The joint UK-EU funding is connected to our EIC-Pathfinder project QRC-4-ESP @qrc-4-esp.bsky.social, for more details see www.qrc-4-esp.eu

Check out the paper for details, we'd love your thoughts and feedback! This research is funded by EPSRC (QTCDF EP/W028344/1), the European Union/EIC (Grant Agreement no. 101129663), and UKRI Horizon Guarantee Scheme (Grant No. 10108296) @ukri.org @ec.europa.eu

Excitingly, this is a simple modification (probably just a few lines of code) for existing platforms. There are also key connections with ongoing research in the quantum version of these systems.

Why does this matter? Enhanced expressivity means our reservoir can solve more complex nonlinear tasks without changing any property of the scattering media, and just by setting the right phase wrapping factor.

The secret: random spectral feature engineering. By allowing phase wrapping, the reservoir generates richer features through Random Fourier features interacting nonlinearly (see Rahimi & Rech's 2007 seminal paper papers.nips.cc/paper_files/... and following works (e.g., arxiv.org/pdf/2006.07310).

This might seem counterintuitive—aren't we losing information? Surprisingly, it boosts expressivity. We show clear improvements on benchmark tasks such as nonlinear time-series prediction and classification, outperforming conventional encoding strategies that don’t work within [0, 2\pi).

This is done to avoid repetition in the outputs corresponding to different data. We instead deliberately stretch the encoding domain, enabling inputs to "wrap" around multiple times in the optical phase.

In this work, we propose a new and counterintuitive approach: non-bijective phase encoding. What does it mean? Typically, in these systems, the data is encoded in the standard period [0, 2\pi).

Photonic Reservoir Computing is a powerful approach for unconventional computing, particularly in systems combining phase modulators and scattering media. Scattering-based reservoirs generally encode the data in the phase of the optical waves.

This theoretical project is the result of lengthy discussions (and long nocturnal writing sessions) among Gerard McCaul, Giulia Marcucci @giuliasnlworld.bsky.social , CEO of LumiAIres Ltd, and us at the EPicX (especially @girishtripathy.bsky.social‬).