Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
IEEE

Reverse Convolutional Precoding of Polar Codes: Design, Analysis, and Decoding Algorithms

Abstract: Convolutionally precoded polar codes known as polarization-adjusted convolutional (PAC) codes are a promising variant of polar codes for short block lengths. The precoding in PAC codes has ...
cmu.edu

Researchers Decode the Welfare Effects of Pricing Algorithms

The National Bureau of Economic Research has published a new working paper by economists Ali Shourideh (Carnegie Mellon University, Tepper School of Business), Maryam Farboodi (Massachusetts Institute ...
IEEE

A Local Re-Decoding Algorithm for LDPC-Coded Modulation With CCDMs

Abstract: This letter proposes to break trapping sets of the low-density parity-check (LDPC) codes used in an LDPC-coded modulation system that employs constant composition distribution matchers ...