Project
PaA-9
Geometric Neuroevolution for Fine-tuning Automatic Speech Recognition
Project Heads
Christoph von Tycowicz, Sebastian Pokutta
Project Members
Martin Hanik
Project Duration
01.01.2026 − 31.12.2027
Located at
ZIB
Description
This project aims to develop a geometry-aware neuroevolution framework by integrating information geometry into evolutionary algorithms for neural network training. By leveraging the geometry of the neuromanifold, we seek to enhance optimization efficiency and overcome challenges like slow and premature convergence. We will apply our optimization techniques to automatic speech recognition, improving the fine-tuning of foundation models. In collaboration with the Freie Universit¨at Berlin library, we will enhance transcription accuracy for audiovisual research resources, reducing errors in entity recognition and strengthening the reliability of automatic speech recognition.
Project Webpages
Related Publications
Related Picture