November 12, 2015 at 1:00 p.m. Marcus Brubaker, Ph.D., will be presenting “Efficient 3D Molecular Structure Estimation with Electron Cryomicroscopy”. Speaker: Marcus Brubaker, Ph.D. Postdoctoral at University of Toronto Day & Time: Thursday, November 12, 2015 1:00 p.m. – 2:00 p.m. Location: Room ENG106, Ryerson University 350 Victoria Street, Toronto, Ontario M5B 2K3 Click here to see the Map – Look for ENG Organizer: Instrumentation & Measurement and Magnetics Chapters at IEEE Toronto Contact: Dr. Maryam Davoudpour: maryam.davoudpour@ieee.org Abstract: Discovering the 3D structure of molecules such as proteins and viruses is a fundamental research problem in biology and medicine. Electron Cryomicroscopy (Cryo-EM) is a promising vision-based technique for structure estimation which attempts to reconstruct 3D structures from 2D images. This talk reviews the computational problems in Cryo-EM which are closely related to classical vision problems such as object detection, multiview reconstruction and computed tomography. Finally, a framework is introduced for reconstruction of 3D molecular structure which exploits modern methods for stochastic optimization and importance sampling. The result is a method which is efficient, robust to initialization and flexible. Biography: Marcus Brubaker received his Ph.D. in Computer Science from the University of Toronto in 2011. After that he worked with Raquel Urtasun as a postdoctoral researcher at Toyota Technological Institute at Chicago and is currently a postdoc at University of Toronto, Scarborough. He also consults with Cadre Research Labs on machine learning and computer vision related projects and teaches at the University of Toronto. He was won a number of fellowships and awards, including OGS and NSERC graduate fellowships as well as an NSERC Postdoctoral Fellowship. His most recent work on autonomous vehicle localization (“Lost! Leveraging the Crowd for Probabilistic Visual Self-Localization,” CVPR 2013) and the estimation of the 3D structure of proteins and viruses (“Building Proteins in a Day,” CVPR 2015) have won awards and attention in the lay press. His interests span computer vision, machine learning and statistics and he works on a range of problems including video-based human motion estimation, physical models of human motion, Bayesian inference, Markov Chain Monte Carlo methods, ballistic forensics, electron cryo-microscopy and autonomous vehicle localization.
Instrumentation & Measurement
Events
Calendar of Events
|
Monday
|
Tuesday
|
Wednesday
|
Thursday
|
Friday
|
Saturday
|
Sunday
|
|---|---|---|---|---|---|---|
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
|
0 events,
|
0 events,
|
0 events,
|
1 event,
-
|
0 events,
|
0 events,
|
0 events,
|
|
0 events,
|
0 events,
|
0 events,
|
1 event,
-
Thursday November 19, 2015 at 1:00 p.m. Mohammad Norouzi, PhD candidate in computer science at the University of Toronto, will be presenting “Compact Discrete Representations for Scalable Similarity Search”. Speaker: Mohammad Norouzi PhD Candidate Day & Time: Thursday, November 19, 2015 1:00 p.m. – 2:00 p.m. Location: Room ENG 106 George Vari Engineering and Computing Centre Ryerson University 245 Church Street Toronto Organizer: IEEE Toronto Computer, Magnetics and Instrument-Measurement Chapters Contact: Maryam Davoudpour, Email:maryam.davoudpour@ieee.org Abstract: Scalable similarity search on images, documents, and user activities benefits generic search, data visualization, and recommendation systems. This talk concerns the design of algorithms and machine learning tools for faster and more accurate similarity search. The proposed techniques advocate the use of discrete codes for representing the similarity structure of data in a compact way. In particular, I will discuss how one can learn to map high-dimensional data onto binary codes with a metric learning approach. Then, I will describe a simple algorithm for fast exact nearest neighbour search in Hamming distance, which exhibits sub-linear query time performance. Going beyond binary codes, I will highlight a compositional generalization of k-means clustering which maps data points onto integer codes with storage and search costs that grow sub-linearly in the number of cluster centers. This representation improves upon binary codes, and provides an even more precise approximation of Euclidean distance. Experimental results are reported on multiple datasets including a dataset of SIFT descriptors with 1B entries. Biography: Mohammad Norouzi is a PhD candidate in computer science at the University of Toronto. His research lies at the intersection of machine learning and computer vision. He is a recipient of a Google US/Canada PhD fellowship in machine learning. He is going to join Google as a research scientist in January 2016. |
0 events,
|
0 events,
|
0 events,
|
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|
0 events,
|