Computational Math in Industry and Beyond Seminar Series (CME 500) - Winter
Mondays at 4:00 - 5:00 PST starting January 11, 2021 to March 15, 2021.
This event is open to:
Computational Math in Industry and Beyond (CME 500)
This seminar series in winter quarter will explore how ICME coursework and research is applied in various organizations around the world. It will feature speakers from ICME affiliate companies and ICME alumni giving technical talks on their use of computational math in their current roles. The CME 500 winter 2021 seminar series is open to all graduate students at Stanford.
Seminars will take place Mondays at 4:00 - 5:00 PST starting January 11, 2021 to March 15, 2021.
Zoom information sent to registrants directly.
Not a registered student, but interested in attending? Fill out this form to receive zoom link for attendance and to receive information on the seminar.
Monday, January 11, 2021
Transfer learning has delivered an “ImageNet moment” for natural language processing (NLP). The explosion of new self-supervised language modeling approaches, coupled with ever increasing data size and compute power, has left the field with a wealth of opportunities to utilize transfer learning to achieve state of the art on a number of tasks. However, directly applying these approaches to real-world problems without a principled approach can often lead to mixed results both from a performance and an engineering standpoint. In this lecture, we will talk through case studies from NLP work at Twilio, and practical advice for making this approach work on real problems.
Luke de Oliveira is Technical Lead for AI and ML at Twilio, where he leads natural language understanding, storage, and runtime orchestration for the Twilio AI business unit. Previously, he was CEO & founder of Vai Technologies, which was acquired by Twilio in 2018. Luke completed his M.S. from Stanford ICME in 2016, and his B.S. in Applied Mathematics from Yale in 2014. He has published numerous academic articles in fields ranging from abstractive summarization to generative models for physics simulation. Luke is active in the startup community though roles as an advisor and investor.
Luke de Oliveira
Technical Lead for AI and ML at Twilio
Monday, January 25, 2021
The Investment Management Fintech Strategies team at Vanguard works with traders, portfolio managers and analysts across Vanguard’s Investment Management Group creating experiments to test new technology and alternative data. In this talk, Karly will go over the process of how she is currently partnering with the trading desk to evaluate how different machine learning techniques can be incorporated into the decision-making process. Specifically, she will cover how they are using alternative data sources to predict bond ratings changes, detail modeling considerations needed when partnering with a trading desk, and go over the modeling design process.
Karly Jerman is a Senior Data Scientist in Vanguard’s Office of Investment Management Fintech Strategies (IMFS), focusing on machine learning and natural language processing techniques. The mission of IMFS is to actively explore new technologies in ways that could dramatically improve Vanguard’s investment performance, make global capital markets work better for investors, and serve as a catalyst for innovation within the Investment Management Group (IMG) at Vanguard. IMG manages over $5.5 trillion in assets globally, across both passive and active strategies in the fixed income, equities, currencies, and derivatives markets.
Before joining Vanguard in 2018, Ms. Jerman earned a M.S. in Management Science and Engineering from Stanford University, and a B.S. and B.A. in Industrial and Systems Engineering from the University of San Diego. Ms. Jerman held positions as a student researcher at the Lucile Packard Children’s Hospital and Thermo Fischer Scientific as well as analytical intern roles at Lockheed Martin, Shiseido, and others.
Senior Data Scientist in Vanguard’s Office of Investment Management Fintech Strategies (IMFS)
Monday, February 1, 2021
In recent years, some of the most talked-about developments in the transportation sector include the use of drones, the introduction of last-mile delivery services, and the use of large-scale mapping data. Along with these new developments comes a host of new problems and trade-offs. We will discuss two such problems and use the "continuous approximation paradigm" to reveal basic insights about those factors that influence them most significantly.
John Gunnar Carlsson is the Kellner Family Associate Professor of Industrial and Systems Engineering at the University of Southern California. He received a Ph.D. in computational mathematics from Stanford University in 2009 and an A.B. in music and mathematics from Harvard College in 2005. He is the recipient of Popular Science magazine's Brilliant 10 Award, the AFOSR Young Investigator Prize, the INFORMS Computing Society (ICS) Prize, and the DARPA Young Faculty Award, and serves as an Associate Editor for Operations Research, Management Science, Transportation Science, and Computers and Operations Research.
John Gunnar Carlsson
Kellner Family Associate Professor of Industrial and Systems Engineering at the University of Southern California
Monday, February 8, 2021
In the past few months, Cisco Research has started focusing on several emerging technology areas of interest to Cisco. In this talk, I will be talking a bit about what these areas are of research interest are, how we are leveraging and engaging with universities for technology, societal, and business impact.
Dr. Ramana Kompella is currently the Head of Research in the Emerging Technology and Incubation group at Cisco. His previous experience includes co-founding successful startups, and providing engineering leadership to build world class products currently in use in several hundreds of customers’ data centers. Prior to his industry roles, he was a tenured faculty at Purdue in Computer Science Department, where he conducted research on systems and networking areas, with multi-million dollar grants from NSF and other industry sources. He co-advised several PhD and Masters students, and has co-authored 70+ publications in top networking and systems conferences such as SIGCOMM. He was the recipient of several awards including the prestigious NSF CAREER award.
Head of Research in the Emerging Technology and Incubation group at Cisco
Monday, February 22, 2021
Recent observations show that both the Greenland and Antarctic ice sheets are losing mass at increasingly rapid rates . In its fourth assessment report (AR4), the Intergovernmental Panel on Climate Change (IPCC) declined to include estimates of future sea-level change from dynamics of the polar ice sheets due to the inability of ice sheet models to mimic or explain observed dynamic behaviors, such as the acceleration and thinning then occurring on several of Greenland’s large outlet glaciers .
In recent years, there has been a push to develop “next generation” land-ice models and codes for integration into global Earth System Models (ESMs) to address this acknowledged limitation. This talk will give an overview of one such next-generation land-ice dynamical code (dycore) known as Albany Land-Ice (ALI) , currently under development at Sandia National Laboratories. Unlike many of its predecessors, ALI: (1) is able to perform realistic, high-resolution, continental scale simulations, (2) is robust, efficient and scalable on next-generation hybrid systems (multi-core, many-core, GPU, Intel Xeon Phi), (3) possesses built-in advanced analysis capabilities (e.g., sensitivity analysis, optimization, uncertainty quantification), and (4) is hooked up to the U.S. Department of Energy’s Energy Exascale Earth System Model (E3SM) as well as NCAR’s Community Earth System Model (CESM).
The ALI dycore is based on the so-called “First-Order Stokes” equations for the ice momentum balance , an attractive alternate to the more expensive “Full Stokes” model. Both the Full Stokes and the First-Order Stokes models assume that ice behaves like a very viscous, shear-thinning, non-Newtonian fluid, similar to lava flow. Following an overview of our land-ice model and project, I will describe some of the algorithms and software we have developed as a part of this project that have contributed to our dycore’s robustness and scalability. These include robust automatic-differentiation-based nonlinear solvers, scalable algebraic-multigrid-based iterative linear solvers , and stable semi-implicit First-Order Stokes-thickness coupling methods. I will also discuss some of the advanced analysis capabilities in ALI, namely a large-scale inversion approach we have developed for obtaining optimal ice initial conditions , our workflow towards quantifying uncertainties in land-ice models, and performance-portability of the ALI code to new and emerging architectures using the Kokkos library [7,9]. I will show results which demonstrate that the ALI dycore is scalable, fast and robust for production-scale land-ice problems on state-of-the-art HPC machines. I will also discuss results from a recent validation study in which ALI was used to simulate the Greenland ice sheet during the period 1991-2013 with realistic climate forcing, and the simulation data were compared with observational data collected by NASA satellites . Finally, I will show some predictive dynamic experiments and simulations we are beginning to perform using ALI.
This work was done in collaboration with Luca Bertagna, Max Carlson, Irina Demeshko, Mike Eldred, Matt Hoffman, John Jakeman, Mauro Perego, Steve Price, Andy Salinger, Chad Sockwell, Ray Tuminaro and Jerry Watkins.
Dr. Irina Tezaur (f.k.a. Dr. Irina Kalashnikova) is a Principal Member of Technical Staff (PMTS) in the Extreme Scales Data Science & Analytics Department (Org. 8759) at Sandia National Laboratories in Livermore, CA. Prior to joining this group, from October 2011 to September 2014, she was SMTS in the Computational Mathematics Department (Org. 1442) at Sandia in Albuquerque, NM. She received her Ph.D. in Computational and Mathematical Engineering (CME) from Stanford University in 2011. Her advisor at Stanford was Professor Charbel Farhat and I was a member of the Farhat Research Group (FRG). Her Bachelors and Masters degrees are in pure mathematics, awarded by the University of Pennsylvania in 2006. Dr. Tezaur’s research interests are numerical solution to PDEs, mixed/hybrid finite element methods, stability and convergence properties of numerical methods, Reduced Order Modeling (ROM) and simulation-based analysis of fluid-structure interaction that she currently applies to climate modeling.
Principal Member of Technical Staff (PMTS) in the Extreme Scales Data Science and Analytics Department (Org. 8759) at Sandia National Laboratories in Livermore, CA
Monday, March 1, 2021
How can we apply technology to improve patient care? This session will provide a brief introduction to some of the solutions being developed at Google Health.
Lan joined Google Health as a Data Scientist in 2019 after completing her doctoral degree. As an ICME student, she was working with Prof. Susan Holmes on statistical methods for analyzing high-dimensional heterogeneous data with applications to genomics and microbiome studies. Before coming to Stanford, she did her undergraduate studies in applied mathematics and economics at Caltech.
Data Scientist, Google Health
Monday, March 15, 2021
AI has been applied for personalization for quite some time in the consumer domain for use cases such as movie recommendations on Netflix, and book and product recommendations on Amazon, and various other retailers. That personalization is primarily based on transaction data, which is structured in nature. The rise of social media data on platforms such as Twitter, Facebook, and Instagram, where users share/broadcast their daily activities and have social conversations with friends, is providing increased access to unstructured user data that can be analyzed (with users' permission) for personalization. This is enabling the development of next-generation User Models power by the latest advances in machine learning, wherein people’s personality traits, in-the-moment emotions, sentiment towards products, services, their communication tones, and intentions can be derived from this social media data. In this talk, I will discuss our work at IBM on building User Models that is being applied in various enterprise use cases including in the development of personalized Chatbots, personalized product & service recommendations. I’ll also discuss the genuine privacy concerns around using such user data and models and the new business models that are emerging, such as ‘bring-your-own-data, to address these concerns.
Rama Akkiraju is an IBM Fellow, Master Inventor and IBM Academy Member, and a Director, at IBM’s Watson Division where she leads the AI operations team with a mission to scale AI for Enterprises. Prior to this, Rama led the AI mission of enabling natural, personalized and compassionate conversations between computers and humans. Rama has been named by Forbes as one of the ‘Top 20 Women in AI Research’ in May 2017, has been featured in ‘A-Team in AI’ by Fortune magazine in July 2018 and named ‘Top 10 pioneering women in AI and Machine Learning’ by Enterprise Management 360. In her career, Rama has worked on agent-based decision support systems, electronic market places, and semantic Web services, for which she led a World-Wide-Web (W3C) standard. Rama has co-authored 4 book chapters, and over 100 technical papers. Rama has 18 issued patents and 25+ pending. She is the recipient of 3 best paper awards in AI and Operations Research. Rama holds a Masters degree in Computer Science and has received a gold medal from New York University for her MBA for highest academic excellence. Rama served as the President for ISSIP, a Service Science professional society for 2018 and continues to actively drive AI projects through this professional society.
IBM Fellow, Master Inventor and IBM Academy Member, and a Director, at IBM’s Watson Division where she leads the AI operations team with a mission to scale AI for Enterprises