System and Method for Multi-task Learning

Patent metadata

Field	Value
Publication number	US-2023222340-A1
Application number	US-202318175262-A
Country	US
Kind code	A1
Filing date	Feb 27, 2023
Priority date	Mar 1, 2019
Publication date	Jul 13, 2023
Grant date	—

Abstract

A method of generating a multi-modal prediction is disclosed herein. A computing system retrieves event data from a data store. The event data includes information for a plurality of events across a plurality of seasons. Computing system generates a predictive model using a mixture density network, by generating an input vector from the event data learning, by the mixture density network, a plurality of values associated with a next play following each play in the event data. The mixture density network is trained to output the plurality of values near simultaneously. Computing system receives a set of event data directed to an event in a match. The set of event data includes information directed to at least playing surface position and current score. Computing system generates, via the predictive model, a plurality of values associated with a next event following the event based on the set of event data.

First claim

What is claimed:

1 . A method of generating a multi-modal prediction, comprising:
identifying, by a computing system, information related to a sporting event, the information comprising dense features of the sporting event and sparse features of the sporting event; generating, by the computing system, dense representations of the sparse features using one or more embedding layers of a machine learning architecture; generating, by the computing system, an input vector comprising the dense features and the dense representations of the sparse features; simultaneously generating, by the computing system using a mixture density layer of the machine learning architecture, a plurality of values associated with a next event to occur based on the input vector; and outputting, by the computing system, a graphical user interface comprising graphical representations of the plurality of values.

2 . The method of claim 1 , wherein the dense features comprise a current score, time remaining, and ball position.

3 . The method of claim 1 , wherein the sparse features comprise ball position, play number, team identifiers, and season identifiers.

4 . The method of claim 3 , wherein generating, by the computing system, the dense representations of the sparse features using the one or more embedding layers of the machine learning architecture comprises:
generating a first dense representation of the ball position by passing the ball position through a first embedding layer; generating a second dense representation of the play number by passing the play number through a second embedding layer; generating a third dense representation of the team identifiers by passing the team identifiers through a third embedding layer; and generating a fourth dense representation of the season identifiers by passing the season identifiers through a fourth embedding layer.

5 . The method of claim 1 , wherein the plurality of values comprises two or more of predictions for expected meters or yards, expected try tackles, expected try set, win probability, or final score line.

6 . The method of claim 1 , wherein generating, by the computing system, the input vector comprising the dense features and the dense representations of the sparse features comprises:
concatenating the dense representations of the sparse features with the dense features.

7 . The method of claim 1 , wherein simultaneously generating, by the computing system using the mixture density layer of the machine learning architecture, the plurality of values associated with the next event to occur based on the input vector comprises:
generating an output vector comprising each of the plurality of values.

8 . A non-transitory computer readable medium comprising one or more sequences of instructions, which, when executed by one or more processors, causes a computing system to perform operations comprising:
identifying, by the computing system, information related to a sporting event, the information comprising dense features of the sporting event and sparse features of the sporting event; generating, by the computing system, dense representations of the sparse features using one or more embedding layers of a machine learning architecture; generating, by the computing system, an input vector comprising the dense features and the dense representations of the sparse features; simultaneously generating, by the computing system using a mixture density layer of the machine learning architecture, a plurality of values associated with a next event to occur based on the input vector; and outputting, by the computing system, a graphical user interface comprising graphical representations of the plurality of values.

9 . The non-transitory computer readable medium of claim 8 , wherein the dense features comprise a current score, time remaining, and ball position.

10 . The non-transitory computer readable medium of claim 8 , wherein the sparse features comprise ball position, play number, team identifiers, and season identifiers.

11 . The non-transitory computer readable medium of claim 10 , wherein generating, by the computing system, the dense representations of the sparse features using the one or more embedding layers of the machine learning architecture comprises:
generating a first dense representation of the ball position by passing the ball position through a first embedding layer; generating a second dense representation of the play number by passing the play number through a second embedding layer; generating a third dense representation of the team identifiers by passing the team identifiers through a third embedding layer; and generating a fourth dense representation of the season identifiers by passing the season identifiers through a fourth embedding layer.

12 . The non-transitory computer readable medium of claim 8 , wherein the plurality of values comprises two or more of predictions for expected meters or yards, expected try tackles, expected try set, win probability, or final score line.

13 . The non-transitory computer readable medium of claim 8 , wherein generating, by the computing system, the input vector comprising the dense features and the dense representations of the sparse features comprises:
concatenating the dense representations of the sparse features with the dense features.

14 . The non-transitory computer readable medium of claim 8 , wherein simultaneously generating, by the computing system using the mixture density layer of the machine learning architecture, the plurality of values associated with the next event to occur based on the input vector comprises:
generating an output vector comprising each of the plurality of values.

15 . A system, comprising:
a processor; and a memory having programming instructions stored thereon, which, when executed by the processor, causes the system to perform operations comprising: identifying information related to a sporting event, the information comprising dense features of the sporting event and sparse features of the sporting event; generating dense representations of the sparse features using one or more embedding layers of a machine learning architecture; generating an input vector comprising the dense features and the dense representations of the sparse features; simultaneously generating, using a mixture density layer of the machine learning architecture, a plurality of values associated with a next event to occur based on the input vector; and outputting a graphical user interface comprising graphical representations of the plurality of values.

16 . The system of claim 15 , wherein the dense features comprise a current score, time remaining, and ball position.

17 . The system of claim 15 , wherein the sparse features comprise ball position, play number, team identifiers, and season identifiers.

18 . The system of claim 17 , wherein generating the dense representations of the sparse features using the one or more embedding layers of the machine learning architecture comprises:
generating a first dense representation of the ball position by passing the ball position through a first embedding layer; generating a second dense representation of the play number by passing the play number through a second embedding layer; generating a third dense representation of the team identifiers by passing the team identifiers through a third embedding layer; and generating a fourth dense representation of the season identifiers by passing the season identifiers through a fourth embedding layer.

Assignees

• Stats Llc

Inventors

• Holbrook Matthew
• Hobbs Jennifer
• Lucey Patrick Joseph

Classifications

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• A63F13/812

  Ball games, e.g. soccer or baseball · CPC title