AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

EM-Network: Oracle Guided Self-distillation for Sequence Learning

Ji Won Yoon, SungHwan Ahn, Hyeonseung Lee, Minchan Kim, Seok Min Kim, Nam Soo Kim

ICML 2023 · Label-efficient and robust learning › Knowledge distillation · approach: Self and data-free distillation · 0 citations (OpenAlex)

Read the paper

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 target-derived oracle guidance and self-distillationConformer-CTC Small · LibriSpeech test-clean, w/o LMWER (%): 4.87 -> 4.29Improved
Add target-derived oracle guidance and self-distillationConformer-CTC Small · LibriSpeech test-other, w/o LMWER (%): 12.05 -> 10.81Improved
Add target-derived oracle guidance and self-distillationdata2vec Base · LibriSpeech test-clean, greedy decodingWER (%): 2.78 -> 2.66Improved
Add target-derived oracle guidance and self-distillationTransformer · IWSLT'14 En-De testBLEU: 29.43 -> 29.94Improved
Add target-derived oracle guidance and self-distillationTransformer · IWSLT'14 De-En testBLEU: 35.99 -> 36.53Improved
Add target-derived oracle guidance and self-distillationBiBERT · WMT'14 newstest2014 En-DeBLEU: 31.26 -> 31.30Improved
Add target-derived oracle guidance and self-distillationBiBERT · WMT'14 newstest2014 De-EnBLEU: 34.94 -> 35.40Improved
Train with online self-distillation rather than offline distillation (ablation)EM-Network with Conformer-CTC Small · LibriSpeech test-cleanWER (%): 4.35 -> 4.29Improved
Train with online self-distillation rather than offline distillation (ablation)EM-Network with Conformer-CTC Small · LibriSpeech test-otherWER (%): 10.97 -> 10.81Improved
Add EM-Network self-distillation to the hybrid CTC/attention modelHybrid CTC/attention Transformer · LibriSpeech test-clean with LMWER (%): 4.1 -> 3.6Improved
Add EM-Network self-distillation to the hybrid CTC/attention modelHybrid CTC/attention Transformer · LibriSpeech test-other with LMWER (%): 9.4 -> 8.8Improved
Use EM-Network fine-tuning rather than data2vec fine-tuningdata2vec · LibriSpeech test-other, 100h fine-tuningWER (%): 9.8 -> 9.8No 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