Retrieval-Guided Reinforcement Learning for Boolean Circuit Minimization

Authors: Animesh Basak Chowdhury, Marco Romanelli, Benjamin Tan, Ramesh Karri, Siddharth Garg

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This study conducts a thorough examination of learning and search techniques for logic synthesis...Our findings showcase substantial enhancements...Furthermore, ABC-RL achieves an impressive up to 9x reduction in runtime...3 EMPIRICAL EVALUATION
Researcher Affiliation Academia Animesh Basak Chowdhury1 Marco Romanelli1 Benjamin Tan2 Ramesh Karri1 Siddharth Garg1 1 New York University 2 University of Calgary
Pseudocode Yes Algorithm 1 ABC-RL: Policy agent pre-training
Open Source Code No The paper provides a reproducibility statement but does not include any explicit statement about releasing their own code or a link to a code repository.
Open Datasets Yes We consider three datasets used by logic synthesis community: MCNC Yang (1991), EPFL arithmetic and EPFL random control benchmarks Amarú et al. (2015).
Dataset Splits Yes We randomly split the 56 total netlists obtained from all three benchmarks into 23 netlists for training 13 for validation (11 MCNC, 1 EPFL-arith, 1 EPFLrand) and remaining 20 for test (see Table 1).
Hardware Specification Yes We performed the training on a server machine with one NVIDIA RTX A4000 with 16GB VRAM.
Software Dependencies No The paper describes various methods and models (e.g., GCN, BERT, Adam), citing their original papers, but it does not specify the version numbers for software libraries or dependencies used in the implementation.
Experiment Setup Yes Agents are trained for 50 epochs using Adam with an initial learning rate of 0.01. In each training epoch, we perform MCTS on all netlists with an MCTS search budget K = 512 per synthesis level...We set T = 100 and δth = 0.007 based on our validation data.