AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

State-Regularized Recurrent Neural Networks to Extract Automata and Explain Predictions

Cheng Wang, Carolin Lawrence, Mathias Niepert

TPAMI 2022 · Architectures and efficiency › RNNs and normalization · approach: RNN training dynamics and gradients · 3 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 state regularization with projection (sr-LSTM-p)LSTM · Balanced parentheses (small train set), d=20, l<=1000error rate: 0.519 -> 0.380Improved
Add state regularization with projection (sr-LSTM-p)LSTM · Balanced parentheses (large train set), d in [10,20], l<=200error rate: 0.341 -> 0.003Improved
Add state regularization without projection (sr-LSTM)LSTM · Balanced parentheses (large train set), d=5, l<=200error rate: 0.002 -> 0.044Hurt
Add state regularization with projection (sr-LSTM-p)LSTM · Palindrome test, max length 100error rate (%): 31.2 -> 10.5Improved
Add state regularization with projection (sr-LSTM-p)LSTM · Palindrome test, max length 500error rate (%): 47.7 -> 29.8Improved
Add projection to state-regularized LSTM (ablation)sr-LSTM · Palindrome test, max length 200error rate (%): 36.0 -> 16.7Improved
Add state regularization to LSTMLSTM · IMDBtest error (%): 10.1 -> 9.2Improved
Use alternative sr-LSTM-p formulations instead of best one (ablation)sr-LSTM-p (k=10) · IMDBtest error (%): 9.2 -> 11.1Hurt
Add state regularization to LSTMLSTM · Sequential MNISTtest error (%): 2.3 -> 0.8Improved
Use alternative sr-LSTM-p formulation instead of best one (ablation)sr-LSTM-p (k=100) · Sequential MNISTtest error (%): 0.8 -> 6.7Hurt
Reduce centroids from k=5 to k=2 (ablation)sr-LSTM-p · Balanced parentheses (small train set), d=10, l<=200error rate: 0.138 -> 0.038Improved
Increase centroids from k=5 to k=50 (ablation)sr-LSTM-p · Balanced parentheses (small train set), d=5, l<=200error rate: 0.015 -> 0.047Hurt

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