Copy
Week of October 24 - October 28, 2016:
View this email in your browser

ICME Weekly Seminar Digest

Please click here for upcoming seminar information as well as other events going on in ICME.
MONDAY, OCTOBER 24, 2016:
CME 500: Departmental Seminar
Speaker: Ron Estrin, Graduate student at ICME 

Title:  Estimating the 2-Norm Forward Error for SYMMLQ and CG

Abstract: When running iterative methods such as SYMMLQ and CG for large sparse symmetric linear systems, Ax=b, often one is interested in knowing how 'close' the current approximation is to the true x. Typically backward error estimates like the residual are used, but sometimes these can be misleading. Instead it would be preferable to have a handle on the distance from the true solution. We provide a cheap estimate of the forward error (in 2-norm) for SYMMLQ and CG, which is an upper bound when A is positive definite. Numerical experiments demonstrate the quality of these estimates.

Bio: Ron Estrin is a 3rd year PhD student at ICME. His advisor(s) include Michael Saunders (and Dominique Orban for this particular project). Ron's academic interests include Numerical Linear Algebra and Optimization. During his undergraduate career, he studied Combined Math and Computer Science at University of British Columbia. Ron also completed internships at Google (2013) and Microsoft (2014, 2015); and was a Gene Golub Fellowship recipient. For fun, Ron enjoys tennis and taekwondo. 

Location: Y2E2-111
Time: 4:30-5:20 p.m.
TUESDAY, OCTOBER 25, 2016:
CME 300: First Year Seminar Series
Join us and hear from two speakers this week:

Speaker: Ariel Schwartzman, Associate Professor of Particle Physics and Astrophysics

Title: Data science and the Large Hadron Collider

Abstract: The Large Hadron Collider (LHC) is the largest and most powerful particle accelerator in the world. High energy collisions of protons at the LHC are used to search for new subatomic particles that can answer some of most fundamental and outstanding questions about the nature of our universe. Particle collisions at the LHC are registered using huge detectors designed to take 100 Mega Pixel images 40 million times per second. 

The data collected by the LHC detectors are extremely rich in information, both in the detail and complexity of each event picture, and in the sheer number of events collected. The large and complex LHC dataset and its associated computational and pattern recognition challenges are ideal for the application of advanced data science techniques. In particular, the spectacular recent advances in the fields of artificial intelligence, computer vision, and deep learning have the potential to transfer the way physicists analyze particle collider data. 

In this seminar I will describe some of the key challenges for the analysis of LHC data  and show how novel analysis methods based on deep neural networks and computer vision can significantly enhance the reconstruction capabilities of the LHC, outperforming state-of-the-art methods currently in use. Finally, I will discuss new, more ambitious, LHC analysis problems would like to solve with the use of sophisticated machine learning methods. 

Speaker: Alan Gous, Adjunct Professor at ICME

This talk will be a combination of a few things, in a very short time:
  • An opportunity to introduce myself as a new member of the faculty at ICME,
  • Some personal history on my involvement with Cariden Technologies, a company I co-founded to develop software for IP network traffic management,
  • Talking about how mathematical optimization is relevant to this area of networking, particularly in traffic estimation and route optimization. (More details in a class I will give in the winter!)
Bio: Alan Gous did a Phd in the Statistics Department at Stanford, working with Brad Efron, Stephen Boyd and others. Most of his time after that was spent developing the math and technology used at Cariden. Alan spent the last couple of years transitioning this work into the SDN (Software-Defined Networks) project at Cisco Systems.

Location: Y2E2-101
Time: 12:30-1:20 p.m.
THURSDAY, OCTOBER 27, 2016:
CME 242: 
Mathematical and Computational Finance
Speaker: Ronnie Sircar, Princeton University

Title: Fracking, Mean Field Games & Hotelling’s Rule under Stochastic Demand

Abstract: The dramatic decline in oil prices, from around $110 per barrel in June 2014 to around $30 in January 2016 highlights the importance of competition between different energy producers.  Indeed, the price drop has been primarily attributed to OPEC's strategic decision (until very recently) not to curb its oil production in the face of increased supply of shale gas and oil in the US, which was spurred by the development of fracking technology. Most dynamic Cournot models focus on supply-side factors, such as increased shale oil, and random discoveries. However declining and uncertain demand from China is a major factor driving oil price volatility. We study Cournot games in a stochastic demand environment, and present asymptotic and numerical results, as well as a modified Hotelling's rule for games with stochastic demand.

Location: 200-205
Time: 4:30-5:50p.m.
 
THURSDAY, OCTOBER 27, 2016:
CME 510: Linear Algebra and Optimization Seminar

Speaker: Xiangrui Meng, Databricks Inc.

Title: Implementing Alternating Least Squares on Apache Spark

Abstract: Apache Spark is the most popular open source project for big data analytics, while alternating least squares (ALS) is among the most popular algorithms for collaborative filtering.  We will briefly introduce Spark and the ALS algorithm, present a scalable implementation of ALS on Apache Spark, and share the lessons learned.  We utilize Spark's in-memory caching and a special partitioning strategy to make ALS efficient and scalable.  Optimized internal data storage and other techniques are used to accelerate the computation and to improve JVM performance.  Spark's implementation of ALS runs 10x faster than Apache Mahout's and it scales up to billions of ratings.

Bio: Xiangrui Meng is an Apache Spark PMC member and a software engineer at Databricks.  His main interests are in developing and implementing scalable algorithms for scientific applications.  At Databricks he has been active in the development and maintenance of Spark MLlib. Previously he worked as an applied research engineer at LinkedIn, where he was main developer of an offline machine learning framework in Hadoop MapReduce.  His PhD work at Stanford was on randomized algorithms for large-scale linear regression. 


Location: Y2E2-101
Time: 4:30-5:30 p.m.
OTHER ICME RELATED SEMINARS:
Applied Math Seminar
Wednesday, October 26, 2016
Speaker: Gautum Iyer, Carnegie Mellon University

Location: 384H
Time: 4:15 p.m.


Visit their seminars page at mathematics.stanford.edu for updates.
Copyright © 2016 ICME, All rights reserved.

Our mailing address is:
Institute for Computational and Mathematical Engineering (ICME)
Huang Engineering Center
475 Via Ortega, Suite B060
Stanford, CA 94305

If you would like to unsubscribe from the ICME seminar digest, please email us at 
icme contact@stanford.edu to be completely removed from our mailing list. Thank you.






This email was sent to <<Email Address>>
why did I get this?    unsubscribe from this list    update subscription preferences
Stanford University-Institute for Computational and Mathematical Engineering · 475 Via Ortega, Suite 060 · Stanford, CA 94305 · USA

Email Marketing Powered by Mailchimp