AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Improving the Performance and Learning Stability of Parallelizable RNNs Designed for Ultra-Low Power Applications

Julien Brandoit, Arthur Fyon, Damien Ernst, Guillaume Drion

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: RNN training dynamics and gradients · 0 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 cumulative update (epsilon 0 -> 1) (ablation)CMRU d=32 · sMNISTaccuracy (%): 35.32 -> 95.92Improved
add cumulative update (epsilon 0 -> 1) (ablation)alphaCMRU d=32 · sMNISTaccuracy (%): 48.86 -> 96.76Improved
anneal epsilon from 1 to 0 instead of constant epsilon=0 (ablation)CMRU d=32 · sMNISTaccuracy (%): 35.32 -> 87.94Improved
train with constant epsilon=1 instead of annealed epsilon (ablation)CMRU d=32 · sMNISTaccuracy (%): 87.94 -> 95.92Improved
anneal epsilon vs constant epsilon=1 (ablation)alphaCMRU d=32 · sMNISTaccuracy (%): 96.76 -> 96.94No clear effect
use CMRU instead of LRU at d=4CMRU (single layer) · sMNISTaccuracy (%): 83.61Improved
use CMRU instead of LRU/minGRU for persistent memoryCMRU r=3 d=32 · Pathfinderaccuracy (%): 50.14 -> 90.20Improved
use CMRU instead of LRU on ListOpsCMRU r=3 d=32 · ListOpsaccuracy (%): 40.45 -> 60.42Improved
use CMRU instead of LRU on IMDbCMRU r=3 d=32 · IMDbaccuracy (%): 66.84 -> 65.18Hurt
use CMRU instead of minGRU on sCIFAR10CMRU r=3 d=32 · sCIFAR10accuracy (%): 60.95 -> 59.82Hurt
combine minGRU with CMRU layersminGRU · Shakespeare char-level LMcross-entropy loss: 1.453 -> 1.442Improved
use CMRU instead of LRU for keyword spottingCMRU r=1 d=16 · KWS Allaccuracy (%): 91.82 -> 89.76Hurt

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