AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Implicit Bias of Linear RNNs

Melikasadat Emami, Mojtaba Sahraee-Ardakan, Parthe Pandit, Sundeep Rangan, Alyson K Fletcher

ICML 2021 · Architectures and efficiency › RNNs and normalization · approach: RNN training dynamics and gradients · 1 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
Use the short-memory-biased RNN parameterization instead of unscaled Conv-1DMacaque S1 reaching models · Limited data (4.5 minutes), x directionR²: 0.6565 -> 0.6462Hurt
Use the short-memory-biased RNN parameterization instead of unscaled Conv-1DMacaque S1 reaching models · Limited data (4.5 minutes), y directionR²: 0.6027 -> 0.5911Hurt
Use the short-memory-biased RNN parameterization instead of unscaled Conv-1DMacaque S1 reaching models · Entire recording, x directionR²: 0.5856 -> 0.6043Improved
Use the short-memory-biased RNN parameterization instead of unscaled Conv-1DMacaque S1 reaching models · Entire recording, y directionR²: 0.3918 -> 0.4257Improved
Increase delay in the target systemRNN and scaled Conv-1D · Synthetic delayed-input taskTest errorHurt
Use the short-memory-biased RNN parameterization instead of unscaled Conv-1DMacaque S1 reaching models · Limited data (4.5 minutes), x and y directionsR²: RNN: 0.6462, 0.5911; Conv-1D: 0.6565, 0.6027Mixed

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