Reconstruction on Trees and Low-Degree Polynomials

Authors: Frederic Koehler, Elchanan Mossel

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Figure 1, we test kernel ridge regression in a simulation in both the case λ2 = 0 and λ2 > 0: consistent with our result, KRR fails to beat the null risk when λ2 = 0; interestingly, it also fails for moderately small values of λ2 as well, which is related to the Open Problem we discuss later. In the figure, KRR is performed using 2000 i.i.d. samples of (x, y) pairs with x the one-hot encoded leaf colorations and y the centered indicator that the root color is 1, as in Theorem 8.
Researcher Affiliation Academia Frederic Koehler Stanford University fkoehler@stanford.edu Elchanan Mossel Massachusetts Institute of Technology elmos@mit.edu
Pseudocode No The paper discusses various algorithms and estimators (e.g., Belief Propagation, simple recursive estimator) but does not provide any structured pseudocode or algorithm blocks.
Open Source Code Yes 3. If you ran experiments... (a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]
Open Datasets No In the figure, KRR is performed using 2000 i.i.d. samples of (x, y) pairs with x the one-hot encoded leaf colorations and y the centered indicator that the root color is 1, as in Theorem 8. This indicates the data was generated for the simulation rather than being a publicly available dataset with a specific name, link, or citation.
Dataset Splits No Bandwidth and ridge penalty are selected via grid search on a validation set. While a validation set is mentioned, the specific split percentages or sample counts for it are not provided.
Hardware Specification No 3. If you ran experiments... (d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [N/A] Only used trivial resources.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions) that would be needed for reproducibility.
Experiment Setup Yes In the figure, KRR is performed using 2000 i.i.d. samples of (x, y) pairs with x the one-hot encoded leaf colorations and y the centered indicator that the root color is 1, as in Theorem 8. Bandwidth and ridge penalty are selected via grid search on a validation set.