Many engineering applications such as infrastructure operation and model predictive control (MPC) involve solving similar optimization problems over and over and over and over again, with slightly varying input parameters. Electric grid optimization, which is influenced by variable renewable energy generation, is a prominent example. In this talk, we consider …
Imagine a drone looking for a safe landing site in a dense forest, or a social robot trying to determine the emotional state of a person by measuring her skin-tremors due to pulse beats. In these applications, it is critical to resolve fine geometric details, such as tree twigs; and …
Distributed Stochastic Gradient Descent (SGD) when run in a synchronous manner, suffers from delays in waiting for the slowest workers (stragglers). Asynchronous methods can alleviate stragglers, but cause gradient staleness that can adversely affect convergence. In this work, we present the first theoretical characterization of the speed-up offered by asynchronous …
The problem of optimizing objective functions that involve expectation or integral operators is common in a number of fields. This problem is commonly addressed using one of three frameworks or hybrids thereof: Sample Average Approximation/Monte Carlo (SAA/MC), Bayesian Optimization (BO), and Stochastic Approximation (SA). While the methods that belong to …
A striking feature of modern supervised machine learning is its consistent use of techniques that interpolate the data. Deep networks, often containing several orders of magnitude more parameters than data points, are trained to obtain near zero error on the training set. Yet, at odds with most theory, they show …
From the confusion surrounding the optimization and generalization of deep networks has arisen an exciting possibility: gradient descent is implicitly regularized, meaning it not only outputs iterates of low error, but moreover iterates of low complexity. This talk starts with a “spectrally-normalized” generalization bound which is small if gradient descent …
In this talk, I will present my work on learning policies for navigation. I will talk about three projects. First, I will show how insights from classical mapping and planning can be operationalized in context of learning based policies to arrive at policies that can effectively leverage statistical regularities in …
A conditional Gaussian framework for uncertainty quantification, data assimilation and prediction of complex nonlinear turbulent dynamical systems will be introduced in this talk. Despite the conditional Gaussianity, the dynamics remain highly nonlinear and are able to capture strongly non-Gaussian features such as intermittency and extreme events. The conditional Gaussian structure …
Tensors of order 3 or greater, known as higher-order tensors, have recently attracted increased attention in many fields. Methods built on tensors provide powerful tools to capture complex structures in data that lower-order methods may fail to exploit. However, extending familiar matrix concepts to higher-order tensors is not straightforward, and …
Bio-sensors are becoming an integral part of our everyday life, implicitly and explicitly. The successful operation and availability of biological circuits and components for data processing makes bacteria strong candidates to use as computing machines. Currently bio-sensors, including bacterial sensors are processed independently off-line, leading to delays, manual errors and …