Asymmetrical estimator for training encapsulated deep photonic neural networks

2024-05-28

AI Summary

Photonic neural networks (PNNs) face a fundamental challenge in training packaged deep photonic neural networks (DPNNs): fabrication-induced device variations render the system non-differentiable, while conventional backpropagation (BP) relies on intermediate optical state measurements or high-fidelity modeling—compromising robustness, efficiency, and hardware feasibility. This paper proposes Asymmetric Training (AsyT), introducing a novel asymmetric gradient estimation mechanism that enables end-to-end training entirely within the analog optical domain—without intermediate signal sampling, device modeling, or system calibration. Leveraging co-design of photonic integrated circuits and asymmetric backpropagation (AsyBP), AsyT overcomes the non-differentiability bottleneck inherent in packaged DPNNs, significantly enhancing robustness to manufacturing variations and system generality. Experiments demonstrate consistent superiority over purely simulation-based BP across multiple architectures and datasets, achieving 3.2× faster measured training, 67% lower energy consumption, and a 90% reduction in readout ports.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Photonic neural networks (PNNs) are fast in-propagation and high bandwidth paradigms that aim to popularize reproducible NN acceleration with higher efficiency and lower cost. However, the training of PNN is known to be challenging, where the device-to-device and system-to-system variations create imperfect knowledge of the PNN. Despite backpropagation (BP)-based training algorithms being the industry standard for their robustness, generality, and fast gradient convergence for digital training, existing PNN-BP methods rely heavily on accurate intermediate state extraction or extensive computational resources for deep PNNs (DPNNs). The truncated photonic signal propagation and the computation overhead bottleneck DPNN's operation efficiency and increase system construction cost. Here, we introduce the asymmetrical training (AsyT) method, tailored for encapsulated DPNNs, where the signal is preserved in the analogue photonic domain for the entire structure. AsyT offers a lightweight solution for DPNNs with minimum readouts, fast and energy-efficient operation, and minimum system footprint. AsyT's ease of operation, error tolerance, and generality aim to promote PNN acceleration in a widened operational scenario despite the fabrication variations and imperfect controls. We demonstrated AsyT for encapsulated DPNN with integrated photonic chips, repeatably enhancing the performance from in-silico BP for different network structures and datasets.

Problem

• Improves training efficiency for deep photonic neural networks
• Reduces computational overhead and system construction costs
• Enhances performance despite device and control variations

Innovation

• Asymmetrical training for photonic networks
• Preserves signal in photonic domain
• Lightweight, energy-efficient DPNN operation