Industrial Mathematics and Statistics Seminar
In light of the recent development of data driven research and industrial activities and their intimate connection to techniques from mathematics and statistics, the purpose of this weekly seminar is to introduce opportunities from industries to mathematics and statistics students. Students from other disciplines are welcome and encouraged to participate. Each week the speaker will present possible ideas, methodology and projects for students to explore. The topics will come from a wide spectrum of theory and practice.
(Department of Statistics),
Mark Daniel Ward
(Department of Statistics),
Nung Kwan (Aaron) Yip
(Department of Mathematics).
Spring 2022, Date, Time and Place:
Thursdays, 11:00am-12:00pm (EST).
Dynamic Pricing Provides Robust Equilibria for Ridesharing
Ridesharing markets, such as those operated by Uber and Lyft, match riders wishing to travel with driver partners seeking flexible earning opportunities. Matching these two groups is a difficult dance across time and space: too few drivers in a neighborhood means riders must wait or find alternate transportation; too many means that drivers idle when they want to earn. Adding to the challenge, drivers may ignore advice from platforms about where and when to drive. If prices are not set well, these challenges destroy efficiency. Building on the presenter's experience as a data scientist at Uber, we present a new algorithmic approach to dynamic pricing and matching that uses ideas from stochastic programming in a game-theoretic setting. Providing new insights into the value of pricing dynamically, we show in a large-market limit that all approximate rider-driver equilibria are approximately welfare-optimal under our dynamic pricing method.
In contrast, a variant whose prices do not react to the realized distribution of drivers has equilibria with poor welfare.
This is joint work with Massey Cashore and Eva Tardos.
Title: Thoughts on Risk Analysis Methods and Deterrence
Abstract: One of the most difficult problems in analyzing risks from strategic interactions between groups or nations is accounting for antagonist behavior and the choices of different actors. How do we account for the actions of individual decision makers in high stakes conflicts in order to make risk informed decisions about policy, capability investments, and defensive actions? How can we understand if deterrence might work or not? This talk will give a brief overview of deterrence challenges and describe how important it is to be able to solve these problems. The talk will also touch on some of the issues with applying risk analytic approaches to such problems, and illustrate one method for doing so.
Yen (Terri) Bui
Title: Merck-Purdue Data Mine Collaboration: Where do the undergrads go?
As big data and analytics are being adopted across industries - Merck is also following suit. Here, we present a method of engagement for academia and industry developing agile solutions and platforms to expedite the integration of data science into pharmaceutical sciences. In this engagement, students and Merck researchers collaborate by exploring technologies pertinent to big-data strategies that will enable data-driven research decisions through the development, design, and implementation of: (1) a biometric platform workflow and integration platform; (2) radio-frequency identification (RFID) software solution; (3) natural language processing (NLP) on scientific text;
While these technologies have been readily adopted in other industries, their usage within the clinical and pharmaceutical research and development sector is still being explored from a compliance and integration perspective. Thus, the value provided in utilizing the Merck-Purdue model of engagement by exploring new libraries and packages within the framework of general
industrial "use cases" can serve as preliminary "data" for early adoption and operationalization.
Future works involve expanded areas of research within Merck Research Lab
and Purdue Faculty involvement.
Optimization of Black-Box Functions in the Presence of
In many engineering applications, one wishes to optimize the performance of a system simulated by software whose intrinsics are not accessible to us. Objective function values are available (and are typically noisy) but derivatives are unknown. We discuss how to solve problems of this kind in practice. Examples in machine learning and engineering design illustrate the challenges to be overcome, particularly when the number of unknowns is large and the model includes constraints that must be respected.
Alfred O. Hero, III
(University of Michigan)
Immuno-mimetic Deep Neural Networks
Biomimetics has played a key role in the evolution of artificial neural networks. Thus far, in-silico metaphors have been dominated by concepts from neuroscience and cognitive psychology. In this talk we introduce a different type of biomimetic model, one that borrows concepts from the immune system, for designing robust deep neural networks. This immuno-mimetic model leads to a new computational framework for robustification of deep neural networks against adversarial attacks. Within this Immuno-Net framework we define a robust adaptive immune-inspired learning system (RAILS) that emulates, in silico, the adaptive biological mechanisms of B-cells that are used to defend a mammalian host against pathogenic attacks.
Xuming Xie (Morgan State University)
Free boundary problems modelling Hele-Shaw flows and diabetic atherosclerosis
The problem of a less viscous fluid displacing a more viscous fluid in a
Hele-Shaw cell has been the subject of numerous studies since Saffman and
Taylor's seminar paper in 1958. The Saman-Taylor instability has been
served as a standard example of pattern formation and stability in
non-equilibrium system. In this talk, we are concerned with some variations
and extensions of Saffman-Taylor finger problem in a Hele-Shaw cell, these
extensions have many applications in industry. We will focus on Hele-Shaw
problems with non-zero kinetic undercooling regularization on the free
Over the last several decades, free boundary problems have been emerging
in the growing research field of mathematical biology. Free boundary
problems, which model the tumor growth, wound healing and atherosclerosis,
have been extensively studied. In this talk, we will discuss a recent free
boundary model for diabetic atherosclerosis. We will address several issues
regarding the mathematical analysis and numerical simulation of this model.
(Partly supported by National Institute of General Medical Sciences of the
National Institutes of Health under Award Number UL1GM118973.)
(Morgan State University)
A Problem about Submodular Functions from Welsh's 1976 Text,
Submodular functions arise in a variety of applications, such as economics, especially applications related to combinatorial optimization. We solve a
problem from Oxford University professor Dominic Welsh's seminal 1976
text, Matroid Theory.
Fall 2021, Date, Time and Place:
Fridays, 2:30pm-3:30pm (EST).
(Bio/Distinguished Professor of Mathematics, Director of Graduate Programs, North Carolina State University)
Phylogenetic Algebraic Geometry
The main problem in phylogenetics is to reconstruct evolutionary relationships between collections of species, typically represented by a phylogenetic tree. In the statistical approach to phylogenetics, a probabilistic model of mutation is used to reconstruct the tree that best explains the data (the data consisting of DNA sequences from homologous genes of the extant species). In algebraic statistics, we interpret these statistical models of evolution as geometric objects in a high-dimensional probability simplex. This connection arises because the functions that parametrize these models are polynomials, and hence we can consider statistical models as algebraic varieties. The goal of the talk is to introduce this connection and explain how the algebraic perspective leads to new theoretical advances in phylogenetics, and also provides new research directions in algebraic geometry. The talk material will be kept at an introductory level, with background on phylogenetics and algebraic geometry.
Spring 2021, Date, Time and Place:
Wednesdays, 3:30pm (EST)
Jeffrey M. Larson (Argonne National Laboratory)
A collection of optimization problems arising in quantum information sciences
This talk highlights a collection of problems where numerical optimization methods can be
applied to problems arising in the quantum sciences. Examples will include 1) optimizing
parameters in variational algorithms, 2) maximizing entanglement of quantum dot/plasmon
simulations, 3) identifying optimal cuts for quantum circuits in order to efficiently
map them onto smaller quantum devices.
(Suggested reading material:
Introduction to Derivative Free Optimization,
by Conn, Scheinberg, Vicente)
Anne McLaren (Cummins Inc.)
Applications of Statistics to Solve Real World Problems
This talk will focus on several case studies in the areas of engine calibration development, reliability parenting using warranty data analysis, and reliability growth modeling. For each, we will discuss the problem statement
and objective and then review the types of data available models used.
(Useful website: Reliability Engineering Resource Website)
Mar. 3: Talk cancelled.
Ricardas Zitikis (Western University, London, Ontario, CA)
Detecting systematic anomalies affecting systems when inputs are stationary time series
We develop an anomaly detection method when systematic anomalies are affecting control systems at the input and/or output stages. The method allows anomaly-free inputs (i.e., those before contamination) to originate from a wide class of stationary random sequences, thus opening up the most diverse possibilities for its applications. To show how the method works on data, and how to interpret results and make decisions, we provide an extensive numerical experiment with anomaly-free inputs following ARMA time series under various contamination scenarios. It is joint work with Ning Sun and Chen Yang.
Erica Zuhr (Onlife Health)
Data Science in the Health and Wellness Industry
This talk will introduce data science problem statements of varying complexity arising in the wellness industry. We will use toy problems to understand solution techniques and review applications of these techniques at Onlife Health. We will also present an overview of the tools commonly used to solve such problems, and touch on applied data science across the broader healthcare industry.
Meg Walters (Allstate)
The data science of car accidents: prevention, detection, and damage assessment
This talk will provide an overview of data science problems in the auto insurance space. We will begin with high frequency sensor data used for prevention and detection of accidents and end with an application of computer vision to assess vehicle damage. We will discuss challenges as well as common techniques used to extract critical insights from an overwhelming amount of data that can be collected from almost any mobile phone.
Pavithra Harsha (IBM)
Dynamic Pricing of Omnichannel Inventories
Omnichannel retail refers to a seamless integration of an e-commerce
channel and a network of brick-and-mortar stores. An example is
cross-channel fulfillment, which allows a store to fulfill online orders
in any location. Another is price transparency, which allows customers to
compare the online price with store prices. This paper studies a new and
widespread problem resulting from omnichannel retail: price optimization
in the presence of cross-channel interactions in demand and supply,
where cross-channel fulfillment is exogenous. We propose two pricing
policies that are based on the idea of "partitions" to the store inventory
that approximate how this shared resource will be utilized. These policies
are practical because they rely on solving computationally tractable mixed
integer programs that can accept various business and pricing rules. In
extensive simulation experiments, they achieve a small optimality gap
relative to theoretical upper bounds on the optimal expected profit. The
good observed performance of our pricing policies results from managing
substitutive channel demands in accordance with partitions that rebalance
inventory in the network. A proprietary implementation of the analytics
that also includes demand estimation was commercialized and available as
part of the IBM Commerce markdown price solution suite. The system results
in an estimated 13.7% increase in clearance-period revenue based on causal
model analysis of the data from a pilot implementation for clearance
pricing at a large U.S. retailer.
Ojas D. Parekh (Sandia)
Quantum discrete optimization
In this talk I will motivate quantum approaches to discrete optimization by highlighting fundamental connections between quantum physics and discrete optimization. I will explain popular quantum discrete optimization techniques such as the Quantum Adiabatic Algorithm and the Quantum Approximate Optimization Algorithm (QAOA). I will discuss the performance and limitations of QAOA by using the NP-hard Maximum Cut problem as an example. Finally, I will discuss a physically motivated quantum generalization of Maximum Cut.
This talk assumes no familiarity with quantum computing or discrete optimization, and is aimed at a general Math/CS audience.
Useful and introductory materials for quantum computation
Abhishek Parab (IRI)
Recommender Systems in Industry
How does Netflix recommend your next movie? How does Youtube curate its "similar" videos or Amazon suggest products you end up buying? Recommender Systems play a crucial role in providing a personalized experience which leads to enhanced user satisfaction and loyalty. I will discuss the mathematics behind these algorithms much of which was largely inspired by the "Netflix Prize", and how my attempts to answer a business question led me to exploring these. Being a Math alumnus myself, I can talk about my transition from pure mathematics to data science and answer related questions.
Jonathan Allen (Lawrence Livermore National Lab)
Data-driven modeling approaches in computational drug discovery
Advances in chemical synthesis are expanding the number of molecules that can be easily made and tested for drug discovery. The exceptionally large ‘makeable’ chemical space means experimental data can only be collected for a tiny fraction of candidate molecules and this can fundamentally limit the type of molecules found using an experimentally driven drug discovery process. New approaches that employ a computational search of the virtual chemical space have the potential to find molecules that meet multiple design criteria and increase the chances of a drug candidate advancing to human clinical trials. This talk will introduce elements of the Accelerating Therapeutics for Opportunities in Medicine (ATOM) drug discovery framework including a generative statistical chemical model and an iterative drug design loop that selectively explores chemical space using data-driven small molecule property prediction models and physics-based scoring functions. Preliminary results show the promise of an iterative design loop, with opportunities for improvement in areas such as quantifying model prediction uncertainty and optimizing chemical search techniques. The aim of this work is to build an open computational framework accessible to the broader research community that can improve the efficiency of drug discovery on new disease targets.
Steve Swanson (formerly at Amazon Web Services)
From Math PhD to Software Developer
After getting his pure math PhD in 1993, Steve taught for two years, then moved (back) to software development. He stayed in that industry until 2019 when he retired from Amazon.com. In this talk, Steve will review some of the joys and pitfalls of working in the software industry. He will also give some opinions on interviewing and what hiring managers are looking for.
Steve will finish with a Q&A session.