PAC-Bayes under potentially heavy tails
Authors: Matthew Holland
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Empirical analysis In this section, we use tightly controlled simulations to investigate how the performance of bx (cf. (3) and Proposition 4) compares with the sample mean and other robust estimators." and "Experimental setup For each experimental setting and each independent trial, we generate a sample x1, . . . , xn of size n, compute some estimator {xi}n i=1 7 bx, and record the deviation |bx Eµ |. The sample sizes range over n {10, 20, 30, . . . , 100}, and the number of trials is 104. |
| Researcher Affiliation | Academia | Matthew J. Holland Institute of Scientific and Industrial Research Osaka University matthew-h@ar.sanken.osaka-u.ac.jp |
| Pseudocode | No | No structured pseudocode or algorithm blocks were found in the paper. |
| Open Source Code | No | The paper does not provide any statement or link regarding the availability of its source code. |
| Open Datasets | No | The paper states 'we generate a sample x1, . . . , xn of size n' from 'Normal family' and 'log-Normal family', indicating synthetic data generation for the experiments rather than using a publicly available dataset with concrete access information. |
| Dataset Splits | No | The paper does not describe specific training, validation, or test dataset splits. The experimental setup involves generating synthetic data samples for robust mean estimation, not typical machine learning model training with predefined splits. |
| Hardware Specification | No | No specific hardware details (such as GPU/CPU models, memory, or cloud instance types) used for running the experiments are mentioned in the paper. |
| Software Dependencies | No | The paper does not provide specific software dependencies, libraries, or solvers with version numbers that would be needed to replicate the experiments. |
| Experiment Setup | Yes | For each experimental setting and each independent trial, we generate a sample x1, . . . , xn of size n, compute some estimator {xi}n i=1 7 bx, and record the deviation |bx Eµ |. The sample sizes range over n {10, 20, 30, . . . , 100}, and the number of trials is 104. We draw data from two distribution families, the Normal family with mean a and variance b2, and the log-Normal family, with log-mean alog and log-variance b2 log, under multiple parameter settings... all algorithms are run with confidence parameter δ = 0.01. |