Processing

Please wait...

Settings

Settings

Goto Application

1. WO2018215665 - TRAINING ACTION SELECTION NEURAL NETWORKS USING LOOK-AHEAD SEARCH

Publication Number WO/2018/215665
Publication Date 29.11.2018
International Application No. PCT/EP2018/063869
International Filing Date 28.05.2018
IPC
G06N 3/00 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
G06N 3/04 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
04Architecture, e.g. interconnection topology
G06N 3/08 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
G06N 5/00 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
5Computer systems using knowledge-based models
CPC
G06N 3/006
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
004Artificial life, i.e. computers simulating life
006based on simulated virtual individual or collective life forms, e.g. single "avatar", social simulations, virtual worlds or particle swarm optimisation
G06N 3/0472
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
04Architectures, e.g. interconnection topology
0472using probabilistic elements, e.g. p-rams, stochastic processors
G06N 3/08
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
G06N 5/003
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
5Computer systems using knowledge-based models
003Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
G06N 7/005
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
7Computer systems based on specific mathematical models
005Probabilistic networks
Applicants
  • DEEPMIND TECHNOLOGIES LIMITED [GB]/[GB]
Inventors
  • SIMONYAN, Karen
  • SILVER, David
  • SCHRITTWIESER, Julian
Agents
  • KUNZ, Herbert
Priority Data
62/511,94526.05.2017US
Publication Language English (EN)
Filing Language English (EN)
Designated States
Title
(EN) TRAINING ACTION SELECTION NEURAL NETWORKS USING LOOK-AHEAD SEARCH
(FR) RÉSEAUX NEURONAUX DE SÉLECTION D'ACTION D'APPRENTISSAGE UTILISANT UNE RECHERCHE ANTICIPÉE
Abstract
(EN)
Methods, systems and apparatus, including computer programs encoded on computer storage media, for training an action selection neural network. One of the methods includes receiving an observation characterizing a current state of the environment; determining a target network output for the observation by performing a look ahead search of possible future states of the environment starting from the current state until the environment reaches a possible future state that satisfies one or more termination criteria, wherein the look ahead search is guided by the neural network in accordance with current values of the network parameters; selecting an action to be performed by the agent in response to the observation using the target network output generated by performing the look ahead search; and storing, in an exploration history data store, the target network output in association with the observation for use in updating the current values of the network parameters.
(FR)
L'invention concerne des procédés, des systèmes et un appareil, y compris des programmes informatiques codés sur un support de stockage informatique, pour l'apprentissage d'un réseau neuronal de sélection d'action. L'un des procédés consiste à recevoir une observation caractérisant un état actuel de l'environnement ; à déterminer une sortie de réseau cible pour l'observation par réalisation d'une recherche anticipée d'états futurs possibles de l'environnement à partir de l'état actuel jusqu'à ce que l'environnement atteigne un état futur possible qui satisfait un ou plusieurs critères de terminaison, la recherche anticipée étant guidée par le réseau neuronal en fonction des valeurs actuelles des paramètres de réseau ; à sélectionner une action à exécuter par l'agent en réponse à l'observation à l'aide de la sortie de réseau cible générée par la réalisation de la recherche anticipée ; et à stocker, dans une mémoire de données d'historique d'exploration, la sortie de réseau cible en association avec l'observation pour une utilisation dans la mise à jour des valeurs actuelles des paramètres de réseau.
Also published as
Latest bibliographic data on file with the International Bureau