Search papers, labs, and topics across Lattice.
This study investigates the limitations imposed by hardware constraints on photonic Bayesian neural networks (BNNs) during inference, specifically focusing on factors like quantization, programming error, and representable mean and variance. By framing photonic BNN inference as constrained stochastic variational inference, the authors conduct a systematic ablation study to identify which constraints can be mitigated through training versus those necessitating hardware changes. The findings yield actionable co-design guidelines that enhance predictive performance and uncertainty quality in photonic BNNs, demonstrating that hardware-aware training can effectively compensate for certain constraints while highlighting the need for targeted modifications in other cases.
Hardware-aware training can recover predictive performance in photonic Bayesian neural networks, but only if the required variational family remains representable.
Classical neural networks frequently produce overconfident predictions on ambiguous or out-of-distribution (OOD) data, a liability that grows with each AI system deployed in safety-critical real-world scenarios. Bayesian neural networks (BNNs) provide a principled framework for uncertainty-aware prediction by replacing deterministic parameters with probability distributions, but repeated sampling increases latency, memory traffic, and energy consumption. Photonic probabilistic computing offers a promising alternative by exploiting intrinsic optical stochasticity for fast and parallel sampling. However, photonic BNNs are not ideal samplers: analog constraints on quantization, programming error, dynamic range, and representable mean and variance restrict the variational families that can be implemented in hardware. In this work, we study which hardware-imposed constraints limit scalable photonic BNN inference, how these constraints can be represented, and which ranges can be tolerated by photonic BNNs beyond small proof-of-concept networks. We formulate photonic BNN inference as constrained stochastic variational inference and perform a systematic ablation study over stochasticity location, stochasticity modality, quantization, programming error, and mean/variance bounds. From these results, we derive concrete co-design guidelines that distinguish hardware constraints that can be compensated by training from those requiring hardware or architecture intervention. We validate these guidelines under coupled, hardware-realistic constraints on Dirty-MNIST, CIFAR-10, and CINIC-10, using Fashion-MNIST and SVHN as OOD benchmarks, showing that hardware-aware training recovers predictive performance and uncertainty quality whenever the required variational family remains representable, whereas violations of representational limits require targeted hardware modifications.
