Optimizing Solution-Samplers for Combinatorial Problems: The Landscape of Policy-Gradient Method

Authors: Constantine Caramanis, Dimitris Fotakis, Alkis Kalavasis, Vasilis Kontonis, Christos Tzamos

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we investigate experimentally the effect of our main theoretical contributions, the entropy regularizer (see Equation (2)) and the fast/slow mixture scheme (see Equation (5)).
Researcher Affiliation Academia Constantine Caramanis UT Austin & Archimedes / Athena RC constantine@utexas.edu Dimitris Fotakis NTUA & Archimedes / Athena RC fotakis@cs.ntua.gr Alkis Kalavasis Yale University alvertos.kalavasis@yale.edu Vasilis Kontonis UT Austin vkonton@gmail.com Christos Tzamos UOA & Archimedes / Athena RC tzamos@wisc.edu
Pseudocode No The paper includes a Python class definition in the appendix (Figure 4), but it is not explicitly labeled as "Pseudocode" or an "Algorithm" block.
Open Source Code Yes For more details we refer to our full code submitted in the supplementary material.
Open Datasets No The paper states: "We generate 100 random 𝐺(𝑛, 𝑝) graphs with 𝑛= 15 nodes and 𝑝= 0.5" and refers to "random 𝑑-regular graphs with 𝑛nodes". These are generated graphs, not specific publicly available datasets with direct access information or formal citations.
Dataset Splits No The paper describes generating graphs and training models but does not provide specific details on train/validation/test splits, such as percentages, sample counts, or predefined split references.
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU types, or cloud computing instance specifications used for the experiments.
Software Dependencies No The paper mentions a "pytorch implementation" but does not specify version numbers for PyTorch or any other software libraries or dependencies used in the experiments.
Experiment Setup Yes We perform 600 iterations and, for the entropy regularization, we progressively decrease the regularization weight, starting from 10, and dividing it by 2 every 60 iterations. We used a fast/slow mixing with mixture probability 0.2 and inverse temperature rho=0.03