Learning the Finer Things: Bayesian Structure Learning at the Instantiation Level

Authors: Chase Yakaboski, Eugene Santos, Jr

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental By leveraging Bayesian Knowledge Bases (BKBs), a framework that operates at the instantiation level and inherently subsumes Bayesian Networks (BNs), we develop both a theoretical MDL score and associated structure learning algorithm that demonstrates significant improvements over learned BNs on 40 benchmark datasets. Further, our algorithm incorporates recent off-the-shelf DAG learning techniques enabling tractable results even on large problems. We then demonstrate the utility of our approach in a significantly under-determined domain by learning gene regulatory networks on breast cancer gene mutational data available from The Cancer Genome Atlas (TCGA).
Researcher Affiliation Academia Thayer School of Engineering at Dartmouth College, Hanover, NH {chase.th, esj}@dartmouth.edu
Pseudocode Yes Algorithm 1: BKB Structure Learning Input: Dataset D, Source Reliabilities R, DAG learning algorithm f and hyperparameters Θ 1: K 2: for τ D = do 3: Gτ f(τ, R, Θ) 4: K K {Gτ} 5: end for 6: return BKB-Fusion(K, R)
Open Source Code Yes For source code visit: https://github.com/di2ag/pybkb.
Open Datasets Yes We then demonstrate the utility of our approach in a significantly under-determined domain by learning gene regulatory networks on breast cancer gene mutational data available from The Cancer Genome Atlas (TCGA). (Tomczak, Czerwińska, and Wiznerowicz 2015)
Dataset Splits Yes We performed a 10-fold classification cross validation on a subset of only 22 datasets due to the increased learning and reasoning time incurred by running cross validation analysis.
Hardware Specification No No specific hardware details such as GPU/CPU models, memory, or cloud instance specifications used for running experiments are provided.
Software Dependencies No The paper mentions using 'GOBNILP' but does not specify any software names with version numbers for reproducibility.
Experiment Setup No The paper mentions 'hyperparameters Θ' in Algorithm 1 and refers to Appendix A for 'naming conventions and feature selection process', but does not explicitly detail specific hyperparameter values (e.g., learning rate, batch size) or other system-level training settings in the main text.