Scaling Up AND/OR Abstraction Sampling
Authors: Kalev Kask, Bobak Pezeshki, Filjor Broka, Alexander Ihler, Rina Dechter
IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper, we introduce AOAS, a new Abstraction Sampling scheme on AND/OR search spaces that allow more flexible use of abstractions by circumventing the properness requirement. We analyze the properties of this new algorithm and, in an extensive empirical evaluation on five benchmarks, over 480 problems, and comparing against other state of the art algorithms, illustrate AOAS s properties and show that it provides a far more powerful and competitive Abstraction Sampling framework. |
| Researcher Affiliation | Academia | Kalev Kask, Bobak Pezeshki, Filjor Broka, Alexander Ihler and Rina Dechter University of California, Irvine, Irvine, CA 92697, USA {kkask, pezeshkb, fbroka, ihler, dechter}@ics.uci.edu |
| Pseudocode | Yes | Algorithm 1: AOAS. |
| Open Source Code | No | The paper provides a link to a general publications page (https://www.ics.uci.edu/~dechter/publications.html) and not a specific source code repository for the described methodology. |
| Open Datasets | Yes | We perform high-throughput experiments on over 480 problems from five well known benchmarks: DBN, Grids, Linkage-Type4, Pedigree, and Promedas. |
| Dataset Splits | No | The paper does not provide specific details on train/validation/test dataset splits. It mentions using 'an empirical estimate based on an average over 100 1hr of abstraction sampling' as a reference, but not formal data splits. |
| Hardware Specification | Yes | All experiments were run for 1 hr on a 2.66 GHz processor with 8 GB of memory (24 GB for the Linkage-Type4 benchmark). |
| Software Dependencies | No | The paper only states 'All algorithms were implemented in C++.' without specific version numbers for the compiler or any libraries. |
| Experiment Setup | Yes | We standardize our experiments by using i-bound 10. ... All experiments were run for 1 hr on a 2.66 GHz processor with 8 GB of memory (24 GB for the Linkage-Type4 benchmark). ... Algorithms were also tested against high performing state-of-the-art Dynamic Importance Sampling (DIS) [Lou et al., 2019] using an equal-time policy. |