Photonic Differential Privacy with Direct Feedback Alignment

Authors: Ruben Ohana, Hamlet Medina, Julien Launay, Alessandro Cappelli, Iacopo Poli, Liva Ralaivola, Alain Rakotomamonjy

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct experiments demonstrating the ability of our learning procedure to achieve solid end-task performance.
Researcher Affiliation Collaboration Ruben Ohana 1,3 , Hamlet J. Medina Ruiz 2, Julien Launay 1,3, Alessandro Cappelli1, Iacopo Poli1, Liva Ralaivola2, Alain Rakotomamonjy2 1Light On, Paris, France 2Criteo AI Lab, Paris, France 3LPENS, École Normale Supérieure, Paris, France
Pseudocode Yes Algorithm 1 Photonic DFA training
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the methodology described in this paper.
Open Datasets Yes We perform our experiments on Fashion MNIST dataset [32]
Dataset Splits Yes We perform our experiments on Fashion MNIST dataset [32], reserving 10% of the data as validation, and reporting test accuracy on a held-out set.
Hardware Specification Yes We run our simulations on cloud servers with a single NVIDIA V100 GPU and an OPU, for a total estimate of 75 GPU-hours.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., PyTorch version, Python version, etc.).
Experiment Setup Yes Optimization is done over 15 epochs with SGD, using a batch size of 256, learning rate of 0.01 and 0.9 momentum.