from 0.98 to 0.999 (DFA, Macro) (ablation)Transformer · WikiText-103validation perplexity: 77.1 -> 55.0Improved
add plateau LR scheduler (DFA, Macro) (ablation)Transformer · WikiText-103validation perplexity: 55.0 -> 52.0Improved

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