US 11,727,270 B2
Cross data set knowledge distillation for training machine learning models
Ji Li, San Jose, CA (US); Amit Srivastava, San Jose, CA (US); Xingxing Zhang, Beijing (CN); Furu Wei, Beijing (CN); and Ming Zhou, Beijing (CN)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed by MICROSOFT TECHNOLOGY LICENSING, LLC, Redmond, WA (US)
Filed on Feb. 24, 2020, as Appl. No. 16/799,091.
Prior Publication US 2021/0264106 A1, Aug. 26, 2021
Int. Cl. G06F 40/40 (2020.01); G06N 3/08 (2023.01); G06F 40/205 (2020.01); G06F 18/214 (2023.01); G10L 15/16 (2006.01); G10L 15/18 (2013.01); G06N 3/088 (2023.01); G06F 40/30 (2020.01)
CPC G06N 3/08 (2013.01) [G06F 18/2148 (2023.01); G06F 40/205 (2020.01); G06F 40/40 (2020.01); G06F 40/30 (2020.01); G06N 3/088 (2013.01); G10L 15/16 (2013.01); G10L 15/18 (2013.01)] 20 Claims
OG exemplary drawing
 
8. A method for training a text-to-content recommendation machine-learning (ML) model, the method comprising:
training a first ML model using a first training data set;
providing as input to the trained first ML model a set of unlabeled unordered training data to generate a transfer data set;
training a pretrained text analysis model using a labeled training data set;
executing the pretrained text analysis model to generate an output; and
utilizing the transfer data set and the output to train the text-to-content recommendation ML model for recommending content based on text,
wherein:
the text-to-content recommendation ML model is used by a text-to-content service provided by a server that receives a text portion as an input and provides the text portion to the text-to-content recommendation ML model and receives a plurality of recommendations for content that correspond with the text portion as an output of the text-to-content recommendation ML model, and
the server provides the plurality of recommendations for content to a backend unit that ranks the plurality of recommendations for presentation to a user.