Schematic of next-character prediction using reservoir computing and transformers (IMAGE)

Saitama University

Schematic of next-character prediction using reservoir computing and transformers

Caption

Schematic illustration of the language-modeling framework used in the study. Input text is first converted into numerical vector representations and then processed by one of three architectures: traditional reservoir computing, attention-enhanced reservoir computing, or a transformer. The resulting output is converted into probabilities for the next character, and the prediction error is evaluated using cross-entropy. By comparing these architectures with similar numbers of trainable parameters, the researchers investigated the trade-off between predictive performance and computational efficiency.

Credit

© 2026 American Physical Society. Source: Felix Köster and Atsushi Uchida, Physical Review Applied 26, 024051 (2026), DOI: 10.1103/sd11-x3ny

Usage Restrictions

Cannot be used without permission.

License

Licensed content

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.