no code implementations • 12 Feb 2024 • Yiyun He, Roman Vershynin, Yizhe Zhu
We present a polynomial-time algorithm for online differentially private synthetic data generation.
no code implementations • 26 Jan 2024 • Girish Kumar, Thomas Strohmer, Roman Vershynin
Much of the research in differential privacy has focused on offline applications with the assumption that all data is available at once.
no code implementations • 26 May 2023 • Yiyun He, Thomas Strohmer, Roman Vershynin, Yizhe Zhu
Differentially private synthetic data provide a powerful mechanism to enable data analysis while protecting sensitive information about individuals.
no code implementations • 31 May 2022 • Kathryn Dover, Zixuan Cang, Anna Ma, Qing Nie, Roman Vershynin
In general applications, other methods can be used for the alignment and dimension reduction modules.
no code implementations • 15 Feb 2022 • Pierre Baldi, Roman Vershynin
The gating mechanisms correspond to multiplicative extensions of the standard model and are used across all current attention-based deep learning architectures.
no code implementations • 19 Feb 2021 • Pierre Baldi, Roman Vershynin
Motivated by biological considerations, we study sparse neural maps from an input layer to a target layer with sparse activity, and specifically the problem of storing $K$ input-target associations $(x, y)$, or memories, when the target vectors $y$ are sparse.
no code implementations • 20 Jan 2020 • Roman Vershynin
Overwhelming theoretical and empirical evidence shows that mildly overparametrized neural networks -- those with more connections than the size of the training data -- are often able to memorize the training data with $100\%$ accuracy.
no code implementations • 28 Oct 2019 • Yan Shuo Tan, Roman Vershynin
In recent literature, a general two step procedure has been formulated for solving the problem of phase retrieval.
no code implementations • 2 Jan 2019 • Pierre Baldi, Roman Vershynin
Here we define the capacity of an architecture by the binary logarithm of the number of functions it can compute, as the synaptic weights are varied.
no code implementations • NeurIPS 2018 • Pierre Baldi, Roman Vershynin
We define the capacity of a learning machine to be the logarithm of the number (or volume) of the functions it can implement.
no code implementations • 30 Jun 2017 • Yan Shuo Tan, Roman Vershynin
We consider the problem of phase retrieval, i. e. that of solving systems of quadratic equations.
no code implementations • 4 Apr 2017 • Yan Shuo Tan, Roman Vershynin
The problem of Non-Gaussian Component Analysis (NGCA) is about finding a maximal low-dimensional subspace $E$ in $\mathbb{R}^n$ so that data points projected onto $E$ follow a non-gaussian distribution.
no code implementations • 31 May 2014 • Can M. Le, Elizaveta Levina, Roman Vershynin
Community detection is one of the fundamental problems of network analysis, for which a number of methods have been proposed.