FIELD: information technology.
SUBSTANCE: invention relates to strategy search in strategic interaction between two or more parties. One method includes: storing a plurality of regret samples in a first data storage, wherein a plurality of regret samples are obtained in two or more iterations of the CRF algorithm when searching for a strategy in strategic interaction between two or more parties; storing a plurality of policy samples in a second data store; updating the parameters of the first neural network to predict the possible action regression value in the side state based on the plurality of regret samples in the first data storage; and updating parameters of the second neural network to predict the possible action strategy value in the side state based on the plurality of policy samples in the second data storage.
EFFECT: technical result is higher rate of CFR algorithm convergence.
22 cl, 12 dwg
Title | Year | Author | Number |
---|---|---|---|
ATTRIBUTING USER ACTION BASED ON BIOMETRIC IDENTITY | 2014 |
|
RU2668984C2 |
ADAPTIVE AUDIO ENHANCEMENT FOR MULTICHANNEL SPEECH RECOGNITION | 2016 |
|
RU2698153C1 |
MODEL-BASED PLAYFIELD REGISTRATION | 2011 |
|
RU2480832C2 |
TRAINING OF DNN-STUDENT BY MEANS OF OUTPUT DISTRIBUTION | 2014 |
|
RU2666631C2 |
RECOGNIZING OF MIXED SPEECH | 2015 |
|
RU2686589C2 |
DISTRIBUTED NETWORK ARCHITECTURE FOR INPUTTING DYNAMIC INFORMATION CONTENT INTO SYNTHESISED MEDIUM | 2008 |
|
RU2468847C2 |
DISTRIBUTED NETWORK ARCHITECTURE FOR INTRODUCTION OF DYNAMIC INFORMATION CONTENT INTO SYNTHESISED ENVIRONMENT | 2008 |
|
RU2468431C2 |
ORTHOGONAL CODING SOURCE AND RECEIVER | 2013 |
|
RU2612896C2 |
DEVICE FOR REMOTELY PLAYING CHESS | 2021 |
|
RU2761770C1 |
METHOD AND SERVER FOR WAVEFORM GENERATION | 2021 |
|
RU2803488C2 |
Authors
Dates
2021-02-20—Published
2019-01-17—Filed