AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Self-Instantiated Recurrent Units with Dynamic Soft Recursion

Aston Zhang, Yi Tay, Yikang Shen, Alvin Chan, Shuai Zhang

NeurIPS 2021 · Architectures and efficiency › RNNs and normalization · approach: Brain-inspired recurrent architectures · 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
Use Self-IRU instead of ON-LSTM on longest sequenceslogical inference model · Logical inference, 12 operationsaccuracy: 76.0 -> 88.0Improved
Use Self-IRU instead of Tree-LSTM with ground-truth syntaxlogical inference model · Logical inference, 12 operationsaccuracy: 87.0 -> 88.0Improved
Use Self-IRU encoder instead of TrellisNet with 8x more parameterssequence encoder · Pixel-wise CIFAR-10accuracy: 73.42 -> 73.01No clear effect
Use Self-IRU instead of Transformersequence encoder · Pixel-wise CIFAR-10accuracy: 62.20 -> 73.01Improved
Replace recurrent decoder of TranX with Self-IRUTranX · Django code generationaccuracy: 77.2 -> 78.3Improved
Replace recurrent decoder of TranX with Self-IRUTranX · Atis semantic parsingaccuracy: 87.7 -> 88.4Improved
Replace recurrent decoder of TranX with Self-IRUTranX · Geo semantic parsingaccuracy: 88.6 -> 88.6No clear effect
Use Self-IRU instead of TCN on JSB Choralesmusic model · JSB Choralesnegative log-likelihood: 8.13 -> 8.12No clear effect
Use Self-IRU instead of TCN on Nottinghammusic model · Nottinghamnegative log-likelihood: 2.95 -> 2.88Improved
Increase max recursion depth from 1 to 3 with linear base transformation (ablation)Self-IRU · Atis semantic parsingaccuracy: 88.40 -> 87.80Hurt
Use LSTM instead of linear base transformation (depth 1) (ablation)Self-IRU · Atis semantic parsingaccuracy: 88.40 -> 86.61Hurt
Use LSTM instead of linear base transformation (depth 1) (ablation)Self-IRU · Django code generationaccuracy: 77.56 -> 78.33Improved

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