Transformer Memory as a Differentiable Search Index (Machine Learning Research Paper Explained)
#dsi #search #google Search engines work by building an index and then looking up things in it. Usually, that index is a separate data structure. In keyword search, we build and store reverse indices. In neural search, we build nearest-neighbor indices. This paper does something different: It directly trains a Transformer to return the ID of the most relevant document. No similarity search over embeddings or anything like this is performed, and no external data structure is needed, as the entire index is essentially captured by the model's weights. The paper experiments with various ways of representing documents and training the system, which works surprisingly well! Sponsor: Diffgram https://diffgram.com?ref=yannic OUTLINE: 0:00 - Intro 0:45 - Sponsor: Diffgram 1:35 - Paper overview 3:15 - The search problem, classic and neural 8:15 - Seq2seq for directly predicting document IDs 11:05 - Differentiable search index architecture 18:05 - Indexing 25:15 - Retrieval and document representation 33:25 - Training DSI 39:15 - Experimental results 49:25 - Comments & Conclusions Paper: https://arxiv.org/abs/2202.06991 Abstract: In this paper, we demonstrate that information retrieval can be accomplished with a single Transformer, in which all information about the corpus is encoded in the parameters of the model. To this end, we introduce the Differentiable Search Index (DSI), a new paradigm that learns a text-to-text model that maps string queries directly to relevant docids; in other words, a DSI model answers queries directly using only its parameters, dramatically simplifying the whole retrieval process. We study variations in how documents and their identifiers are represented, variations in training procedures, and the interplay between models and corpus sizes. Experiments demonstrate that given appropriate design choices, DSI significantly outperforms strong baselines such as dual encoder models. Moreover, DSI demonstrates strong generalization capabilities, outperforming a BM25 baseline in a zero-shot setup. Authors: Yi Tay, Vinh Q. Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Gupta, Tal Schuster, William W. Cohen, Donald Metzler Links: Merch: store.ykilcher.com TabNine Code Completion (Referral): http://bit.ly/tabnine-yannick YouTube: https://www.youtube.com/c/yannickilcher Twitter: https://twitter.com/ykilcher Discord: https://discord.gg/4H8xxDF BitChute: https://www.bitchute.com/channel/yann... LinkedIn: https://www.linkedin.com/in/ykilcher BiliBili: https://space.bilibili.com/2017636191 If you want to support me, the best thing to do is to share out the content :) If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this): SubscribeStar: https://www.subscribestar.com/yannick... Patreon: https://www.patreon.com/yannickilcher Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2 Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n