Novel Upper Bounds for the Constrained Most Probable Explanation Task

Authors: Tahrima Rahman, Sara Rouhani, Vibhav Gogate

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our proposed scheme along several dimensions including quality of the bounds and computation time required on various benchmark graphical models and how it can be used to find heuristic, near-optimal feasible solutions in an example application pertaining to robust estimation and adversarial attacks on classifiers. Empirically, we investigate the quality and computational efficiency of our proposed bounding techniques on a variety of CMPE tasks formulated on cutset networks [30] and graphical models used in the UAI competitions [13, 15]. We conducted experiments on the MNIST dataset [22] using the univariate and grid distance functions described above and linear support vector machine as the classifier.
Researcher Affiliation Academia Tahrima Rahman Sara Rouhani Vibhav Gogate The University of Texas at Dallas {tahrima.rahman, sara.rouhani, vibhav.gogate@utdallas.edu}
Pseudocode Yes Algorithm 1: UB-MPE (M1, M2,q) and Algorithm 2: UB-KP (M1, M2,q,i)
Open Source Code No The paper does not explicitly state that source code for the described methodology is released or provide a link to a code repository.
Open Datasets Yes We learned cutset networks on four high-dimensional datasets AD, BBC, Book and Reuters. We used these networks as M1... We also experimented with several high treewidth networks used in the UAI competitions [13, 15]... We conducted experiments on the MNIST dataset [22] using the univariate and grid distance functions described above and linear support vector machine as the classifier.
Dataset Splits No The paper states: "The dataset has 70,000 examples. We used 70% of the examples for training and 30% for testing." It does not explicitly mention a separate validation split.
Hardware Specification No The paper does not provide specific details regarding the hardware used for running experiments, such as GPU/CPU models, memory, or cloud computing instance types.
Software Dependencies No The paper does not provide specific version numbers for software dependencies, such as programming languages, libraries, or solvers used in the experiments.
Experiment Setup Yes We ran JG for a maximum of 40 iterations or until convergence. We ran the outer-loop of the MPE-based as well as MCKP-based bounding algorithms for a maximum of 100 iterations or until convergence. For each network, we experimented with 5 values of q such that the chosen values lie roughly in the 10-th, 20-th, 50-th, 80-th and 90-th percentile respectively. In our experiments, we used diminishing step size which reduces α by periodically dividing it by a positive constant.