公开(公告)号：US20200320435A1

公开(公告)日：2020-10-08

申请号：US16842265

申请日：2020-04-07

Applicant: SRI International

Inventor： Pedro Daniel Barbosa Sequeira ,  Melinda T. Gervasio ,  Chih-hung Yeh

IPC: G06N20/00 ,  G06N5/04

Abstract: Techniques are disclosed for applying a multi-level introspection framework to interaction data characterizing a history of interaction of a reinforcement learning agent with an environment. The framework may apply statistical analysis and machine learning methods to interaction data collected during the RL agent's interaction with the environment. The framework may include a first (“environment”) level that analyzes characteristics of one or more tasks to be solved by the RL agent to generate elements, a second (“interaction”) level that analyzes actions of the RL agent when interacting with the environment to generate elements, and a third (“meta-analysis”) level that generates elements by analyzing combinations of elements generated by the first level and elements generated by the second level.

公开(公告)号：US20240338569A1

公开(公告)日：2024-10-10

申请号：US18504923

申请日：2023-11-08

Applicant: SRI International

Inventor： Pedro Daniel Barbosa Sequeira ,  Melinda T. Gervasio

IPC: G06N3/092

CPC classification number: G06N3/092

Abstract: In an example, a method includes, collecting interaction data comprising one or more interactions between one or more Reinforcement Learning (RL) agents and an environment; analyzing interestingness of the interaction data along one or more interestingness dimensions; determining competency of the one or more RL agents along the one or more interestingness dimensions based on the interestingness of the interaction data; and outputting an indication of the competency of the one or more RL agents.

公开(公告)号：US11941012B2

公开(公告)日：2024-03-26

申请号：US16443618

申请日：2019-06-17

Applicant: SRI International

Inventor： Karen L. Myers ,  Melinda T. Gervasio

IPC: G06F7/02 ,  G06F16/00 ,  G06F16/23 ,  G06F16/2458 ,  G06F16/9035

CPC classification number: G06F16/2465 ,  G06F16/2365 ,  G06F16/9035

Abstract: In general, the disclosure describes techniques for identifying sequences of user actions from event data and logs of user actions for at least one user of a computing system. In one example, a system includes a sequence mining unit that processes event data and logs of user actions for at least one user of a computing system to obtain a set of one or more candidate action sequences each comprising a sequence of one or more user actions. A sequence filtering unit of the system applies, to the set of one or more candidate action sequences, one or more filters informed by a model of user actions for an application domain to obtain a set of one or more filtered action sequences to improve a quality of action sequences identified by the system. An output device of the system outputs an indication of the set of one or more filtered action sequences usable for generating at least one automated workflow or information usable for improving a workflow.

公开(公告)号：US20200233865A1

公开(公告)日：2020-07-23

申请号：US16443618

申请日：2019-06-17

Applicant: SRI International

Inventor： Karen L. Myers ,  Melinda T. Gervasio

IPC: G06F16/2458 ,  G06F16/23 ,  G06F16/9035

Abstract: In general, the disclosure describes techniques for identifying sequences of user actions from event data and logs of user actions for at least one user of a computing system. In one example, a system includes a sequence mining unit that processes event data and logs of user actions for at least one user of a computing system to obtain a set of one or more candidate action sequences each comprising a sequence of one or more user actions. A sequence filtering unit of the system applies, to the set of one or more candidate action sequences, one or more filters informed by a model of user actions for an application domain to obtain a set of one or more filtered action sequences to improve a quality of action sequences identified by the system. An output device of the system outputs an indication of the set of one or more filtered action sequences usable for generating at least one automated workflow or information usable for improving a workflow.

公开(公告)号：US11597394B2

公开(公告)日：2023-03-07

申请号：US16222623

申请日：2018-12-17

Applicant: SRI International

Inventor： Chih-hung Yeh ,  Boone Adkins ,  Melinda T. Gervasio ,  Karen L. Myers ,  Rodrigo de Salvo Braz

IPC: G06N20/00 ,  B60W40/09 ,  G05D1/00 ,  G07C5/08 ,  G05D1/02

Abstract: In general, the disclosure describes various aspects of techniques for evaluating decisions determined by autonomous devices. A device comprising a memory and a processor may be configured to perform the techniques. The memory may store first state data representative of a first observational state detected by an autonomous device, and first action data representative of one or more first actions the autonomous device performs responsive to detecting the first observational state. The processor may execute a computation engine configured to identify, based on the first action data, a first inflection point representative of changing behavior of the autonomous device. The computation engine may further be configured to determine, based on the first inflection point, first explanatory data representative of portions of the first state data on which the autonomous device relied that explain the changing behavior of the autonomous device, and output the first explanatory data.

公开(公告)号：US11568246B2

公开(公告)日：2023-01-31

申请号：US16810324

申请日：2020-03-05

Applicant: SRI International

Inventor： Chih-hung Yeh ,  Melinda T. Gervasio ,  Karen L. Myers ,  Daniel J. Sanchez ,  Matthew Crossley

IPC: G06F40/30 ,  G06N3/08 ,  G06N3/04

Abstract: Techniques are disclosed for training a machine learning model to perform actions within an environment. In one example, an input device receives a declarative statement. A computation engine selects, based on the declarative statement, a template that includes a template action performable within the environment. The computation engine generates, based on the template, synthetic training episodes. The computation engine further generates experiential training episodes, each experiential training episode collected by a machine learning model from past actions performed by the machine learning model. Each synthetic training episode and experiential training episode comprises an action and a reward. A machine learning system trains, with the synthetic training episodes and the experiential training episodes, the machine learning model to perform the actions within the environment.