Quantum Optics Seminar: Wolfram Pernice

Embracing Uncertainty: A Photonic Neuromorphic Approach to Probabilistic Computing

Unlike artificial neural networks (ANNs), which focus on maximizing accuracy, biological systems excel at handling uncertainty. This ability is critical for adaptability and efficiency, yet traditional ANNs, implemented on deterministic hardware, fail to capture the full probabilistic nature of inference.

To address this limitation, Bayesian neural networks (BNNs) replace deterministic parameters with probability distributions, allowing us to distinguish between epistemic uncertainty (due to limited data) and aleatoric uncertainty (arising from noise). By incorporating Bayesian inference, BNNs provide a more robust approach, particularly in cases of incomplete data. However, processing probabilistic models remains a challenge for conventional digital hardware, which relies on deterministic von Neumann architectures that separate memory from computation.

To overcome these limitations, I propose a neuromorphic computing approach that harnesses hardware noise as a computational resource rather than a constraint. In electronic crossbar arrays, memristors exhibit inherent stochasticity, making them suitable for probabilistic inference. However, sequential sampling and variability in memristive materials present obstacles. By transitioning to photonic computing, we can achieve parallel probabilistic operations using chaotic light, an ideal entropy source for true random number generation.

I will present a photonic neuromorphic architecture that leverages chaotic light fields for single-shot probabilistic computing. Using non-volatile phase-change materials, we encode and process probabilistic information efficiently. Finally, I will demonstrate Bayesian inference in a LeNet-5-based model for image recognition, benchmarking accuracy and uncertainty handling on an incomplete MNIST dataset. This approach paves the way for energy-efficient, high-speed probabilistic machine learning beyond the limitations of conventional hardware.