US Patent Application 18031789. A LOOK AHEAD STRATEGY FOR TRIE-BASED BEAM SEARCH IN GENERATIVE RETRIEVAL simplified abstract

From WikiPatents
Jump to navigation Jump to search

A LOOK AHEAD STRATEGY FOR TRIE-BASED BEAM SEARCH IN GENERATIVE RETRIEVAL

Organization Name

Microsoft Technology Licensing, LLC==Inventor(s)==

[[Category:Jian Jiao of Bellevue WA (US)]]

[[Category:Yeyun Gong of Beijing (CN)]]

[[Category:Nan Duan of Beijing (CN)]]

[[Category:Ruofei Zhang of Sunnyvale CA (US)]]

[[Category:Ming Zhou of Beijing (CN)]]

A LOOK AHEAD STRATEGY FOR TRIE-BASED BEAM SEARCH IN GENERATIVE RETRIEVAL - A simplified explanation of the abstract

This abstract first appeared for US patent application 18031789 titled 'A LOOK AHEAD STRATEGY FOR TRIE-BASED BEAM SEARCH IN GENERATIVE RETRIEVAL

Simplified Explanation

- This patent application describes a system and method for generating a keyword sequence from an input query. - The system uses a machine learning model that includes an encoder and a decoder. - The encoder takes the input query and encodes it into a source sequence representation. - The decoder then generates a keyword sentence based on the source sequence representation. - The decoder calculates a modified generation score for each prediction token, taking into account the prediction token generation score and a maximum generation score for a suffix of each prediction token. - The decoder selects the prediction token with the highest modified generation score and adds it to the previously decoded partial hypothesis. - This process continues until the desired keyword sequence is generated.


Original Abstract Submitted

Systems and methods are provided for generating a keyword sequence from an input query. A first text sequence corresponding to an input query may be received and encoded into a source sequence representation using an encoder of a machine learning model. A keyword sentence may then be generated from the source sequence representation using a decoder of the machine learning model. The decoder may generate a modified generation score for a plurality of prediction tokens, wherein the modified generation score is based on the respective prediction token generation score and a maximum generation score for a suffix of each prediction token. The decoder may then select the prediction token of the plurality of prediction tokens based on the modified generation score, and add the selected prediction token to the previously decoded partial hypothesis provided by the decoder.