HAMON: Passive Optical Sequence Mixing for Long-Horizon Forecasting
Alper Yıldırım
cs.LG, cs.AI, cs.AR
TL;DR
Researchers created a new way to predict future trends in data using light waves instead of traditional computer chips, showing that physical optical processes can perform as well as or better than standard digital forecasting models.
Summary
In the world of data forecasting, simple mathematical models often perform surprisingly well, leading researchers to question whether complex digital systems are always necessary for predicting future trends. This paper introduces HAMON, a system that replaces traditional digital processing with a physical optical core. Instead of using software to mix and analyze data sequences, the system encodes historical data onto an optical aperture. As light passes through a series of trainable phase masks, the physical properties of the light itself—specifically free-space diffraction—shape the forecast directly into an output field. This means the actual prediction happens during a single pass of light through the device, without needing a digital layer to mix the data. When tested against standard forecasting benchmarks, HAMON proved to be highly competitive. It outperformed the strongest digital models on several datasets, including ETTm2, where it improved mean squared error by up to 14%. While it trailed behind digital baselines on some of the more complex, high-channel-count datasets like Traffic and Electricity, its performance on others suggests that physical optical systems are a viable alternative for time-series forecasting. The authors conducted several tests to confirm that these predictions are indeed generated by the optical field rather than a hidden digital process. By demonstrating that a passive physical core can handle sequence mixing, the study provides a clear roadmap for developing specialized optical hardware for high-speed, energy-efficient data forecasting.
Abstract
Simple linear and frequency-domain models remain surprisingly competitive in long-horizon time-series forecasting, and recent mechanistic evidence suggests that standard forecasting benchmarks may not require the dense superposed representations that make transformers powerful in other domains. This raises a substrate-level question: if the core forecasting operator is often low-complexity and approximately linear, does it need to be implemented as learned digital temporal mixing? We introduce HAMON, a passive diffractive optical forecasting core in which historical values are encoded onto an optical aperture, future positions are left dark, and cascaded trainable phase masks with free-space diffraction shape the forecast directly in the output field. At inference, prediction is performed by a single passive optical propagation pass with no trainable digital sequence-mixing layer. Across standard benchmarks, HAMON outperforms the strongest digital baselines considered on ETTm2 at all horizons and on ETTh2 at all but the longest horizon, improving MSE by up to 14\% and doing so consistently across horizons rather than at isolated points. It is competitive on Weather and trails the strongest baselines on the remaining ETT settings and on the high-channel-count Traffic and Electricity datasets. Phase encoding, intensity-compatible readout, and phase-scrambling ablations, together with a TorchOptics cross-simulator check, indicate that the forecasts arise from the data-bearing optical field rather than from a digital forecasting head. Because the passive core uses standard Fourier optics, HAMON defines a concrete target for optical hardware and for passive physical sequence mixing.