Reproduce “Rejection-Sampled Universal Quantization for Smaller Quantization Errors” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chih Wei Ling, Cheuk Ting Li
Rejection-Sampled Universal Quantizers (RSUQ) and their layered extension (LRSUQ) propose a new randomized vector quantization scheme that reduces maximum and mean squared quantization errors compared to existing lattice quantizers for specific dimensions. These quantizers further enable exact, near-optimal simulation of additive noise channels by precisely controlling the error distribution.
Key insights
- RSUQ achieves lower maximum error and mean squared error (MSE) compared to all known lattice quantizers for several specific dimensions (e.g., n=5-48 for max error, n=35-47 for MSE).
- The performance of RSUQ, in terms of normalized entropy and error characteristics, is robust and less sensitive to the specific underlying lattice chosen, simplifying its practical application.
- LRSUQ establishes layered entropy as the approximate 'high SNR limit' for one-shot additive noise channel simulation, achieving near-optimal communication cost within an additive constant of log e bits.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.