Deepmind Technologies Limited

Google Inc.

Todd Andrew Hester

Michael C Ryan

Scott A McGaraghan

Nicolas Manfred Otto Heess

Allen Joseph Minich

Scott Ruffner

Jonathan Crimins

Amanda Sahl

Richard Andrew Evans

Jonathan Karl Scholz

William Greene

Thomas Lampe

Bilal Piot

Thomas Rothoerl

Gabriel Dulac-Arnold

Jim Gao

Wang Fumin

Praveen Subramani

Martin Riedmiller

Olivier Pietquin

Matej Vecerik

A method of training an action selection neural network to perform a demonstrated task using a supervised learning technique. The action selection neural network is configured to receive demonstration data comprising actions to perform the task and rewards received for performing the actions. The action selection neural network has auxiliary prediction task neural networks on one or more of its intermediate outputs. The action selection policy neural network is trained using multiple combined losses, concurrently with the auxiliary prediction task neural networks.

Training action-selection neural networks from demonstrations using multiple losses

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for improving operational efficiency within a data center by modeling data center performance and predicting power usage efficiency. An example method receives a state input characterizing a current state of a data center. For each data center setting slate, the state input and the data center setting slate are processed through an ensemble of machine learning models. Each machine learning model is configured to receive and process the state input and the data center setting slate to generate an efficiency score that characterizes a predicted resource efficiency of the data center if the data center settings defined by the data center setting slate are adopted t. The method selects, based on the efficiency scores for the data center setting slates, new values for the data center settings.

Optimizing data center controls using neural networks

A thermostat may include one or more memory devices comprising a stored setpoint schedule, one or more temperature sensors configured to provide temperature sensor measurements, and a processing system configured to be in operative communication the one or more memory devices to determine a setpoint temperature, and in still further operative communication with a heating, ventilation, and air conditioning (HVAC) system to control the HVAC system based at least in part on the setpoint temperature and the temperature sensor measurements. The processing system may be configured to control the HVAC system by receiving an indication that a total instantaneous energy usage rate for a structure in which the thermostat is installed is projected to exceed a threshold amount; and altering the stored setpoint schedule to reduce an energy usage rate of the HVAC system.

Architecture for thermostat control during peak intervals

A thermostat my include a stored setpoint schedule, temperature sensors providing temperature sensor measurements; and a processing system configured to control an HVAC system based at least in part on the setpoint temperature and the temperature sensor measurements. The processing system may be configured to control the HVAC system by receiving an indication of a first time interval, where energy is available to the HVAC system at a first rate during the first time interval, energy is available to the HVAC system at a second rate during a second time interval that is outside of the first time interval, and the first rate is higher than the second rate; identifying a first one or more setpoints in the plurality of setpoints of the stored setpoint schedule that occur in the first time interval; and decreasing a temperature component of at least one of the first one or more setpoints.

Architecture for thermostat control during time-of-use intervals

A thermostat may be programmed to operate during cold-weather months by operating in a heat mode where the thermostat periodically causes the HVAC system to activate a heating function to heat an enclosure based at least in part on the stored setpoint schedule; receiving an indication from a thermostat management server of a demand response event while the thermostat is operating in the heat mode; determining whether the thermostat stores a user setting indicating that the thermostat should preheat the enclosure prior to reaching a scheduled setpoint in the stored setpoint schedule; and preheating the enclosure prior to the demand response event based on a determination that the thermostat stores the user setting indicating that the thermostat should preheat the enclosure prior to reaching the scheduled setpoint in the stored setpoint schedule.

Thermostat algorithms and architecture for efficient operation at low temperatures

Growing community of inventors

London, United Kingdom

Todd Andrew Hester