Regularly scheduled seminar in the fall of 2019
Digital adverting is arguably the largest and most ubiquitous application of machine learning. Learning algorithms pick the ads we see by inferring information about who we are and what we might buy. Graph datasets, due to their simplicity, play a central role in facilitating this inference. Internet Device Graphs are …
Learning a policy using only observational data is challenging because the distribution of states it induces at execution time may differ from the distribution observed during training. In this work, we propose to train a policy while explicitly penalising the mismatch between these two distributions over a fixed time horizon. …
We present a generalization of the collaborative filtering algorithm for the task of tensor estimation, i.e. estimating a low-rank 3-order n-by-n-by-n tensor from noisy observations of randomly chosen entries in the sparse regime. Not only does the algorithm have desirable computational properties, it also provably achieves sample complexity that (nearly) …
Adaptive experimental design (AED), or active learning, leverages already-collected data to guide future measurements, in a closed loop, to collect the most informative data for the learning problem at hand. In both theory and practice, AED can extract considerably richer insights than any measurement plan fixed in advance, using the …
Recently, the surge of interest in deep neural learning has dramatically improved image and signal processing, which has fueled breakthroughs in many domains such as drug discovery, genomics, and automatic translation. These advances have been further applied to scientific computing and, in particular, to electronic structure computations. In this case, …
Inverse problems of various flavors span all of science, engineering, and design. Over the last five years, approaches based on neural networks have emerged as the tool of choice for solving such problems. However, a clear theoretical understanding of how well such approaches perform — together with quantitative sample-complexity and …
Several important families of computational and statistical results in machine learning and randomized algorithms rely on statistical independence of data. The scope of such results include the Johnson-Lindenstrauss Lemma (JLL), the Restricted Isometry Property (RIP), regression models, and stochastic optimization. In this talk, we will discuss a new result on …
In this talk, I will present the role of side information, simulators, and GANs for learning with scarce data. In the first part, I will talk about the role of similarity graphs in recommendation systems. In the second part, the role of simulators and GANs will be discussed.
When high-quality labeled training data are unavailable, an alternative is to learn from training sources that are biased in some way. This talk will cover my group’s recent work on three problems where a learner has access to multiple biased sources. First, we consider the problem of classification given multiple …
Fitting a model to a collection of observations is one of the quintessential questions in statistics. The standard assumption is that the data was generated by a model of a given type (e.g., a mixture model). This simplifying assumption is at best only approximately valid, as real datasets are typically …