More Like this

1. Estimation of the covariance structure of heavy-tailed distributions

   Minsker, Stanislav ; Wei, Xiaohan ( December 2017 , NIPS)

   We propose and analyze a new estimator of the covariance matrix that admits strong theoretical guarantees under weak assumptions on the underlying distribution, such as existence of moments of only low order. While estimation of covariance matrices corresponding to sub-Gaussian distributions is well-understood, much less in known in the case of heavy-tailed data. As K. Balasubramanian and M. Yuan write, "data from real-world experiments oftentimes tend to be corrupted with outliers and/or exhibit heavy tails. In such cases, it is not clear that those covariance matrix estimators .. remain optimal" and "what are the other possible strategies to deal with heavy tailed distributions warrant further studies." We make a step towards answering this question and prove tight deviation inequalities for the proposed estimator that depend only on the parameters controlling the intrinsic dimension'' associated to the covariance matrix (as opposed to the dimension of the ambient space); in particular, our results are applicable in the case of high-dimensional observations. 

   more » « less
2. Constant-Expansion Suffices for Compressed Sensing with Generative Priors. In the

   Daskalakis, C ; Rohatgi, D ; Zampetakis, M ( January 2020 , 34th Annual Conference on Neural Information Processing Systems (NeurIPS), NeurIPS 2020)

   null (Ed.)

   Generative neural networks have been empirically found very promising in providing effective structural priors for compressed sensing, since they can be trained to span low-dimensional data manifolds in high-dimensional signal spaces. Despite the non-convexity of the resulting optimization problem, it has also been shown theoretically that, for neural networks with random Gaussian weights, a signal in the range of the network can be efficiently, approximately recovered from a few noisy measurements. However, a major bottleneck of these theoretical guarantees is a network expansivity condition: that each layer of the neural network must be larger than the previous by a logarithmic factor. Our main contribution is to break this strong expansivity assumption, showing that constant expansivity suffices to get efficient recovery algorithms, besides it also being information-theoretically necessary. To overcome the theoretical bottleneck in existing approaches we prove a novel uniform concentration theorem for random functions that might not be Lipschitz but satisfy a relaxed notion which we call "pseudo-Lipschitzness." Using this theorem we can show that a matrix concentration inequality known as the Weight Distribution Condition (WDC), which was previously only known to hold for Gaussian matrices with logarithmic aspect ratio, in fact holds for constant aspect ratios too. Since the WDC is a fundamental matrix concentration inequality in the heart of all existing theoretical guarantees on this problem, our tighter bound immediately yields improvements in all known results in the literature on compressed sensing with deep generative priors, including one-bit recovery, phase retrieval, low-rank matrix recovery, and more. 

   more » « less
3. Robust and Tuning-Free Sparse Linear Regression via Square-Root Slope

   Minsker, S ; . Ndaoud, M. ; Wang, W. ( October 2022 , arXivorg)

   We consider the high-dimensional linear regression model and assume that a fraction of the responses are contaminated by an adversary with complete knowledge of the data and the underlying distribution. We are interested in the situation when the dense additive noise can be heavy-tailed but the predictors have sub-Gaussian distribution. We establish minimax lower bounds that depend on the the fraction of the contaminated data and the tails of the additive noise. Moreover, we design a modification of the square root Slope estimator with several desirable features: (a) it is provably robust to adversarial contamination, with the performance guarantees that take the form of sub-Gaussian deviation inequalities and match the lower error bounds up to log-factors; (b) it is fully adaptive with respect to the unknown sparsity level and the variance of the noise, and (c) it is computationally tractable as a solution of a convex optimization problem. To analyze the performance of the proposed estimator, we prove several properties of matrices with sub-Gaussian rows that could be of independent interest. 

   more » « less
4. Comparing Decentralized Gradient Descent Approaches and Guarantees

   https://doi.org/10.1109/ICASSP49357.2023.10096994  

   Moothedath, Shana ; Vaswani, Namrata ( January 2023 , 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP))

   This work studies our recently developed decentralized algorithm, decentralized alternating projected gradient descent algorithm, called Dec-AltProjGDmin, for solving the following low-rank (LR) matrix recovery problem: recover an LR matrix from independent column-wise linear projections (LR column-wise Compressive Sensing). In recent work, we presented constructive convergence guarantees for Dec-AltProjGDmin under simple assumptions. By "constructive", we mean that the convergence time lower bound is provided for achieving any error level ε. However, our guarantee was stated for the equal neighbor consensus algorithm (at each iteration, each node computes the average of the data of all its neighbors) while most existing results do not assume the use of a specific consensus algorithm, but instead state guarantees in terms of the weights matrix eigenvalues. In order to compare with these results, we first modify our result to be in this form. Our second and main contribution is a theoretical and experimental comparison of our new result with the best existing one from the decentralized GD literature that also provides a convergence time bound for values of ε that are large enough. The existing guarantee is for a different problem setting and holds under different assumptions than ours and hence the comparison is not very clear cut. However, we are not aware of any other provably correct algorithms for decentralized LR matrix recovery in any other settings either. 

   more » « less
5. A Tail-Index Analysis of Stochastic Gradient Noise in Deep Neural Networks

   Simsekli, Umut ; Sagun, Levent ; Gurbuzbalaban, Mert ( June 2019 , Proceedings of Machine Learning Research)

   The gradient noise (GN) in the stochastic gradient descent (SGD) algorithm is often considered to be Gaussian in the large data regime by assuming that the classical central limit theorem (CLT) kicks in. This assumption is often made for mathematical convenience, since it enables SGD to be analyzed as a stochastic differential equation (SDE) driven by a Brownian motion. We argue that the Gaussianity assumption might fail to hold in deep learning settings and hence render the Brownian motion-based analyses inappropriate. Inspired by non-Gaussian natural phenomena, we consider the GN in a more general context and invoke the generalized CLT (GCLT), which suggests that the GN converges to a heavy-tailed -stable random variable. Accordingly, we propose to analyze SGD as an SDE driven by a Lévy motion. Such SDEs can incur ‘jumps’, which force the SDE transition from narrow minima to wider minima, as proven by existing metastability theory. To validate the -stable assumption, we conduct extensive experiments on common deep learning architectures and show that in all settings, the GN is highly non-Gaussian and admits heavy-tails. We further investigate the tail behavior in varying network architectures and sizes, loss functions, and datasets. Our results open up a different perspective and shed more light on the belief that SGD prefers wide minima. 

   more » « less