Radio propagation used for recurrent neural computation and model tuning
RAOA: Alternating-Operator Neural Computation with Programmable Radio Propagation
Machine LearningEmerging TechnologiesLogic in Computer Science
Summary
This paper explores using programmable radio waves not just to send signals, but as a way to perform repeated computer calculations. The authors created a method called RAOA, which switches between updating a problem and mixing information to improve solutions over several steps. They tested this idea on different tasks, including solving math problems and improving language models, showing that it can help models adapt without adding extra complexity. While they demonstrated that programmable radio waves could approximate these calculations in simulations, they did not build physical hardware yet.
What this means in practice
- •For optimization engineers: Use repeated programmable radio wave computations to improve solution quality in discrete optimization problems without increasing system complexity.
- •For machine learning engineers: Adapt pretrained language models efficiently by inserting RAOA residual layers that fine-tune behavior close to shallow neural networks with minimal parameter increase.
Tested on simulated data.
Authors
Toshiaki Koike-Akino
Abstract
Can programmable radio propagation serve as computational depth rather than only as a communication channel or one-shot analog transform? We introduce the Radio Alternating Operator Ansatz (RAOA), a recurrent computing architecture that alternates an energy-derived problem update with a mixing update over a persistent latent state. Recomputing the problem field after each mix makes repeated passes compositional even when the same learned controls are reused across depth. We evaluate this idea through exact discrete optimization, constrained programmable-propagation simulation, and pretrained-model adaptation. On discrete objectives, repeated execution can improve solution quality without increasing the learned-control count, and the same formulation handles higher-order interactions directly. A passive phase-only free-space model further shows that the required operators can be approximated by programmable propagation while retaining useful downstream behavior despite realization error. When inserted as a zero-initialized residual adapter, RAOA adapts pretrained language models with WikiText performance close to a matched shallow MLP across three model families, while reasoning-task transfer remains model-dependent. Together, these results connect alternating-operator computation, programmable radio propagation, and neural adaptation within one recurrent framework. The RF realization evidence is simulation-based rather than a hardware demonstration.