Deepmind technologies limited (20240185070). TRAINING ACTION SELECTION NEURAL NETWORKS USING LOOK-AHEAD SEARCH simplified abstract
Contents
TRAINING ACTION SELECTION NEURAL NETWORKS USING LOOK-AHEAD SEARCH
Organization Name
Inventor(s)
Julian Schrittwieser of London (GB)
TRAINING ACTION SELECTION NEURAL NETWORKS USING LOOK-AHEAD SEARCH - A simplified explanation of the abstract
This abstract first appeared for US patent application 20240185070 titled 'TRAINING ACTION SELECTION NEURAL NETWORKS USING LOOK-AHEAD SEARCH
Simplified Explanation
The patent application describes methods, systems, and apparatus for training an action selection neural network.
- Receiving an observation characterizing the current state of the environment.
- Determining a target network output by performing a look ahead search of possible future states of the environment.
- Selecting an action for the agent based on the target network output.
- Storing the target network output in an exploration history data store for updating network parameters.
Potential Applications: - Autonomous vehicles - Robotics - Gaming industry
Problems Solved: - Enhancing decision-making processes in complex environments - Improving efficiency and accuracy of action selection
Benefits: - Increased performance and adaptability - Enhanced learning capabilities - Optimized decision-making processes
Commercial Applications: - Autonomous driving systems - Real-time strategy games - Industrial automation
Prior Art: No specific information on prior art related to this technology is provided in the abstract.
Frequently Updated Research: There may be ongoing research in the field of reinforcement learning and neural networks that could be relevant to this technology.
Questions about the technology: Question 1: How does the look ahead search process improve decision-making in the neural network? Answer: The look ahead search allows the network to anticipate future states and make more informed decisions based on potential outcomes.
Question 2: What are the key advantages of using an exploration history data store in updating network parameters? Answer: Storing target network outputs in the exploration history data store helps in refining the network parameters over time based on past observations and actions.
Original Abstract Submitted
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.
