Citation Context

... augmented with privacy preserving capabilities. Further, optical projection and classification have been integrated (without any privacy preservation) as in [13]. Some of these algorithms are linear =-=[73, 1, 65, 12]-=- and, in future work, we may consider implementing these within our optical frame work. 2. Single Aperture Privacy Preserving Optics We introduce two optical designs that perform privacy preserving co...

Citation Context

...ns an open problem. Techniques that protect privacy of the model include privacypreserving probabilistic inference [38], privacy-preserving speaker identification [36], and computing on encrypted data [3, 6, 55]. By contrast, our objective is to collaboratively train a neural network that can be used privately and independently by each participant. Differential privacy [19] is a popular approach to privacy-preserving machine learning. It has been applied to boosting [21], principal component analysis [10], linear and logistic regression [8, 57], support vector machines [41], risk minimization [9, 53], and continuous data processing [43]. Recent results show that a noisy variant of stochastic gradient descent achieves optimal error for minimizing Lipschitz convex functions over `2-bounded sets [4], and that randomized “dropout,” used to prevent overfitting, cal also strengthen the privacy guarantee in a simple 1-layer neural network [29]. To the best of our knowledge, none of the previous work addressed the problem of collaborative deep learning with multiple participants using distributed stochastic gradient descent. Aggregation of independently trained neural networks using differential ...

Citation Context

...y. Convergence rates are calculated for stochastic algorithms that enforce privacy by signal obfuscation. In particular, the proofs below extend previous work on privacyaware optimization [11], [12], =-=[13]-=-. In this competition to derive estimates of the optimal programs the agents will undertake, we prove that introducing random perturbations in the dual gradient method is much more detrimental to an e...