Code for "Learning Structured Distributions From Untrusted Batches: Faster and Simple"

These are scripts for the paper "Learning Structured Distributions From Untrusted Batches: Faster and Simple" (preprint).

Contents

• preamble.py: contains main dependencies and global flags
• fed.py: implementation of our algorithm
• sample.py: code for generating synthetic data from untrusted batches
• experiments.py: wrapper code to run all of our experiments
• data/ is pre-populated with our experimental data.
• To reproduce the plots in this work, run python plots.py
• To re-generate data for experiment number i from scratch, run python experiments.py i for any choice of i = 1,2,3,4. See paper for estimated runtimes.