Invention Application
- Patent Title: MULTI-TASK NEURAL NETWORK SYSTEMS WITH TASK-SPECIFIC POLICIES AND A SHARED POLICY
-
Application No.: US17486842Application Date: 2021-09-27
-
Publication No.: US20220083869A1Publication Date: 2022-03-17
- Inventor: Razvan Pascanu , Raia Thais Hadsell , Victor Constant Bapst , Wojciech Czarnecki , James Kirkpatrick , Yee Whye Teh , Nicolas Manfred Otto Heess
- Applicant: DeepMind Technologies Limited
- Applicant Address: GB London
- Assignee: DeepMind Technologies Limited
- Current Assignee: DeepMind Technologies Limited
- Current Assignee Address: GB London
- Main IPC: G06N3/08
- IPC: G06N3/08 ; G06N3/10 ; G06N5/04

Abstract:
A method is proposed for training a multitask computer system, such as a multitask neural network system. The system comprises a set of trainable workers and a shared module. The trainable workers and shared module are trained on a plurality of different tasks, such that each worker learns to perform a corresponding one of the tasks according to a respective task policy, and said shared policy network learns a multitask policy which represents common behavior for the tasks. The coordinated training is performed by optimizing an objective function comprising, for each task: a reward term indicative of an expected reward earned by a worker in performing the corresponding task according to the task policy; and at least one entropy term which regularizes the distribution of the task policy towards the distribution of the multitask policy.
Public/Granted literature
- US11983634B2 Multi-task neural network systems with task-specific policies and a shared policy Public/Granted day:2024-05-14
Information query