Deepmind Technologies Limited

Other

Gdm Holding LLC

Tom Schaul

David Silver

John Quan

Hado Philip Van Hasselt

Matteo Hessel

Koray Kavukcuoglu

Joao Ferdinando Gomes De Freitas

Volodymyr Mnih

Maxwell Elliot Jaderberg

Wojciech Czarnecki

Misha Man Ray Denil

Sergio Gomez Colmenarejo

Matthew William Hoffman

David Benjamin Pfau

Marcin Andrychowicz

Karen Simonyan

Erich Konrad Elsen

Simon Osindero

Karol Gregor

Mark Ring

William Clinton Dabney

Georg Ostrovski

Daniel George Horgan

Fengning Ding

Karel Lenc

Daniel J Mankowitz

Diana Luiza Borsa

David Szepesvari

Methods, systems, and apparatus, including computer programs encoded on computer storage media for training machine learning models. One method includes obtaining a machine learning model, wherein the machine learning model comprises one or more model parameters, and the machine learning model is trained using gradient descent techniques to optimize an objective function; determining an update rule for the model parameters using a recurrent neural network (RNN); and applying a determined update rule for a final time step in a sequence of multiple time steps to the model parameters.

Training machine learning models by determining update rules using neural networks

A method of training an action selection neural network for controlling an agent interacting with an environment to perform different tasks is described. The method includes obtaining a first trajectory of transitions generated while the agent was performing an episode of the first task from multiple tasks; and training the action selection neural network on the first trajectory to adjust the control policies for the multiple tasks. The training includes, for each transition in the first trajectory: generating respective policy outputs for the initial observation in the transition for each task in a subset of tasks that includes the first task and one other task; generating respective target policy outputs for each task using the reward in the transition, and determining an update to the current parameter values based on, for each task, a gradient of a loss between the policy output and the target policy output for the task.

Continual reinforcement learning with a multi-task agent

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for prediction of an outcome related to an environment. In one aspect, a system comprises a state representation neural network that is configured to: receive an observation characterizing a state of an environment being interacted with by an agent and process the observation to generate an internal state representation of the environment state; a prediction neural network that is configured to receive a current internal state representation of a current environment state and process the current internal state representation to generate a predicted subsequent state representation of a subsequent state of the environment and a predicted reward for the subsequent state; and a value prediction neural network that is configured to receive a current internal state representation of a current environment state and process the current internal state representation to generate a value prediction.

Environment prediction using reinforcement learning

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a neural network used to select actions performed by a reinforcement learning agent interacting with an environment. In one aspect, a method includes maintaining a replay memory, where the replay memory stores pieces of experience data generated as a result of the reinforcement learning agent interacting with the environment. Each piece of experience data is associated with a respective expected learning progress measure that is a measure of an expected amount of progress made in the training of the neural network if the neural network is trained on the piece of experience data. The method further includes selecting a piece of experience data from the replay memory by prioritizing for selection pieces of experience data having relatively higher expected learning progress measures and training the neural network on the selected piece of experience data.

Training neural networks using a prioritized experience memory

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for controlling an agent. One of the methods includes sampling a behavior modulation in accordance with a current probability distribution; for each of one or more time steps: processing an input comprising an observation characterizing a current state of the environment at the time step using an action selection neural network to generate a respective action score for each action in a set of possible actions that can be performed by the agent; modifying the action scores using the sampled behavior modulation; and selecting the action to be performed by the agent at the time step based on the modified action scores; determining a fitness measure corresponding to the sampled behavior modulation; and updating the current probability distribution over the set of possible behavior modulations using the fitness measure corresponding to the behavior modulation.

Modulating agent behavior to optimize learning progress

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a reinforcement learning system. The method includes: training an action selection policy neural network, and during the training of the action selection neural network, training one or more auxiliary control neural networks and a reward prediction neural network. Each of the auxiliary control neural networks is configured to receive a respective intermediate output generated by the action selection policy neural network and generate a policy output for a corresponding auxiliary control task. The reward prediction neural network is configured to receive one or more intermediate outputs generated by the action selection policy neural network and generate a corresponding predicted reward. Training each of the auxiliary control neural networks and the reward prediction neural network comprises adjusting values of the respective auxiliary control parameters, reward prediction parameters, and the action selection policy network parameters.

Reinforcement learning with auxiliary tasks

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. The neural network has a plurality of differentiable weights and a plurality of non-differentiable weights. One of the methods includes determining trained values of the plurality of differentiable weights and the non-differentiable weights by repeatedly performing operations that include determining an update to the current values of the plurality of differentiable weights using a machine learning gradient-based training technique and determining, using an evolution strategies (ES) technique, an update to the current values of a plurality of distribution parameters.

Learning non-differentiable weights of neural networks using evolutionary strategies

Training machine learning models by determining update rules using recurrent neural networks

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for reinforcement learning using goals and observations. One of the methods includes receiving an observation characterizing a current state of the environment; receiving a goal characterizing a target state from a set of target states of the environment; processing the observation using an observation neural network to generate a numeric representation of the observation; processing the goal using a goal neural network to generate a numeric representation of the goal; combining the numeric representation of the observation and the numeric representation of the goal to generate a combined representation; processing the combined representation using an action score neural network to generate a respective score for each action in the predetermined set of actions; and selecting the action to be performed using the respective scores for the actions in the predetermined set of actions.

Selecting reinforcement learning actions using goals and observations

Building a forecast for an autonomous agent at least comprises assigning a selected parameter of the autonomous agent to a scalar variable, adding a new policy to a set of policies where the new policy maps internal states of the autonomous agent to actions of the autonomous agent in which the mapping may optimize the scalar variable, and adding a new forecast to a set of forecasts where the forecast at least comprises a prediction regarding future values of the scalar variable following execution of the new policy, regardless whether the agent ever actually chooses to take actions in accordance with said new policy. A state of the autonomous agent may be evaluated following completion of each of the agent's actions by comparing the agent's state information with the predicted values of one or more forecasts. Whether to build an additional forecast may be determined based on the evaluation.

Method for creating predictive knowledge structures from experience in an artificial agent

Growing community of inventors

London, United Kingdom

Tom Schaul