[Paper Review] Probabilistic Photonic Computing with Chaotic Light

Frank Brückerhoff‐Plückelmann, Hendrik Borras+11 more|arXiv (Cornell University)|Jan 31, 2024

Neural Networks and Reservoir Computing4 citations

TL;DR

This paper presents a photonic computing architecture that leverages chaotic light as a physical entropy source to enable high-speed probabilistic inference via Bayesian neural networks. By using incoherent photonic crossbars to sample from probabilistic distributions in parallel at 70.4 GS/s, the system simultaneously performs image classification and uncertainty quantification, demonstrating a seamless integration of physical randomness and computational speed.

ABSTRACT

Biological neural networks effortlessly tackle complex computational problems and excel at predicting outcomes from noisy, incomplete data, a task that poses significant challenges to traditional processors. Artificial neural networks (ANNs), inspired by these biological counterparts, have emerged as powerful tools for deciphering intricate data patterns and making predictions. However, conventional ANNs can be viewed as "point estimates" that do not capture the uncertainty of prediction, which is an inherently probabilistic process. In contrast, treating an ANN as a probabilistic model derived via Bayesian inference poses significant challenges for conventional deterministic computing architectures. Here, we use chaotic light in combination with incoherent photonic data processing to enable high-speed probabilistic computation and uncertainty quantification. Since both the chaotic light source and the photonic crossbar support multiple independent computational wavelength channels, we sample from the output distributions in parallel at a sampling rate of 70.4 GS/s, limited only by the electronic interface. We exploit the photonic probabilistic architecture to simultaneously perform image classification and uncertainty prediction via a Bayesian neural network. Our prototype demonstrates the seamless cointegration of a physical entropy source and a computational architecture that enables ultrafast probabilistic computation by parallel sampling.

Motivation & Objective

• To address the limitations of deterministic computing in handling uncertainty in artificial neural network predictions.
• To enable high-speed probabilistic inference by exploiting physical entropy sources in photonic hardware.
• To demonstrate parallel sampling of probabilistic outputs using chaotic light and incoherent photonic processing.
• To integrate a physical entropy source directly into a computational architecture for real-time uncertainty-aware machine learning.
• To achieve ultrafast, scalable probabilistic computing by leveraging wavelength-division multiplexing in photonic crossbars.

Proposed method

• Utilizes a chaotic light source as a physical entropy generator to produce stochastic input signals.
• Employs an incoherent photonic crossbar architecture to perform parallel, wavelength-multiplexed computations across multiple channels.
• Applies the chaotic light signals to a Bayesian neural network implemented in photonic hardware for probabilistic inference.
• Employs electronic interface to sample the photonic output at 70.4 GS/s, limited only by electronic bandwidth.
• Leverages the inherent statistical properties of chaotic light to sample from complex probability distributions without explicit random number generation.
• Uses wavelength-division multiplexing to enable independent computation across multiple parallel channels, increasing throughput.

Experimental results

Research questions

• RQ1 Can chaotic light serve as a scalable, high-bandwidth physical entropy source for probabilistic computing?
• RQ2 Can photonic crossbar architectures enable parallel sampling of probabilistic outputs at multi-gigasample-per-second rates?
• RQ3 Can a photonic system simultaneously perform image classification and uncertainty quantification using Bayesian neural networks?
• RQ4 How does the integration of physical randomness with photonic computation affect inference speed and accuracy?
• RQ5 To what extent can incoherent photonic processing replace electronic random number generation in probabilistic AI workloads?

Key findings

• The system achieves a sampling rate of 70.4 GS/s, limited only by the electronic interface, demonstrating ultrafast probabilistic sampling.
• The photonic architecture enables simultaneous image classification and uncertainty prediction using a Bayesian neural network.
• Chaotic light provides a high-bandwidth, intrinsic source of entropy that eliminates the need for external random number generators.
• The incoherent photonic crossbar supports multiple independent computational wavelength channels, enabling parallel probabilistic inference.
• The prototype demonstrates seamless cointegration of physical entropy and computational architecture for real-time probabilistic computing.
• The approach enables high-speed, uncertainty-aware inference by directly leveraging the statistical properties of chaotic light in a photonic framework.