Skip to:

Past Events

View events by calendar year
Enter one or more keywords to search for
Filter by event type
Tuesday, October 20, 2015 -
12:30pm to 1:30pm

 

 Speaker: Alison Marsden, Associate Professor of Pediatrics and Bioengineering and, by courtesy, Mechanical Engineering at Stanford University

About the Speaker: Alison Marsden is affiliated with the Department of Pediatrics, Department of Bioengineering, and Institute of Computational and Mathematical Engineering at Stanford University. Marsden's lab, The Cardiovascular Biomechanics Computation Lab at Stanford, develops novel computational methods for the study of cardiovascular disease progression, surgical methods, and medical devices. Their interests include: cardiovascular disease and biofluid mechanics, shape optimization for complex flows, pediatric cardiology and congenital heart disease, vascular surgery, derivative-free optimization methods, uncertainty quantification, multiscale modeling, vascular design principles, vascular growth and remodeling, thrombosis, cardiovascular device design, Kawasaki Disease, bypass graft optimization, and ventricular assist devices. 

Please visit: http://web.stanford.edu/~amarsden/MarsdenLab/Research.html for the latest information on her research, publications, and other news.

 

Friday, October 16, 2015 -
11:00am to 1:00pm

Data Visualization Center Open House

Join us for the HIVE Xperience, an open house of Stanford's immersive visualization environment

11:00am- 1:00pm

Friday, October 16, 2015

Huang Engineering Center, room 050

The HIVE Xperience is your chance to see the HIVE in action:

  • Discover the nuances that can be uncovered in viewing your research images in large-scale and ultra-high resolution
  • See samples of collaborative visualization and research projects already taking place in the HIVE
  • Bring your own images!  See what your high-resolution images look like on the HIVE by bringing your files with you to the event (we recommend you upload your images to Stanford Box for easy access https://itservices.stanford.edu/service/box)

No R.S.V.P. is needed; light refreshments will be served

The HANA Immersive Visualization Environment (HIVE), was built in partnership between the Institute for Computational and Mathematical Engineering (ICME), the Army High Performance Computing Research Center (AHPCRC), and with the generous support of SAP.  Located in Huang 050, the HIVE may be reserved by Stanford faculty, students, and staff, free of charge, for research visualization projects and courses.

The HIVE features a 10-foot-tall by 24-feet-wide display with 13440x5400 resolution and 72 million total active pixels.  Researchers may use multiple displays simultaneously to investigate various aspects of data collection, simulation, and visualization and to zoom in to see detail at previously unheard-of-levels.  Learn more:  hive.stanford.edu

 

 
Thursday, October 15, 2015 -
4:30pm to 5:45pm

Speaker: TBD

Thursday, October 15, 2015 -
4:30pm to 5:30pm

Speaker: Tim Levandoski, Eurex

 

Tuesday, October 13, 2015 -
12:30pm to 1:15pm

Speaker: Ron Dror, Associate Professor of Computer Science at Stanford University

Title: "Computational Biology in Three (or Four) Dimensions"

Ron Dror is an Associate Professor of Computer Science and (by courtesy) Molecular and Cellular Physiology at Stanford University, where he is also affiliated with the Institute for Computational and Mathematical Engineering, Bio-X, ChEM-H, and the Biophysics Program. Dr. Dror's research at Stanford addresses a broad set of computational biology problems related to the spatial organization and dynamics of biomolecules and cells.

Before joining Stanford in March 2014, Dr. Dror served as second-in-command of D. E. Shaw Research, a hundred-person company, having joined in 2002 as its first hire. At DESRES, he focused on high-performance computing and biomolecular simulation—in particular, developing technology that accelerates molecular dynamics simulations by orders of magnitude, and applying these simulations to the study of protein function, protein folding, and protein-drug interactions (part of a project highlighted by Science as one of the top 10 scientific breakthroughs of 2010).

Dr. Dror earned a PhD in Electrical Engineering and Computer Science at MIT, an MPhil in Biological Sciences as a Churchill Scholar at the University of Cambridge, and both a BA in Mathematics and a BS in Electrical and Computer Engineering at Rice University, summa cum laude. As a student, he worked in genomics, vision, image analysis, and neuroscience. He has been awarded a Fulbright Scholarship and fellowships from the National Science Foundation, the Department of Defense, and the Whitaker Foundation, as well as a Gordon Bell Prize and several Best Paper awards.

Visit Ron Dror's website at: http://cs.stanford.edu/people/rondror/

Monday, October 12, 2015 -
4:30pm to 5:30pm

Speaker: Gianluca Geraci

 

Title: “Towards multi-fidelity approaches for multi-level Monte Carlo techniques”
G. Geraci, H. Fairbanks, A. Doostan and G. Iaccarino
 
Abstract: In the last few years efficient algorithms have been proposed for the propagation of uncertainties through numerical codes. Despite the high efficiency reached, UQ studies of real engineering devices still remain challenging due to the large number of computationally expensive multi-physics simulations required. Recently, the multi-level Monte Carlo method (MLMC) emerged as a novel technique able to retain the robustness of the MC method, but increasing also its efficiency. Therefore, the MLMC technique is able to quantify statistics requiring a small number of deterministic realizations compared to MC. Nonetheless, the overall numerical cost still remains  prohibitive in presence of multi physics problems. In this talk, an overview on the ongoing activities related  to the acceleration of MLMC will be provided. The pivotal idea is to further reduce the MLMC numerical cost  exploiting (in different ways) the information produced by realizations of less accurate, but  computationally inexpensive, models.
Friday, October 9, 2015 -
4:00pm to 7:00pm

Career Info Night and Networking Reception with Industry Partners

Meet ICME partners during this informal event that combines small group discussions and a networking reception.  Hear directly from partners about opportunities for students and graduates at their organizations and enjoy informal networking opportunities with partners before the Xtend interview day. 

To find out more about this event, visit the Xtend student website, or look at the full Xtend timeline, complete with workshops and key deadlines.

Thursday, October 8, 2015 -
4:30pm to 5:45pm

Speaker: Youngsoo Choi, Department of Mechanical Engineering at Stanford University

Title: "Gradient-based Constrained Optimization Using a Database of Linear Reduced-Order Models"
 
Abstract: A novel methodology for accelerating the solution of PDE-constrained optimization is introduced.  It is based on offline construction of a database of local ROMs, and online interpolation within the database. The online flexibility of the ROM database approach makes it amenable to speeding up optimization-intensive applications such as robust optimization, multi-objective optimization, and multi-start strategies for locating global optima.  The accuracy of the ROM database model can be tuned in the offline phase through a greedy procedure.  In this work, a novel greedy algorithm based on a saturation assumption is introduced to speed-up the construction procedure.  The ROM database approach is applied to a realistic wing design problem and leads to a large online speed-up.
Wednesday, October 7, 2015 -
9:00am to 4:00pm

Registration is NOW OPEN for this year's NVIDIA Tech Talks event and Hands-On Labs.  

Stanford is a NVIDIA Center of Excellence and each year we get together in the Huang Engineering Center with the morning full of tech talks and afternoon with hands-on labs.  This year we have an exciting line-up of experts and entrepreneurs in the fields of data science.

  • What is the event:  Tech Talks and Hands on Labs in Analytics, Deep Learning and Computer Vision.
  • Who is it for: Undergraduate, graduate students, postdocs, researchers, and professors.
  • This year's theme:  End to End Analytics in Research, Industry, and Business
  • Wednesday, October 7, 2015, 9:00am- 4:00pm
  • Stanford University's Huang Engineering Center- Mackenzie Room
  • Coffee and lunch provided to attendees 
  • Please bring a laptop for remote access to GPU Cluster – no GPU required in your laptop

Register for all or part of the day:  http://www.eventbrite.com/e/nvidia-tech-talks-and-hands-on-labs-at-stanford-university-tickets-18729102249?aff=Stanford 


Agenda:

Tech Talks – 9AM to 11:45AM

Plenary talks where you will hear from fellow researchers and industry startups on tools, research discoveries, best practices and trends using GPUs for data science.

  • Deep Learning, Open Research Questions – Boris Ginsburg, Deep Learning Chief Scientist at NVIDIA
  • Cognitive Computing Platforms – Patrick Ehlen, Chief Scientist at Loop AI Labs
  • Self Driving Cars – Urs Muller, Chief Architect at NVIDIA
  • Caffe-on-Spark – Cyprien Noel, Machine Learning and Distributed Systems Engineer at Yahoo 
  • Character-level language models with RNNs – Andrej Karpathy, Stanford Computer Science Ph.D. Student

Demos and Lunch – 11:50AM to 1PM

Live demos of latest deep learning and computer vision technology and connect with NVIDIA engineers.

Hands-on Labs – 1PM to 4PM

Hands-on exercises to get you started using GPUs for accelerating scientific applications in the accelerated computing lab or for designing your neural network in the deep learning lab.

  • Track 1 – Introduction to Accelerated Computing on GPUs, or                                                             
  • Track 2 – Introduction to Deep Learning on GPUs
Tuesday, October 6, 2015 -
12:30pm to 1:30pm

 

Speaker: Simone Marras

Title: "Numerical Challenges In Computational Fluid Dynamics For the Atmosphere and the Ocean"
 
Abstract: In the first part of this talk we will report on the evolution of numerical weather prediction (NWP) and underline the numerical issues that still represent the most active topics of interest among researchers in the field of atmospheric modeling. The discussion will cover the connection between NWP high performance computing, turbulence modeling for atmospheric flows, and grid generation for high-resolution orographic weather. In the second part of the talk we will address current research in tsunami mitigation and oceanic coastal flows. The numerical challenges in this fascinating multiscale problems will be introduced.
Monday, October 5, 2015 -
4:30pm to 5:30pm

Speaker: Zach Riggins del Rosario

Title: "The PI Active Subspace and Classical Dimension Analysis"
 
Abstract: How do we understand a model if the simulation runs for the age of the universe? The study of Dimension Reduction helps us to make impossible problems tractable. This study aims to defray the dreaded Curse of Dimensionality by reducing the the number of input parameters, through various strategies. One such strategy to achieve Dimension Reduction is the Active Subspace procedure.
 
Active Subspaces study the gradient of a function, in order to identify linear combinations of parameters that account for variability in some Quantity of Interest. In problems of practical interest, Paul Constantine and his colleagues have found that problems with as many as 50 dimensions reduce to a 1 dimensional Active Subspace -- a huge reduction! Though Active Subspaces are powerful, their structure of linear combinations is difficult to interpret -- and in some some physical systems -- not well suited to describe the inputs. Using classical results from Dimensional Analysis -- namely, the Buckingham Pi Theorem -- we develop an alternative formulation of Active Subspaces. This new formulation, called the Pi Active Subspace, seeks products of parameters and chases down the governing dimensionless parameters of a system.
Monday, October 5, 2015 -
2:30pm to 3:30pm

 

Speaker: Peter Carr

Title: "Analogies Between Bond Yields and Implied Volatilities"

Abstract: Intuitions about bond yields can be used to model implied volatilities and vice versa. In particular, we link the shape of the yield curve to the graph of (normal) implied volatilities across strikes.

About the Speaker: Dr. Peter Carr is a Managing Director at Morgan Stanley with 15 years of experience in the derivatives industry. He was also a finance professor for 8 years at Cornell University, after obtaining his PhD from UCLA in 1989. He is presently the Executive Director of the Math Finance program at NYU's Courant Institute, the Treasurer of the Bachelier Finance Society, and a trustee for the Museum of Mathematics in New York. He has over 70 publications in academic and industry-oriented journals and serves as an associate editor for 8 journals related to mathematical finance. He was selected as Quant of the Year by Risk Magazine in 2003 and shared in the ISA Medal for Science in 2008. Last December, the International Association of Financial Engineers (IAFE) and Sungard jointly announced that they have selected Dr. Carr as its 2010 Financial Engineer of the Year. Visit his website at: http://math.nyu.edu/research/carrp/

Thursday, October 1, 2015 -
4:30pm to 5:45pm

Speaker: Jennifer Scott, Rutherford Appleton Laboratory, England

Tuesday, September 29, 2015 -
12:30pm to 1:15pm

Speaker: Mykel Kochenderfer, Assistant Professor of Aeronautics and Astronautics at Stanford University

Thursday, September 24, 2015 -
4:30pm to 5:45pm

Speaker: Eric Darve, Associate Professor of Mechanical Engineering at Stanford University

Tuesday, September 22, 2015 -
12:30pm to 1:15pm

No seminar this week. CME 300 begins on September 29,2015.

Thursday, September 17, 2015 -
9:00am to 4:00pm

Sign-up now for the ICME Summer Refresher!  This review session spans four days and covers basic mathematical material relevant to first-year ICME classes and applied mathematics topics in science and engineering. The ICME Summer Refresher is suitable for incoming ICME graduate students and other graduate students with a technical background. This year, we are offering courses in numerical linear algebra, partial differential equations, statistics and probability, and discrete mathematics and algorithms. Students can enroll in any or all of these courses. 

Details on each session are available here: http://web.stanford.edu/~lanhuong/refresher/

Register here: https://docs.google.com/forms/d/18MTEj7VAtpwFpkQjnLMx2ca2-DWc5hwNdNTI5LE...

Wednesday, September 16, 2015 -
9:00am to 4:00pm

Sign-up now for the ICME Summer Refresher!  This review session spans four days and covers basic mathematical material relevant to first-year ICME classes and applied mathematics topics in science and engineering. The ICME Summer Refresher is suitable for incoming ICME graduate students and other graduate students with a technical background. This year, we are offering courses in numerical linear algebra, partial differential equations, statistics and probability, and discrete mathematics and algorithms. Students can enroll in any or all of these courses. 

Details on each session are available here: http://web.stanford.edu/~lanhuong/refresher/

Register here: https://docs.google.com/forms/d/18MTEj7VAtpwFpkQjnLMx2ca2-DWc5hwNdNTI5LE...

Tuesday, September 15, 2015 -
9:00am to 4:00pm

Sign-up now for the ICME Summer Refresher!  This review session spans four days and covers basic mathematical material relevant to first-year ICME classes and applied mathematics topics in science and engineering. The ICME Summer Refresher is suitable for incoming ICME graduate students and other graduate students with a technical background. This year, we are offering courses in numerical linear algebra, partial differential equations, statistics and probability, and discrete mathematics and algorithms. Students can enroll in any or all of these courses. 

Details on each session are available here: http://web.stanford.edu/~lanhuong/refresher/

Register here: https://docs.google.com/forms/d/18MTEj7VAtpwFpkQjnLMx2ca2-DWc5hwNdNTI5LE...

Monday, September 14, 2015 -
9:00am to 4:00pm

Sign-up now for the ICME Summer Refresher!  This review session spans four days and covers basic mathematical material relevant to first-year ICME classes and applied mathematics topics in science and engineering. The ICME Summer Refresher is suitable for incoming ICME graduate students and other graduate students with a technical background. This year, we are offering courses in numerical linear algebra, partial differential equations, statistics and probability, and discrete mathematics and algorithms. Students can enroll in any or all of these courses. 

Details on each session are available here: http://web.stanford.edu/~lanhuong/refresher/

Register here: https://docs.google.com/forms/d/18MTEj7VAtpwFpkQjnLMx2ca2-DWc5hwNdNTI5LE...

Monday, July 27, 2015 - 9:00am to Friday, August 28, 2015 - 5:00pm

ICME will offer a series of week-long workshops on the Fundamentals of Data Science in July and August 2015. 

Registration will be on a first-come first serve basis to students and ICME External Partners. Other community members are welcome to attend, space permitting.  Please note that these are not Stanford for-credit courses. 

This summer, the workshops include

Stats Week: July 27 (Monday) - July 31 (Friday)

Week: Aug 18 (Tuesday) - Aug 21 (Friday)

Week: Aug 24 (Monday) - Aug 28 (Friday)

Details are available here:  https://icme.stanford.edu/academics/summer-workshops 

Monday, June 29, 2015 - 8:00am to Thursday, July 2, 2015 - 5:00pm

The 2015 SIAM Conference on Mathematical and Computational Issues in the Geosciences will take place at Stanford University this summer, June 29- July 2, 2015.  

Short Courses

The day before the conference, Sunday, June 28, 2015,  Stanford will host a special series of short courses, the Summer Sunday Shortcourse Sensation, which are free of charge and open for registration now: https://www.surveymonkey.com/r/BQC92VS.  


About the Conference

From points of view ranging from science to public policy, there is a growing interest in modeling and simulation of geosystems and their applications. Some examples include petroleum exploration and recovery, underground waste disposal and cleanup of hazardous waste, earthquake prediction, weather prediction, and global climate change. Such modeling is fundamentally interdisciplinary; physical and mathematical modeling at appropriate scales, physical experiments, mathematical theory, probability and statistics, numerical approximations, and large-scale computational algorithms all have important roles to play.

This conference facilitates communication between scientists of varying backgrounds and work environments facing similar issues in different fields, and provides a forum in which advances in parts of the larger modeling picture can become known to those working in other parts. These kinds of interactions are needed for meaningful progress in understanding and predicting complex physical phenomena in the geosciences. 

Register now at https://www.siam.org/meetings/gs15/regform.php-- pre-registration deadline is before June 1, 2015.  

ICME Director Margot Gerritsen is Co-Chair of the conference organizing committee along with along with Knut-Andreas Lie, SINTEF, Norway, and Iuliu Sorin Pop, Eindhoven University of Technology, The Netherlands.

More information is available on the SIAM website at http://www.siam.org/meetings/gs15/

Friday, June 5, 2015 -
4:00pm to 5:00pm

No TGIF on 6/5. TGIF will return in Autumn 2015.

Thursday, June 4, 2015 -
4:15pm to 5:30pm

There is no seminar on 6/4. CME 510 will return in Autumn 2015.

Tuesday, June 2, 2015 -
12:15pm to 1:15pm

There is no seminar on 6/2. CME 300 will resume in Autumn 2015.

Monday, June 1, 2015 -
4:15pm to 5:30pm

CME 500 will return in Autumn 2015.

Friday, May 29, 2015 -
4:00pm to 5:00pm

Topic: What we are looking for in qualifying exams.

Thursday, May 28, 2015 -
4:15pm to 5:30pm

Speaker: Jim Lambers, University of Southern Mississippi
                   James.Lambers@usm.edu
                   http://www.math.usm.edu/lambers/

Approximation of the Scattering Amplitude using Nonsymmetric Saddle-Point Matrices

We look at iterative methods for solving the forward (Ax = b) and adjoint (A'y = g) systems of linear equations to approximate the scattering amplitude, defined by g'x = y'b.  Based on an idea first proposed by Gene Golub, we use a conjugate gradient-like iteration for a nonsymmetric saddle-point matrix that is constructed to have a real positive spectrum.  Numerical experiments show that this method is more consistent than known methods for computing the scattering amplitude such as GLSQR or QMR.  We then demonstrate that when combined with known preconditioning techniques, the proposed method exhibits more rapid convergence than state-of-the-art iterative methods for nonsymmetric systems.  We also examine the use of techniques from "matrices, moments, and quadrature", adapted to the nonsymmetric saddle-point case, to compute the scattering amplitude directly without solving either system explicitly.

Tuesday, May 26, 2015 -
12:15pm to 1:15pm

Speaker: Matthias Ihme
High-Order Numerical Methods and Physical Modeling for Chemically Reacting and Turbulent Flows

Monday, May 25, 2015 -
4:15pm to 5:30pm

Memorial Day (No seminar)

Friday, May 22, 2015 -
4:00pm to 5:00pm

No TGIF this week due to the ICME Xpo and Xtravaganza

Friday, May 22, 2015 - 1:00pm
This May at the ICME Xpo, get an up-close and inside look at current research and future plans for ICME faculty and students.   ICME is engaged with over 50 faculty from 18 departments throughout Stanford.  This is a unique opportunity to see how computational mathematics, data science, scientific computing, and related fields are applied across a wide range of domain areas.  
 

ICME Xpo features an afternoon poster session followed by a series of faculty vision talks and promises plenty of opportunities to connect with ICME faculty and students, alumni, and partners from industry and laboratories.  

Sample topics to be presented at the poster session include: robot motion optimization, neural networks, single-molecule numerical analysis, optimization of net present value under uncertainty, motif based spectral clustering, and distributed acoustic sensing.  

View the agenda for the event here.

View a full list of the Xpo poster session presentations here.  


List of faculty members presenting at the faculty vision talks:
Margot Gerritsen, Director of the Institute for Computational and Mathematical Engineering and 
Associate Professor of Energy Resources Engineering
Eric Darve, Associate Professor of Mechanical Engineering
Ramesh Johari, Associate Professor of Management Science and Engineering
Ron Dror, Associate Professor of Computer Science and the Institute for Computational and Mathematical Engineering
Jack Poulson, Assistant Professor of Mathematics and the Institute for Computational and Mathematical Engineering
Vijay Pande, Professor of Chemistry
Chris Re, Assistant Professor of Computer Science
Lexing Ying, Professor of Mathematics and the Institute for Computational and Mathematical Engineering
Alison Marsden, Associate Professor of Pediatrics and the Institute for Computational and Mathematical Engineering


An annual ICME event:

Friday, May 22, 2015, 1:00- 6:30pm
1:00- 3:00 Poster Session
3:00- 5:00 Faculty Vision Talks
5:00- 6:30 Reception
 
Huang Engineering Center, Mackenzie Room (rm. 300).  Parking/ Directions
 
Who should attend:
 
 
Registration for this event is now closed. 
 
Following the Xpo, ICME will celebrate the end of the year with our annual Xtravaganza celebration, complete with student awards, music, food, and lots of wonderful company.  Join us as we bid farewell to another fantastic year and look forward to all of the great things to come in 2015-16.  
 
Questions? Contact us at icme-contact@stanford.edu or 650-724-3313.
Thursday, May 21, 2015 -
4:15pm to 5:30pm

Speaker: Esmond Ng, LBNL
                  EGNg@lbl.gov
                  http://crd.lbl.gov/~EGNg

Sparse Gaussian Elimination: Myths and Facts

Solving sparse systems of linear equations is at the heart of many large-scale scientific and engineering computation.  Sparse direct methods, which are based on Gaussian elimination, are known to be memory bound and hence are often discarded as the methods of choice for large-scale computation.  Yet, they are reliable because they terminate after a finite number of operations.  We will consider several common criticisms of sparse Gaussian elimination and comment on their validity.  The talk will touch on several aspects of sparse Gaussian elimination, such as efficiency, data structures, graph theory, complexity analysis, and computer architectures.
Tuesday, May 19, 2015 -
12:15pm to 1:15pm

Speakers: Michal Kosinski, postdoctoral scholar in the Department of Computer Science at Stanford University

                      Reza Zadeh, consulting professor in the Institute for Computational and Mathematical Engineering


Michal Kosinski
Predicting Personality from Digital Footprints

Personality traits form a key driver behind people’s behavior, cognitions, motivations, and emotions; therefore, assessing others’ personality is a basic social skill and a crucial element of successful social interactions. However, based on a sample of over a million participants, I show that personality judgments made by computers―and based on generic and pervasive digital footprints (Facebook Likes)―are more accurate than those made by participants’ friends, family members, and even romantic partners. Furthermore, compared with humans, computers achieve higher inter-judge agreement and superior external validity (i.e. are better at predicting life outcomes). In some cases, computer-based personality judgments are even more valid than self-reported personality scores. I conclude by discussing the consequences of computers outpacing humans in this basic social-cognitive skill.

Reza Zadeh
Distributed Algorithms

Reza Zadeh is a consulting professor in the Institute for Computational and Mathematical Engineering, conducting research and teaching courses targeting doctorate students. His focus is Machine Learning, Distributed Computing, and Discrete Applied Mathematics.
Monday, May 18, 2015 -
4:15pm to 5:30pm

Speaker: Chao Chen
Black-box fast multipole methods in dislocation dynamic simulation

We introduce two new black-box fast multipole methods (FMM) to speed up the computation of pairwise dislocation interactions in both isotropic and anisotropic formulations. One advantage of these methods is that both can be applied to kernels that are only known only numerically, i.e., the analytical expressions are not available or too complicated. The results of our tests show that both of our new methods achieve great speedup compared to existing implementations in ParaDis, a parallel dislocation dynamic simulation code.
Friday, May 15, 2015 -
4:00pm to 5:00pm

Topic: Academic careers, the good and the bad

Thursday, May 14, 2015 -
4:15pm to 5:30pm

Speaker: Paul Constantine, Colorado School of Mines
                 pconstan@mines.edu
                       
http://inside.mines.edu/~pconsta

Active subspaces for dimension reduction in approximation, optimization, and integration
 
Scientists and engineers use computer simulations to study relationships between a physical model's input parameters and its outputs. However, thorough parameter studies---e.g., constructing response surfaces, optimizing, or integrating---are challenging, if not impossible, when the simulation is expensive and the model has several inputs. To enable studies in these instances, the engineer may attempt to reduce the dimension of the model's input parameter space. Active subspaces are an emerging set of dimension reduction tools that identify important directions in the parameter space. I will describe techniques for discovering a model's active subspace and propose methods for exploiting the reduced dimension to enable
otherwise infeasible parameter studies.  For more info, see www.activesubspaces.org
 
Tuesday, May 12, 2015 -
12:15pm to 1:15pm

CME 300 has been canceled this week and will resume on 5/19.

Monday, May 11, 2015 -
4:15pm to 5:30pm

Speaker: Xiaotong Suo
An Ordered Lasso and Sparse Time-lagged Regression

We consider a regression scenario where it is natural to impose an order constraint on the coefficients. We propose an order-constrained version of L1-regularized regression (lasso) for this problem, and show how to solve it efficiently using the well-known Pool Adjacent Violators Algorithm as its proximal operator. The main application of this idea is to time-lagged regression, where we predict an outcome at time t from features at the previous K time points. In this setting it is natural to assume that the coefficients decay as we move farther away from t, and hence the order constraint is reasonable. Potential application areas include financial time series and prediction of dynamic patient outcomes based on clinical measurements. We illustrate this idea on real and simulated data.
Friday, May 8, 2015 -
4:00pm to 5:00pm

Topic: Tips for fellowship proposal writing

Thursday, May 7, 2015 -
4:15pm to 5:30pm

Speaker: Damek Davis, UCLA
                       damek@math.ucla.edu

                  http://www.math.ucla.edu/~damek/

A Three-Operator Splitting Scheme and its Optimization Applications

For over 50 years, operator-splitting schemes have been used to solve PDE, feasibility problems, and more recently, large-scale problems in data analysis.  Despite much development, it is known that most existing splitting schemes reduce to one of three basic schemes, each introduced between 15 and 36 years ago.

We introduce a new splitting scheme that extends the Douglas-Rachford and forward-backward splitting schemes to monotone inclusions with three operators, one of which is cocoercive.  We discuss why this algorithm works, derive several special cases, including a simple three-block ADMM algorithm, and introduce an acceleration that achieves the optimal rate of convergence for strongly monotone inclusions. Finally, we discuss several applications and future research directions.

Tuesday, May 5, 2015 -
12:15pm to 1:15pm

Speaker: Hung Le

Monday, May 4, 2015 -
4:15pm to 5:30pm

Speaker: Fayadhoi Ibrahima and Ruoxi Wang


Fayadhoi Ibrahima
Multi-point distribution function of saturation for stochastic two-phase flow

We present a procedure to analytically obtain the multi-point Cumulative Distribution Function (CDF) of the water-saturation for the stochastic Buckley-Leverett problem when the total-velocity field is the random input. The multi-point CDF is determined by solving a simpler intermediate PDE, then taking the ensemble average of its solution. We illustrate the multi-point distribution method in one spatial dimension and for two points. The two-point CDF can notably be used to compute the covariance function of the water-saturation, which is an essential component of data assimilation. We compare the novel two-point CDF method with Monte Carlo simulations.

Ruoxi Wang

Structured Block Basis Factorization for Scalable Kernel Matrix Evaluation
 
Kernel matrices are popular in machine learning and scientific computing, but they are limited by their quadratic complexity in both construction and storage. It is well-known that as one varies the kernel parameter, \emph{e.g.}, the width parameter in radial basis function kernels, the kernel matrix changes from a smooth low-rank kernel to a diagonally-dominant and then fully-diagonal kernel. Low-rank approximation methods have been widely-studied, mostly in the first case, to reduce the memory storage and the cost of computing matrix-vector products. Here, we use ideas from scientific computing to propose an extension of these methods to situations where the matrix is not well-approximated by a low-rank matrix. In particular, we construct an efficient block low-rank approximation method---which we call the Block Basis Factorization---and we show that it has $\mathcal{O}(n)$ complexity in both time and memory. Our method works for a wide range of kernel parameters, extending the domain of applicability of low-rank approximation methods, and our empirical results demonstrate the stability (small standard deviation in error) and superiority over current state-of-art kernel approximation algorithms.
Friday, May 1, 2015 -
4:00pm to 5:00pm

Speaker: Claude Reichard from the Technical Communications Program
Topic: Writing resources, common challenges when writing technical reports/papers/proposals, and how to write efficient posters

Please note that the location for this TGIF has changed. This TGIF will take place in the ICME lobby in Huang, B060.

Thursday, April 30, 2015 -
4:15pm to 5:30pm

Speaker: Holly Jin, Cardinal Optimization
                            http://cardinalopt.com/
                          
  holly.jin@cardinalopt.com

Fleet Resource Allocation and Route Optimizer
 
Cardinal Optimization provides resource allocation and route optimization software for Fleet Management Systems, where there is a need to schedule a fleet of vehicles for a given list of tasks involving each specified point on a map region.  Our patented algorithms are designed for balancing workload optimally across vehicles and generating optimal routes for them.  We will discuss the mathematical methods and their implementation and deployment.
Tuesday, April 28, 2015 -
12:15pm to 1:15pm

Speakers: Yinyu Ye and Gianluca Iaccarino


Yinyu Ye is a professor in Management Science & Engineering and, by courtesy, Electrical Engineering. Professor Ye's research interests lie in the areas of optimization, complexity theory, algorithm design and analysis, and applications of mathematical programming, operations research and system engineering. He is also interested in developing optimization software for various real-world applications.

Gianluca Iaccarino is a professor in the Department of Mechanical Engineering and ICME. Professor Iaccarino will be presenting a talk titled "Predictive engineering: supercomputers, big-data, and the crystal ball”. His research interests include numerical methods for fluid mechanics, physical models for laminar/turbulent flows, and uncertainty quantification in computational science.

 

Monday, April 27, 2015 -
4:15pm to 5:30pm

Speaker: Austin Benson
Spectral Clustering with Tensors

Two of the fundamental analyses of networks are clustering (partitioning, community detection) and the frequency of network motifs, or patterns of links between nodes. However, these analyses are disjoint as community structure is typically defined by link relationships and ignore motifs. Here, we unify these two ideas through motif-based spectral clustering.  Our framework represents motifs through adjacency tensors and uses Markov chains derived from these tensors to find motif-based clusters.
Friday, April 24, 2015 -
4:00pm to 5:00pm

Speaker: Michael Minion
Topic: Tools for writing papers more efficiently

Thursday, April 23, 2015 -
4:15pm to 5:30pm

Speaker: Harald Steck, Netflix
                  hsteck@netflix.com
                  
 http://videolectures.net/harald_steck/

Matrix Factorization for Movie Recommendations

The Netflix recommender system for movies and TV shows is comprised of an ensemble of models.  The talk will focus on matrix factorization models.  Users' feedback data (eg, played or rated titles) can be represented in a matrix involving users and movies/TV shows.  Such a matrix has several interesting properties: (1) it is sparse (ie each user rated only a small number of titles), (2) it is tall and thin (ie there are many more users than titles), and (3) there are various selection biases in the data.  The latter means that there is information in which entries are present in the sparse matrix (besides the information in the entries' values).  An example of a selection bias is that a user tends to rate items that they like or know, resulting in an under-representation of low rating values in the data. Another example is that users tend to rate movies with similar release-years together.  I will discuss different matrix factorization models tailored to these properties of the data.  The models are optimized by stochastic gradient descent toward a personalized ranking of the movies for each user, rather than toward predicting missing entries in the matrix.

Tuesday, April 21, 2015 -
12:15pm to 1:15pm

Speaker: Parviz Moin
Computation of Turbulence & Predictive Science at Stanford

Parviz Moin is the founding director of the Center for Turbulence Research at Stanford and Ames. Established in 1987 as a research consortium between NASA and Stanford, Center for Turbulence Research is devoted to fundamental studies of turbulent flows. Center of Turbulence Research is widely recognized as the international focal point for turbulence research, attracting diverse groups of researchers from engineering, mathematics and physics.

Professor Moin pioneered the use of direct and Large Eddy Simulation techniques for the study of turbulence physics, control and modelling concepts and has written widely on the structure of turbulent shear flows. His current interests include: interaction of turbulent flows and shock waves, aerodynamic noise and hydroacoustics, aerooptics, combustion, numerical analysis, turbulence control, large eddy simulation and parallel computing. He is an Editor of the Annual Review of Fluid Mechanics and Associate Editor of Physics of Fluids, Journal of Computational Physics.

Pages