Skip to content Skip to navigation

ICME Stories

The students, faculty and alumni of ICME are a diverse, interdisciplinary group, with interests and research in bioinformatics, geosciences, computational finance, and more. But the one thing that unites them is a love for computational math and scientific computing.  Here are some of their stories:

Big Math in Action.

Eileen Martin, ICME PhD Candidate

Using Fast Algorithms to Analyze Permafrost

“A lot of the work I do is motivated by geoscience examples. I’m working with a team trying to track where permafrost is thawing underneath infrastructure. The surface underneath a lot of runways and railways in arctic areas is really heterogeneous. There are some patches that are frozen solid permafrost, and other patches that are thawed and kind of squishy. If a vibration is sent through the surface, it will travel at different speeds depending on whether the permafrost is solid or thawed. So we’re developing a system using low-cost fiber-optics as sensors to detect vibrations that are traveling through the earth, say, vibrations due to cars driving along a road or airplanes on a runway. We interpret all of those random vibrations and extract coherent signals that help us create a map of the surface underneath the ground. From an ICME perspective, we’ve had to develop fast algorithms to more efficiently process the dense vibration data we are getting from our sensors, because we want to be able to continuously process new data and detect new changes in the permafrost quickly without requiring huge amounts of computing resources. Eventually this process could help manage infrastructure in arctic areas: A technician could see from his or her desktop computer whether the ground is stable in many different locations along a road or runway. If the road is unstable, they will know and be able reinforce the ground before it gives way and the road crumbles.”

—Eileen Martin, ICME PhD Candidate

Alison Marsden, ICME Faculty

Improving Outcomes in Pediatric Cardiology

Alison Marsden's lab is focused on developing tools for simulating the cardiovascular system, with a particular focus on blood flow in the vasculature and the heart. Alison and her students build patient specific models from medical image data to customize models for individual patients, which represents their particular vascular anatomy. They can then use those models to run simulations of blood flow, which allows us to do virtual surgery, virtual treatment planning or risk assessment. 

Gabriel Maher, ICME PhD Candidate

Teaching a Computer to Analyze MRI Data

“I am working with the Cardiovascular Biomechanics Computation Lab. This lab is working to build and run simulations of the cardiovascular system, so we can better understand how it works, how it responds to surgeries, biomedical implants, etc. In order to perform these simulations, we currently take MRI or CT scan data and manually construct 3D models of the cardiovascular system. The process takes a very long time. I am working to figure out how we can automate the construction of accurate cardiovascular models to speed up this process. This comes down to an artificial intelligence question of how do we teach a computer to recognize where blood vessels are in MRI or CT scan data. Essentially, we are trying to teach a computer how to analyze MRI data, and then we’re working to optimize this process using machine learning, artificial intelligence, deep learning and comprehensive algorithms. If we’re able to successfully automate 3D cardiovascular model construction, this would make it easier to study and come to better understandings of the causes of particular cardiovascular diseases, or how particular surgeries affect the cardiovascular system. With better simulations, we could also potentially optimize surgical processes. Simulations allow a surgical team to run through a couple of potential surgery options, and then choose the best possible surgery for a patient.”

--Gabriel Maher, ICME PhD Candidate

Using Machine Learning to Better Understand Text

“We’re both interested in machine learning. We care a lot about software and design, proper software architecture and making sure we can use our research in the real world. Essentially, what we’re trying to do is categorize and understand text. When we get a piece of text, we want to be able to identify what it is and what it’s talking about. Is this text positive? Negative? Biased? The way we’re approaching the problem is to build algorithms and systems that can be applied to any of those problems. We are trying to build algorithms that are generic enough that they can work in multiple domains – sentiment analysis, humor detection, political bias detection, text categorization for Wikipedia, etc. We want to enable computers to match how good humans are at understanding text. To do this, we use deep learning, a very expressive and powerful part of machine learning, to build our algorithms. Deep learning allows us to learn and imbue the notion of abstraction – something humans are good at. In terms of real-world applications, we’re looking at a lot of things like measuring customer satisfaction based on reviews or tweets, categorizing news or web pages into different topics of interest, finding the most useful reviews of a product, identifying bias in text. There are really limitless possibilities here.”

--Alfredo Lainez and Luke de Oliverira, MS students

Creating Intellegent Automation Systems

Every few minutes, a new customer orders a meal from a 4- or 5- star restaurant via, a new SF startup that is rapidly expanding to multiple cities. When an order arrives, the Caviar operations team finds a suitable courier to send to the restaurant and deliver food to the customer. I took on a consulting position at Caviar to automate this process, thereby enabling the operations team to more easily distribute hundreds of orders per meal period among their spatially distributed couriers. Working closely with the talented Caviar operations and engineering teams, I was able to develop and deploy a novel intelligent automation system that fits their existing workflow. 

The problem is complex from a mathematical and algorithmic perspective, and I drew heavily upon the theory and tools gleaned through my rigorous graduate coursework taken through the Stanford ICME. For example, CME305, a challenging graduate course on Discrete Mathematics and Algorithms taught by Professor Reza Zadeh, provided a broad survey of relevant graph theory and methodology that was instrumental in my ability to rigorously formulate and solve the problem. Professor Percy Liang's excellent course on Artificial Intelligence provided me with a powerful toolkit for developing and evaluating intelligent systems. Together, these and other classes I have taken at the ICME gave me the ability to tackle this extremely complex and rewarding computational problem.

--Dave Deriso, MS