# MGS: Markov Greedy Sums for Accurate Low-Bitwidth Floating-Point Accumulation

Source: [MGS: Markov Greedy Sums for Accurate Low-Bitwidth Floating-Point Accumulation](<https://www.sota2.com/research/paper/68fa77ec1e209131cea76078>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T04:29:27.078Z

Authors: Vikas Natesh, H.T. Kung, David Kong

Published: 2025-04-12T04:19:03.000Z

## Abstract

We offer a novel approach, MGS (Markov Greedy Sums), to improve the accuracy of low-bitwidth floating-point dot products in neural network computations. In conventional 32-bit floating-point summation, adding values with different exponents may lead to loss of precision in the mantissa of the smaller term, which is right-shifted to align with the larger term's exponent. Such shifting (a.k.a. 'swamping') is a significant source of numerical errors in accumulation when implementing low-bitwidth dot products (e.g., 8-bit floating point) as the mantissa has a small number of bits. We avoid most swamping errors by arranging the terms in dot product summation based on their exponents and summing the mantissas without overflowing the low-bitwidth accumulator. We design, analyze, and implement the algorithm to minimize 8-bit floating point error at inference time for several neural networks. In contrast to traditional sequential summation, our method has significantly lowered numerical errors, achieving classification accuracy on par with high-precision floating-point baselines for multiple image classification tasks. Our dMAC hardware units can reduce power consumption by up to 34.1\\% relative to conventional MAC units.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2504.09072>)

[PDF](<https://arxiv.org/pdf/2504.09072.pdf>)
