AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Bivariate Beta-LSTM

Kyungwoo Song, JoonHo Jang, Seung jae Shin, Il-Chul Moon

AAAI 2020 · Architectures and efficiency › RNNs and normalization · approach: LSTM design and memory · 2 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 flexible, correlated Beta-distribution input and forget gates with a structured priorLSTM · CR, SUBJ, MR, TREC, MPQA, SSTaccuracy: 84.66±2.42; 93.25±0.85; 81.59±0.91; 94.80±0.47; 89.66±0.44; 88.94±0.43Improved
Add bBeta-LSTM(5G+p) gatesSRU(8-layers) · CR, SUBJ, MR, TREC, MPQA, SSTaccuracy: 87.00±2.24 -> 87.21±0.78; 93.76±0.61 -> 93.97±0.55; 83.14±1.53 -> 83.51±1.42; 94.52±0.34 -> 94.80±0.47; 90.39±0.55 -> 90Improved
Add bBeta-LSTM(5G+p) gatesR-Transformer · JSB, Muse, Nottingham, Pianonegative log-likelihood: 8.26±0.03 -> 8.24±0.01; 7.00±0.03 -> 6.19±0.02; 2.24±0.01 -> 2.13±0.08; 7.44±0.03 -> 7.32±0.03Improved
Add bBeta-LSTM(5G+p) gatesBN-LSTM · sMNIST and pMNISTerror rate: 1.05% -> 0.76%; 4.26±0.50 -> 3.90±0.25Improved
Add bBeta-LSTM(5G+p) gatesh-detach(0.4), Show & Tell with Resnet152 · MS-COCOB-4: 30.3 -> 30.8Improved
Add bBeta-LSTM(5G+p) gatesh-detach(0.4), Show Attend Tell with Resnet152 · MS-COCOCIDEr: 101.2 -> 103.2Improved
Use independent Beta-distributed gates instead of correlated gates (ablation)bBeta-LSTM · CRaccuracy: 84.45±1.87 -> 84.66±2.42Hurt
Restrict gate correlation to positive values using three Gamma variables (ablation)bBeta-LSTM · CRaccuracy: 83.63±2.14 -> 84.66±2.42Hurt
Remove the structured prior (ablation)bBeta-LSTM(5G+p) · CRaccuracy: 84.48±1.96 -> 84.66±2.42Hurt
Use bBeta-LSTM(5G+p) instead of LSTM gatesShow & Tell with Resnet152 · MS-COCOMETEOR: 24.8 -> 24.7No clear effect
Use bBeta-LSTM(5G+p) instead of h-detach(0.4) gatesShow & Tell with Resnet152 and h-detach · MS-COCOMETEOR: 25.2 -> 25.2No clear effect

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