US 11,868,724 B2
Generating author vectors
Quoc V. Le, Sunnyvale, CA (US); and Brian Patrick Strope, Palo Alto, CA (US)
Assigned to GOOGLE LLC, Mountain View, CA (US)
Filed by GOOGLE LLC, Mountain View, CA (US)
Filed on Mar. 14, 2022, as Appl. No. 17/654,660.
Application 17/654,660 is a continuation of application No. 16/824,216, filed on Mar. 19, 2020, granted, now 11,275,895.
Application 16/824,216 is a continuation of application No. 15/991,531, filed on May 29, 2018, granted, now 10,599,770, issued on Mar. 24, 2020.
Application 15/991,531 is a continuation of application No. 15/206,777, filed on Jul. 11, 2016, granted, now 9,984,062, issued on May 29, 2018.
Claims priority of provisional application 62/191,120, filed on Jul. 10, 2015.
Prior Publication US 2022/0198145 A1, Jun. 23, 2022
Int. Cl. G06F 40/289 (2020.01); G06F 16/31 (2019.01); G06F 16/35 (2019.01)
CPC G06F 40/289 (2020.01) [G06F 16/31 (2019.01); G06F 16/35 (2019.01)] 17 Claims
OG exemplary drawing
 
1. A method performed by a system of one or more computers, comprising:
obtaining a request including an input sequence of words; and
obtaining a predicted next word for the input sequence of words from a machine-learned model, the machine-learned model being configured to take an author identifier identifying the author and the input sequence of words as input, the predicted next word being provided as a response to the request;
wherein the machine-learned model includes:
an encoder neural network configured to take as input the input sequence of words and output an alternative representation of the input sequence, and
a combining layer configured to take as input the alternative representation of the input sequence and an author vector generated based on the author identifier and output a combined representation of the input sequence.