Faster cryptonets: Leveraging sparsity for real-world encrypted inference

E Chou, J Beal, D Levy, S Yeung, A Haque… - arxiv preprint arxiv …, 2018 - arxiv.org
Homomorphic encryption enables arbitrary computation over data while it remains
encrypted. This privacy-preserving feature is attractive for machine learning, but requires …

The loss surface of deep and wide neural networks

Q Nguyen, M Hein - International conference on machine …, 2017 - proceedings.mlr.press
While the optimization problem behind deep neural networks is highly non-convex, it is
frequently observed in practice that training deep networks seems possible without getting …

Safetynets: Verifiable execution of deep neural networks on an untrusted cloud

Z Ghodsi, T Gu, S Garg - Advances in Neural Information …, 2017 - proceedings.neurips.cc
Inference using deep neural networks is often outsourced to the cloud since it is a
computationally demanding task. However, this raises a fundamental issue of trust. How can …

The mechanism of prediction head in non-contrastive self-supervised learning

Z Wen, Y Li - Advances in Neural Information Processing …, 2022 - proceedings.neurips.cc
The surprising discovery of the BYOL method shows the negative samples can be replaced
by adding the prediction head to the network. It is mysterious why even when there exist …

When is a convolutional filter easy to learn?

SS Du, JD Lee, Y Tian - arxiv preprint arxiv:1709.06129, 2017 - arxiv.org
We analyze the convergence of (stochastic) gradient descent algorithm for learning a
convolutional filter with Rectified Linear Unit (ReLU) activation function. Our analysis does …

On connected sublevel sets in deep learning

Q Nguyen - International conference on machine learning, 2019 - proceedings.mlr.press
This paper shows that every sublevel set of the loss function of a class of deep over-
parameterized neural nets with piecewise linear activation functions is connected and …

Optimization landscape and expressivity of deep CNNs

Q Nguyen, M Hein - International conference on machine …, 2018 - proceedings.mlr.press
We analyze the loss landscape and expressiveness of practical deep convolutional neural
networks (CNNs) with shared weights and max pooling layers. We show that such CNNs …

On the loss landscape of a class of deep neural networks with no bad local valleys

Q Nguyen, MC Mukkamala, M Hein - arxiv preprint arxiv:1809.10749, 2018 - arxiv.org
We identify a class of over-parameterized deep neural networks with standard activation
functions and cross-entropy loss which provably have no bad local valley, in the sense that …

Adding one neuron can eliminate all bad local minima

S Liang, R Sun, JD Lee… - Advances in Neural …, 2018 - proceedings.neurips.cc
One of the main difficulties in analyzing neural networks is the non-convexity of the loss
function which may have many bad local minima. In this paper, we study the landscape of …

Understanding the loss surface of neural networks for binary classification

S Liang, R Sun, Y Li, R Srikant - International Conference on …, 2018 - proceedings.mlr.press
It is widely conjectured that training algorithms for neural networks are successful because
all local minima lead to similar performance; for example, see (LeCun et al., 2015; …