Learning Structured Distributions From Untrusted Batches: Faster and Simpler

Authors: Sitan Chen, Jerry Li, Ankur Moitra

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically evaluated our algorithm on synthetic data. Our experiments fall under two types: (a) learning an arbitrary distribution in Aℓnorm and B) learning a structured distribution in total variation distance.
Researcher Affiliation Collaboration Sitan Chen MIT sitanc@mit.edu Jerry Li MSR jerrl@microsoft.com Ankur Moitra MIT moitra@mit.edu
Pseudocode Yes The pseudocode for LEARNWITHFILTER and 1DFILTER is given in Algorithm 1 and 2.
Open Source Code Yes All code can be found at https://github.com/ secanth/federated.
Open Datasets No The paper uses 'synthetic data' which was randomly generated by the authors, rather than a publicly available dataset with concrete access information.
Dataset Splits No The paper does not specify traditional training, validation, or test dataset splits. The problem is framed as learning from 'batches' of samples, some of which are corrupted, and the evaluation is performed on these generated batches.
Hardware Specification Yes The experiments were conducted on a Mac Book Pro with 2.6 GHz Dual-Core Intel Core i5 processor and 8 GB of RAM.
Software Dependencies No The paper mentions 'SCS solver in CVXPY' but does not specify version numbers for these software components.
Experiment Setup Yes Throughout, ω = 0 and ℓ= 10. For each trial, we randomly generated µ... In (i), we fixed k = 1000, ϵ = 0.4, N ℓ/ϵ2. In (ii), we fixed ϵ = 0.4, n = 64, N ℓ/ϵ2. In (iii) we fixed n = 64, k = 1000, N ℓ/0.42; in (iv) we fixed n = 128, k = 1000, ϵ = 0.4.