Collaborative Symmetricity Exploitation for Offline Learning of Hardware Design SolverDownload PDF

Anonymous

22 Sept 2022, 12:35 (modified: 18 Nov 2022, 08:06)ICLR 2023 Conference Blind SubmissionReaders: Everyone
Keywords: Symmetricity, Offline learning, Hardware design solver
TL;DR: We propose offline learning method with symmetric learning for hardware design
Abstract: This paper proposes \textit{collaborative symmetricity exploitation} (CSE) framework to train a solver for the decoupling capacitor placement problem (DPP) benchmark, one of the significant hardware design problems. Due to the sequentially coupled multi-level property of the hardware design process, the design condition of DPP changes depending on the design of higher-level problems. Also, the online evaluation of real-world electrical performance through simulation is extremely costly. Thus, a data-efficient offline learning method to train a solver (i.e., contextualized policy) with high generalization capability over changing task conditions is necessary. In this paper, we apply the CSE framework to train a DPP solver using a limited number of offline expert data. Leveraging the symmetricity for offline learning of hardware design solver has two major advantages: it increases data-efficiency by reducing the solution space and improves generalization capability by capturing the invariant nature present regardless of changing conditions. The proposed CSE is composed of two learning schemes: expert exploitation and self-exploitation. Expert exploitation induces symmetricity during the imitation learning process with offline expert data and self-exploitation induces symmetricity during the consistency learning process with self-generated data. Extensive experiments verified that CSE with zero-shot inference outperforms the neural baselines and iterative conventional design methods on the DPP benchmark. Furthermore, CSE showed promising extrapolation capability as it greatly outperforms the expert method used to generate the offline data for training. Scalability and flexibility of the proposed method were also verified for practical use of CSE in industry.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics
Submission Guidelines: Yes
Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning
11 Replies

Loading