公开(公告)号：US11620576B1

公开(公告)日：2023-04-04

申请号：US16908359

申请日：2020-06-22

Applicant: Amazon Technologies, Inc.

Inventor： Yunzhe Tao ,  Sahika Genc ,  Tao Sun ,  Sunil Mallya Kasaragod

IPC: G06N20/00 ,  G06N5/02 ,  G06N5/045 ,  G06N20/20

Abstract: A training system may create and train a machine learning model with knowledge transfer. The knowledge transfer may transfer knowledge that is acquired by another machine learning model that has been previously trained to the machine learning model that is under training. The knowledge transfer may include a combination of representation transfer and instance transfer, the two of which may be performed alternatingly. The instance transfer may further include a filter mechanism to selectively identify instances with a satisfactory performance to implement the knowledge transfer.

公开(公告)号：US12026598B2

公开(公告)日：2024-07-02

申请号：US18193023

申请日：2023-03-30

Applicant: Amazon Technologies, Inc.

Inventor： Yunzhe Tao ,  Sahika Genc ,  Tao Sun ,  Sunil Mallya Kasaragod

IPC: G06N20/00 ,  G06N5/045 ,  G06N20/20

CPC classification number: G06N20/00 ,  G06N5/045 ,  G06N20/20

Abstract: A training system may create and train a machine learning model with knowledge transfer. The knowledge transfer may transfer knowledge that is acquired by another machine learning model that has been previously trained to the machine learning model that is under training. The knowledge transfer may include a combination of representation transfer and instance transfer, the two of which may be performed alternatingly. The instance transfer may further include a filter mechanism to selectively identify instances with a satisfactory performance to implement the knowledge transfer.

公开(公告)号：US20230419113A1

公开(公告)日：2023-12-28

申请号：US18465916

申请日：2023-09-12

Applicant: Amazon Technologies, Inc.

Inventor： Sahika Genc ,  Sravan Babu Bodapati ,  Tao Sun ,  Sunil Mallya Kasaragod

IPC: G06N3/08 ,  G06F17/16 ,  G06F16/904 ,  G05D1/02

CPC classification number: G06N3/08 ,  G06F17/16 ,  G06F16/904 ,  G05D1/0221 ,  G05D2201/0213

Abstract: A data source configured to provide a representation of an environment of one or more agents is identified. Using a data set obtained from the data source, a neural network-based reinforcement learning model with one or more attention layers is trained. Importance indicators generated by the attention layers are used to identify actions to be initiated by an agent. A trained version of the model is stored.

公开(公告)号：US12265924B1

公开(公告)日：2025-04-01

申请号：US16908486

申请日：2020-06-22

Applicant: Amazon Technologies, Inc.

Inventor： Tao Sun ,  Yunzhe Tao ,  Sahika Genc ,  Sunil Mallya Kasaragod ,  Kaiqing Zhang

IPC: G06N5/04 ,  G06N5/043 ,  G06N20/00

Abstract: Techniques for robust multi-agent reinforcement learning (MARL) are described. An exemplary method includes initializing a plurality of parameters for a plurality of agents including at least policy parameters and action-value (Q) parameters; performing robust multi-agent reinforcement learning to learn polices for the agents, wherein in the learned polices no agent has an incentive to deviate, the agents include an implicit agent that is to select a worst-case at any given time during the learning process; and at least one agent utilizing its learned policy.

公开(公告)号：US11900244B1

公开(公告)日：2024-02-13

申请号：US16588789

申请日：2019-09-30

Applicant: Amazon Technologies, Inc.

Inventor： Sahika Genc ,  Sravan Babu Bodapati ,  Tao Sun ,  Sunil Mallya Kasaragod

IPC: G06N3/08 ,  G05D1/02 ,  G06F17/16 ,  G06F16/904

CPC classification number: G06N3/08 ,  G05D1/0221 ,  G06F16/904 ,  G06F17/16 ,  G05D2201/0213

Abstract: A data source configured to provide a representation of an environment of one or more agents is identified. Using a data set obtained from the data source, a neural network-based reinforcement learning model with one or more attention layers is trained. Importance indicators generated by the attention layers are used to identify actions to be initiated by an agent. A trained version of the model is stored.

公开(公告)号：US20230252355A1

公开(公告)日：2023-08-10

申请号：US18193023

申请日：2023-03-30

Applicant: Amazon Technologies, Inc.

Inventor： Yunzhe Tao ,  Sahika Genc ,  Tao Sun ,  Sunil Mallya Kasaragod

IPC: G06N20/00 ,  G06N5/045 ,  G06N20/20

CPC classification number: G06N20/00 ,  G06N5/045 ,  G06N20/20

Abstract: A training system may create and train a machine learning model with knowledge transfer. The knowledge transfer may transfer knowledge that is acquired by another machine learning model that has been previously trained to the machine learning model that is under training. The knowledge transfer may include a combination of representation transfer and instance transfer, the two of which may be performed alternatingly. The instance transfer may further include a filter mechanism to selectively identify instances with a satisfactory performance to implement the knowledge transfer.