REALM: Retrieval-Augmented Language Model Pre-Training (Paper Explained)

#ai #tech #science Open Domain Question Answering is one of the most challenging tasks in NLP. When answering a question, the model is able to retrieve arbitrary documents from an indexed corpus to gather more information. REALM shows how Masked Language Modeling (MLM) pretraining can be used to train a retriever for relevant documents in an end-to-end fashion and improves over state-of-the-art by a significant margin. OUTLINE: 0:00 - Introduction & Overview 4:30 - World Knowledge in Language Models 8:15 - Masked Language Modeling for Latent Document Retrieval 14:50 - Problem Formulation 17:30 - Knowledge Retriever Model using MIPS 23:50 - Question Answering Model 27:50 - Architecture Recap 29:55 - Analysis of the Loss Gradient 34:15 - Initialization using the Inverse Cloze Task 41:40 - Prohibiting Trivial Retrievals 44:05 - Null Document 45:00 - Salient Span Masking 50:15 - My Idea on Salient Span Masking 51:50 - Experimental Results and Ablations 57:30 - Concrete Example from the Model Paper: https://arxiv.org/abs/2002.08909 Code: https://github.com/google-research/language/tree/master/language/realm My Video on GPT-3: https://www.youtube.com/watch?v=SY5PvZrJhLE My Video on BERT: https://www.youtube.com/watch?v=-9evrZnBorM My Video on Word2Vec: https://www.youtube.com/watch?v=yexR53My2O4 Abstract: Language model pre-training has been shown to capture a surprising amount of world knowledge, crucial for NLP tasks such as question answering. However, this knowledge is stored implicitly in the parameters of a neural network, requiring ever-larger networks to cover more facts. To capture knowledge in a more modular and interpretable way, we augment language model pre-training with a latent knowledge retriever, which allows the model to retrieve and attend over documents from a large corpus such as Wikipedia, used during pre-training, fine-tuning and inference. For the first time, we show how to pre-train such a knowledge retriever in an unsupervised manner, using masked language modeling as the learning signal and backpropagating through a retrieval step that considers millions of documents. We demonstrate the effectiveness of Retrieval-Augmented Language Model pre-training (REALM) by fine-tuning on the challenging task of Open-domain Question Answering (Open-QA). We compare against state-of-the-art models for both explicit and implicit…