AI Paper Atlas

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

Distilling Knowledge from Reader to Retriever for Question Answering

Gautier Izacard, Edouard Grave

ICLR 2021 · Label-efficient and robust learning › Knowledge distillation · approach: Patient consistent distillation · 39 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
Distill reader cross-attention scores into a retriever, initialized from DPRFusion-in-Decoder with T5 large · NaturalQuestions testEM: 51.4 -> 54.4Improved
Distill reader cross-attention scores into a retriever, initialized from DPRFusion-in-Decoder with T5 large · TriviaQA testEM: 67.6 -> 72.5Improved
Initialize the retriever training passages from DPR rather than BM25Proposed reader-to-retriever distillation · NaturalQuestions testR@20: 80.0 -> 84.3Improved
Initialize the retriever training passages from DPR rather than BM25Proposed reader-to-retriever distillation · TriviaQA testR@20: 81.4 -> 83.6Improved
Perform an additional self-training iterationProposed method initialized with BM25 · NaturalQuestions validationDev EM: 41.2 -> 46.8 -> 47.9 -> 46.2Mixed
Perform an additional self-training iterationProposed method initialized with BM25 · TriviaQA validationDev EM: 65.3 -> 66.7 -> 67.5 -> 67.7Improved
Use KL-divergence rather than mean squared error to train the retriever (ablation)Proposed method initialized with BM25 · NaturalQuestions validationR@20: 61.2 -> 76.4Improved
Use KL-divergence rather than max-margin loss, gamma=1 (ablation)Proposed method initialized with BM25 · NaturalQuestions validationR@20: 73.6 -> 76.4Improved
Use KL-divergence rather than max-margin loss, gamma=0.2 (ablation)Proposed method initialized with BM25 · NaturalQuestions validationR@20: 73.5 -> 76.4Improved
Use KL-divergence rather than max-margin loss, gamma=0.1 (ablation)Proposed method initialized with BM25 · NaturalQuestions validationR@20: 73.5 -> 76.4Improved
Average attention over the first output token only rather than all output tokens (ablation)Proposed method initialized with BM25 · NaturalQuestions validationR@20: 75.3 -> 76.4Improved
Average attention over passage tokens rather than taking their maximum (ablation)Proposed method initialized with BM25 · NaturalQuestions validationR@20: 72.5 -> 76.4Improved

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