We present a result on the convergence of weight normalized training of artificial neural networks. In the analysis, we consider over-parameterized 2-layer networks with rectified linear units (ReLUs) initialized at random and trained with batch gradient descent and a fixed step size. The proof builds on recent theoretical works that bound the trajectory of parameters from their initialization and monitor the network predictions via the evolution of a ''neural tangent kernel'' (Jacot et al. 2018). We discover that training with weight normalization decomposes such a kernel via the so called ''length-direction decoupling''. This in turn leads to two convergence regimes. From the modified convergence we make a few curious observations including a natural form of ''lazy training'' where the direction of each weight vector remains stationary.

Mini Course Part 4 of 4. The course is a very short introduction to regularity for linear elliptic pdes of second order. We start with equations with regular coefficients and the difference quotient method of Nirenberg. We then treat the case of coefficients that are merely measurable and bounded, putting forward the basics of De Giorgi-Nash-Moser theory. If time permits, we present some characterizations of Hölder spaces which are very useful in regularity theory.

Mini Course Part 3 of 4. The course is a very short introduction to regularity for linear elliptic pdes of second order. We start with equations with regular coefficients and the difference quotient method of Nirenberg. We then treat the case of coefficients that are merely measurable and bounded, putting forward the basics of De Giorgi-Nash-Moser theory. If time permits, we present some characterizations of Hölder spaces which are very useful in regularity theory.

Computational Bioscience Research Center at King Abdullah University of Science and Technology is pleased to announce the KAUST Research Conference on Digital Health 2020.

Mini Course Part 2 of 4. The course is a very short introduction to regularity for linear elliptic pdes of second order. We start with equations with regular coefficients and the difference quotient method of Nirenberg. We then treat the case of coefficients that are merely measurable and bounded, putting forward the basics of De Giorgi-Nash-Moser theory. If time permits, we present some characterizations of Hölder spaces which are very useful in regularity theory.

Mini Course Part 1 of 4. The course is a very short introduction to regularity for linear elliptic pdes of second order. We start with equations with regular coefficients and the difference quotient method of Nirenberg. We then treat the case of coefficients that are merely measurable and bounded, putting forward the basics of De Giorgi-Nash-Moser theory. If time permits, we present some characterizations of Hölder spaces which are very useful in regularity theory.

Abstract

The model under consideration in this work describes a vibrating structure of an interfac

This talk will be a gentle introduction to proximal splitting algorithms to minimize a sum of possibly nonsmooth convex functions. Several such algorithms date back to the 60s, but the last 10 years have seen the development of new primal-dual splitting algorithms, motivated by the need to solve large-scale problems in signal and image processing, machine learning, and more generally data science. No background will be necessary to attend the talk, whose goal is to present the intuitions behind this class of methods.

The talk will discuss how recent advances in wireless computing and communication nodes can be harnessed to serve the multitude of deployment scenarios required to empower communities of the future with smart and connected systems. In this talk, we address fundamental questions that should be asked when contemplating future smart and connected systems, namely, How, Where and What? (1) How can we design computing and communication nodes that best utilize resources in a way that is cognizant of both the abilities of the platform, as well as the requirements of the network? (2) Where are the nodes deployed? By understanding the context of deployment, one can architect unique solutions that are currently unimaginable. With the transformation to diverse applications such as body area networking, critical infrastructure monitoring, precision agriculture, autonomous driving, etc., the need for innovative solutions becomes even more amplified. (3) What benefit can be inferred from the data gathered by nodes in the capacity of computing, communication, and sensing?

Abstract

The analysis of solar irradiance ha

In this talk, I will describe two recent results on detecting and understanding backdoor attacks on deep learning systems. I will first present Neural Cleanse (IEEE S&P 2019), the first robust tool to detect a wide range of backdoors in deep learning models. We use the idea of perturbation distances between classification labels to detect when a backdoor trigger has created shortcuts to misclassification to a particular label.  Second, I will also summarize our new work on Latent Backdoors (CCS 2019), a stronger type of backdoor attack that is more difficult to detect and survives retraining in commonly used transfer learning systems. Latent backdoors are robust and stealthy, even against the latest detection tools (including neural cleanse).

I will present an overview of our activities around estimation problems for partial and fractional differential equations. I will present the methods and the algorithms we develop for the state, source and parameters estimation and illustrate the results with some simulations and real applications.

This research formally models, analyzes and maximizes equity of transplant waiting times and probabilities using queuing theory and network flows, based on Rawls' theory of justice. The presented formal models address inequities resulting from blood type incompatibilities, which are interrelated to ethnic differences in patient and donor rates.

The talk will give an overview of some recent developments of statistical models based on stochastic partial differential equations. We will in particular focus on equations with non-local differential operators or non-Gaussian driving noise, and explain when any why such models are useful. As motivating applications, analysis of longitudinal medical data and ocean waves will be considered.

We consider the problem of recovering a low-rank signal matrix in the presence of a general, unknown additive noise; more specifically, noise where the eigenvalues of the sample covariance matrix have a general bulk distribution. We assume given an upper bound for the rank of the assumed orthogonally invariant signal, and develop a selector for hard thresholding of singular values, which adapts to the unknown correlation structure of the noise.

A variety of intriguing patterns in eigenvalues were observed and speculated about in ML conference papers. We describe the work of Vardan Papyan showing that the traditional subdisciplines, properly deployed, can offer insights about these objects that ML researchers had.

Adil Salim is mainly interested in stochastic approximation, optimization, and machine learning. He is currently a Postdoctoral Research Fellow working with Professor Peter Richtarik at the Visual Computing Center (VCC) at King Abdullah University of Science and Technology (KAUST).

Abstract

In this talk, we discuss a mean-fie

Roy Maxion will give three lectures focusing broadly on different aspects of an increasingly important topic: reproducibility. Reproducibility tests the reliability of an experimental result and is one of the foundations of the entire scientific enterprise.

We often hear that certain foods are good for you, and a few years later we learn that they're not. A series of results in cancer research was examined to see if they were reproducible. A startling number of them - 47 out of 53 - were not. Matters of reproducibility are now cropping up in computer science, and given the importance of computing in the world, it's essential that our own results are reproducible -- perhaps especially the ones based on complex models or data sets, and artificial intelligence or machine learning. This lecture series will expose attendees to several issues in ensuring reproducibility, with the goal of teaching students (and others) some of the crucial aspects of making their own science reproducible. Hint: it goes much farther than merely making your data available to the public.

Roy Maxion will give three lectures focusing broadly on different aspects of an increasingly important topic: reproducibility. Reproducibility tests the reliability of an experimental result and is one of the foundations of the entire scientific enterprise.

We often hear that certain foods are good for you, and a few years later we learn that they're not. A series of results in cancer research was examined to see if they were reproducible. A startling number of them - 47 out of 53 - were not. Matters of reproducibility are now cropping up in computer science, and given the importance of computing in the world, it's essential that our own results are reproducible -- perhaps especially the ones based on complex models or data sets, and artificial intelligence or machine learning. This lecture series will expose attendees to several issues in ensuring reproducibility, with the goal of teaching students (and others) some of the crucial aspects of making their own science reproducible. Hint: it goes much farther than merely making your data available to the public.

In this talk, we explore a graphical model representation for the stochastic coefficients relying on the specification of the sparse precision matrix. Sparsity is encouraged in an L1-penalized likelihood framework. Estimation exploits a majorization-minimization approach. The result is a flexible nonstationary spatial model that is adaptable to very large datasets.

Roy Maxion will give three lectures focusing broadly on different aspects of an increasingly important topic: reproducibility. Reproducibility tests the reliability of an experimental result and is one of the foundations of the entire scientific enterprise.

We often hear that certain foods are good for you, and a few years later we learn that they're not. A series of results in cancer research was examined to see if they were reproducible. A startling number of them - 47 out of 53 - were not. Matters of reproducibility are now cropping up in computer science, and given the importance of computing in the world, it's essential that our own results are reproducible -- perhaps especially the ones based on complex models or data sets, and artificial intelligence or machine learning. This lecture series will expose attendees to several issues in ensuring reproducibility, with the goal of teaching students (and others) some of the crucial aspects of making their own science reproducible. Hint: it goes much farther than merely making your data available to the public.

Abstract

Many physical models are characteri

Abstract

There are many intriguing aspects a

Abstract

Abstract text.