Title:
SINGLE STEP DESIGN SPACE EXPLORATION SYSTEM
Document Type and Number:
WIPO Patent Application WO/2023/212751
Kind Code:
A3
Abstract:
A system and method for design space exploration where the design space includes a bounded set of dimensions. An intelligent agent is trained to produce optimal design configurations for a design space using feedback to a neural network. The feedback is provided by a design evaluator. The neural network consumes a constant input tensor and produces a policy as a set of conditional probability density functions for sampling each design dimension of the design space. Policy gradient algorithms in deep reinforcement learning evaluate the feedback to update the neural network. Multiple neural architectures may be utilized.
Inventors:
MORTAZAVI MASOOD (US)
YAN NING (US)
QIN TIANCHENG (US)
YAN NING (US)
QIN TIANCHENG (US)
Application Number:
PCT/US2023/070271
Publication Date:
December 07, 2023
Filing Date:
July 14, 2023
Export Citation:
Assignee:
FUTUREWEI TECHNOLOGIES INC (US)
International Classes:
G06F30/27; G06N3/02; G06F30/337; G06F115/02
Domestic Patent References:
| WO2022147583A2 | 2022-07-07 | |||
| WO2018128653A1 | 2018-07-12 |
Foreign References:
| US20220108054A1 | 2022-04-07 |
Attorney, Agent or Firm:
VIERRA, Larry E. (US)
Download PDF: