Explicit and data-Efficient Encoding via Gradient Flow

Kyriakos Flouris, Anna Volokitin, Gustav Bredell, Ender Konukoglu · 2024

Read Explicit and data-Efficient Encoding via Gradient Flow by Kyriakos Flouris, Anna Volokitin, Gustav Bredell, Ender Konukoglu (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-00864.txt.

Description

Kyriakos Flouris, Anna Volokitin, Gustav Bredell, and Ender Konukoglu encode data by flowing it downhill into a compact code, skipping the encoder network, in work from 2024. Autoencoders squeeze data through a learned encoder, but that learned squeeze can be sloppy. In the physical sciences, sloppy codes cost real accuracy.

Their method keeps only a decoder and pushes each data point along a gradient flow into the code space, as a differential equation demands. Training uses an adjoint trick that avoids pricey integrals at small cost in accuracy. A second-order flow, echoing a fast descent method, speeds convergence. Stiff cases get an adaptive solver. Data-thrifty tests favor the approach.

Catalog record

CREATORKyriakos Flouris, Anna Volokitin, Gustav Bredell, Ender Konukoglu
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-00864

Browse the full library · About the Tower of Babel