【New Publication】General Transform: A Unified Framework for AdaptiveTransform to Enhance Representations

https://arxiv.org/abs/2505.04969

Gekko et al. from Quemix have published a paper proposing a novel transformation method, the General Transform (GT), designed to enhance the representational power of machine learning models.

The paper introduces a framework that replaces traditional fixed transforms—such as the discrete Fourier transform—with a learnable combination of transforms that adapts to both the dataset and the task. The weights of the transforms are optimized jointly with the model, enabling general-purpose and efficient feature extraction.

Experiments on image recognition and natural language processing tasks demonstrate consistent performance improvements over conventional transform-based approaches, indicating the broad applicability of GT in future machine learning applications.

This work will be presented as a poster at Q2B 2025 Tokyo (May 15–16, 2025), an international conference on the business applications of quantum technology, where Quemix is participating as a platinum sponsor.