Upcoming Seminars

No upcoming events found.

Previous Seminars

Physics-informed Machine Learning For Climate, Urban, And Environmental Sciences

Thu, Mar 14, 2019, 12:30 pm to 1:30 pm
At the confluence of scientific simulation and modern machine learning there exists an opportunity to develop a “middle path” that leverages the strengths of both approaches to build machine-learning emulators of numerical simulation models. From the perspective of machine learning, incorporating simulation data may significantly reduce the need...

Cosmology with galaxy surveys: from precision to accuracy with data-driven models

Tue, Mar 12, 2019, 3:00 pm to 4:00 pm
Immense surveys of the night sky are being conducted to test the origins and evolutions of the Universe as a whole, using distant galaxies and quasars (bright galaxies with active supermassive black holes at their center). This requires modeling the spatial and spectral statistics of those sources in detail using millions of noisy observations,...

Fitting Convex Sets to Data

Mon, Mar 11, 2019, 4:30 pm to 5:30 pm
A number of problems in signal processing may be viewed conceptually as fitting a convex set to data.  In vision and learning, the task of identifying a collection of features or atoms that provide a concise description of a dataset has been widely studied under the title of dictionary learning or sparse coding.  In convex-geometric terms, this...

The Complexities of Astronomical Data

Fri, Mar 8, 2019, 2:00 pm to 3:00 pm
The analysis of astronomical data set reveals an astonishing diversity of astrophysical processes but also a dizzying array of observational complications. I will introduce several approaches to deal with complications like missing data, non-linear error propagation, and overlapping objects. I will concentrate on proximal optimization algorithms...

Astrophysical Inference with Complex, Stochastic Time Series

Tue, Feb 26, 2019, 3:00 pm to 4:00 pm
Astronomical time series—measurements of a celestial source’s brightness as a function of time—are key to probing the physical processes governing the behaviour of these sources on a large range of scales, from small asteroids in our solar system, to supermassive black holes at the centres of galaxies. Modern time domain surveys like the Zwicky...

Binary star science with sparse, noisy, and missing data

Fri, Feb 22, 2019, 2:00 pm to 3:00 pm
Many open questions in astrophysics hinge on a better understanding of binary star populations (their occurrence rate, orbital parameter distributions, and how these vary with environment).Yet, most of what we know about binary star statistics comes from just a few hundred stars nearest to the sun. Contemporary astronomical surveys have the...

Machine Learning and Causal Inference for Heterogeneous Treatment Effects

Fri, Feb 22, 2019, 12:00 pm to 1:00 pm
This talk will review recently developed methods to apply machine learning methods to causal inference problems, including the problems of estimating heterogeneous treatment effects, for example, in A/B testing, as well as in estimating optimal treatment assignment policies.

Decoding the Milky Way Galaxy

Fri, Feb 15, 2019, 2:00 pm to 3:00 pm
Stars residing in the Milky Way halo hold the key to the origin of our Galaxy. The stellar halo has traditionally been inaccessible as it contains only one percent of stars in the Galaxy. However, novel data sets have enabled us to confidently identify thousands of halo stars in a high-dimensional space of stellar positions, velocities and...

Sparse matrices in sparse analysis

Wed, Feb 13, 2019, 4:30 pm
In this talk, I will give two vignettes on the theme of sparse matrices in sparse analysis. The first vignette covers work from compressive sensing in which we want to design sparse matrices (i.e., matrices with few non-zero entries) that we use to (linearly) sense or measure compressible signals.

Rethinking the Role of Optimization in Learning

Mon, Feb 11, 2019, 4:30 pm
In this talk, I will overview our recent progress towards understanding how we learn large capacity machine learning models, especially deep neural networks. In the modern practice of deep learning, many successful models have far more trainable parameters compared to the number of training examples.