Events

"Finding the perfect salesperson: Advanced job matching models in Bravado"

Presentation by Yoni Acriche
Co-Founder and Chief Data Scientist
Bravado, Austin

3:30 pm - Seminar
4:30 pm - Reception, coffee and cookies

About the seminar:

The performance of the sales team directly affects companies’ revenue, and the alternative cost of a bad hire is often much greater than the hire’s annual salary. However, only 43% of sales hires hit their revenue targets in 2021, and the average tenure of a salesperson decreased to less than 12 months. Given this, it’s no surprise that Sales is considered one of the most challenging roles to hire. Bravado, a Series B startup, aims to ease the process and minimize the risks associated with hiring sales employees. By building the world’s largest professional network for salespeople, Bravado uses a novel data-driven approach to make hiring easier, increase transparency, and help candidates and companies find better fits. In this presentation, I will cover the data architecture of Bravado’s sales hiring matching solution, discuss the strategy of picking the right Machine Learning approach, and explain in detail the neural network model that powers the matching algorithm and its empirical results.

Bio:

Yoni Acriche is the Co-Founder and Chief Data Scientist of Bravado, a series B startup funded by Tiger Global, Redpoint, and Freestyle VC. In Bravado, Yoni focuses on leveraging Machine Learning to make the community members more successful, and increase the company’s operational efficiency. Prior, Yoni was the Head of Data Science at Salespredict (acquired by eBay), and the Head of Seller Experience Research at eBay. His research focuses on using large-scale neural network language models to increase the efficiency of online marketplaces. His work was published in top Machine Learning conferences such as ACL, WSDM, and ACM, as well as in the Harvard Business Review.

Papers:

Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
Product Titles-to-Attributes As a Text-to-Text Task

Seminar video:

A recording of this seminar is available to University of Minnesota faculty, staff and students. Watch the seminar.

"NextGen Robotics with Simulation and Pretraining"

Presentation by Ashish Kapoor
General Manager
Autonomous Systems and Robotics, Microsoft Research, Redmond

3:30 pm - Seminar
4:30 pm - Reception, cookies and coffee

About the seminar:

There are fundamental shifts happening due to advances in Deep Learning as well as our ability to harness large amounts of computation on the cloud. Additionally, we are also seeing how Deep Machine Learning is changing the foundation of numerical computing and simulation. We build upon these observations and propose a different way to build robots and autonomous systems. Specifically, I’ll discuss large scale data-driven and differentiable simulations and how they can help us design and train robotic systems. Secondly, I’ll also highlight approaches that utilize large scale pre-trained models and discuss implications towards how we might build robots in the future.

Papers:

Teaching a robot to see and navigate with simulation

Just say the magic word: using language to program robots

PACT: Perception-Action Causal Transformer for Autoregressive Robotics Pretraining

COMPASS: COntrastive Multimodal Pretraining for Autonomous Systems

Bio:

Ashish Kapoor is the Chief Scientist of the Autonomous Systems and Robotics Group  at Microsoft, Redmond. His recent research focuses on safety, pre-training of deep models and data-driven simulation for Autonomous Systems. Ashish has a PhD from MIT and a Bachelor of Technology from Indian Institute of Technology Delhi.

"Human in the Loop Automation: Ride-Hailing With Remote (Tele-) Drivers"

Presentation by Xiaotang Yang
PhD Candidate, Department of Industrial and Systems Engineering
University of Minnesota

3:30 pm - Seminar
4:30 pm - Reception, snacks and beverages

About the seminar:

Tele-driving refers to a novel concept where drivers can remotely operate vehicles (without being physically in the vehicle). By putting the human back "in the loop," tele-driving has emerged recently as a more viable alternative to fully automated vehicles, with ride-hailing (and other on-demand transportation-enabled services) being an important application. Because remote drivers can be operated as a shared resource (any driver can be assigned to any customer regardless of trip origin or destination), tele-driving has the potential to reduce the severity of the spatial mismatch between vehicle supply and customer demand that is often experienced in these services and the number of drivers needed. In this paper, we quantify the potential gains that could be realized by switching from in-person drivers to remote drivers. We compare a traditional ride-hailing system with one with tele-drivers in three regimes defined by vehicle capacity. We find that: (1) if customers are impatient, a system with appropriately selected driver capacity may significantly improve the service level (the fraction of demand that is served), or significantly reduce the number of drivers while maintaining a similar service level; and (2) if customers are patient, a system with remote drivers may stabilize an otherwise unstable system with in-vehicle drivers or significantly reduce the number of drivers while maintaining a similar service level (as measured by the expected delay experienced by customers).

Bio:

Xiaotang Yang is a Ph.D. candidate at the University of Minnesota -- Twin Cities in the Department of Industrial and Systems Engineering, advised by Saif Benjaafar. Her research interests are in the area of operations management with a focus on on-demand service platforms, smart mobility, and the sharing economy. She aims to understand the impact of new business models and innovative technologies in the context of these applications on multiple stakeholders, including customers,independent workers, and firms. Her research involves building quantitative models to (i) support decision making by the various parties involved and (ii) draw managerial insights and implications for public policy.

What is Industrial and Systems Engineering?

Many students don't know! Industrial and Systems Engineers blend mathematical modeling, engineering thinking, and business practices to optimize system performance. In fact, ISyE is one of the fastest-growing programs in the College of Science and Engineering. Learn more about this exciting major at our upcoming info session. If you cannot attend, you can still RSVP and a member of our department will get in touch.

RSVP to the Info Session

"Reasons for Peace: Negotiations and the Holding Principle"

Presentation by Jorge Ramos-Mercado
PhD Candidate, Department of Economics
University of Minnesota

3:30 pm - Seminar
4:30 pm - Reception, snacks and beverages

About the seminar:

In the standard reputational-bargaining model (Abreu and Gul 2000), two strategic, impatient players reach agreements in finite time. I extend their setting by adding hidden effort to preserve a declining surplus. This extension accounts for negotiations ending without agreement despite multiple rounds of bargaining. Even when effort is not very costly, players exert low effort in expectation that their opponent appropriates an out-sized share of the surplus. Hence, the surplus may be destroyed before players reach an agreement.  I, nonetheless, find that the risk of surplus destruction improves the expected bargaining outcomes, because observing the endogenously determined surplus minimizes strategic delay. This model is further consistent with the 27 percentage point decline, since 1914, in the share of wartime negotiations effectively concluding wars.

Paper:

Negotiations and the Holding Principle (PDF)

Bio:

Jorge David Ramos-Mercado is a PhD candidate in the Department of Economics at the University of Minnesota—Twin Cities. He is a microeconomist who studies the role of dynamic learning in applied game theory and market design. His previous work finds applications ranging from art auctions, wartime diplomacy, and fiscal policy determination. Prior to going to graduate school, Jorge worked at the Center for Retirement Research at Boston College where he studied late career changes over time and changes in mortality inequality by race and socio-economic status. Lastly, he was born in San Juan, Puerto Rico but has lived all over the United States.

What is Industrial and Systems Engineering?

Many students don't know! Industrial and Systems Engineers blend mathematical modeling, engineering thinking, and business practices to optimize system performance. In fact, ISyE is one of the fastest-growing programs in the College of Science and Engineering. Learn more about this exciting major at our upcoming info session. If you cannot attend, you can still contact our department and a staff member will get in touch.

Join via Zoom

"The Role of Social Networks in Economic Mobility"

Presentation by Mike Bailey
Research Science Manager
Core State Science, Meta

3:30 pm - Virtual Seminar via Zoom

Seminar will be also be viewable in Walter Library room 402.

About the seminar:

Social capital—the strength of an individual’s social network and community—has been identified as a potential determinant of outcomes ranging from education to health. We use data on 21 billion friendships in the US to measure and analyze different types of social capital including connectedness between different types of people, social cohesion, and civic engagement. We demonstrate the importance of distinguishing these forms of social capital by analyzing their associations with economic mobility across areas. The share of high-SES friends among individuals with low SES—which we term economic connectedness—is among the strongest predictors of upward income mobility identified to date. In a different paper we use social network data in India to show the importance of social networks to labor migrants and find that increasing social connectedness across space may have considerable economic gains, improving average wages by 3% (24% for the bottom wage-quartile) in a migration model.

Papers:

Social capital I: measurement and associations with economic mobility
Social capital II: determinants of economic connectedness
Social Networks and Spatial Mobility: Evidence from Facebook in India

Bio:

Mike Bailey is a senior social scientist at Meta on the Core Data Science team. His work focuses on the role of social networks on economic opportunity including migration, health, education, and social capital and has been featured in top scientific journals such as Nature and the Journal of Political Economy and covered by outlets such as The Economist and The New York Times.

"Long-run Performance of Approximation Algorithms"

Presentation by Paul Milgrom
Shirley and Leonard Ely Professor of Humanities and Sciences
Department of Economics, Stanford University, Palo Alto

3:30 pm - Seminar
4:30 pm - Reception, snacks and beverages

About the seminar:

We study investment incentives created by truthful mechanisms that allocate resources using approximation algorithms. Even for some high-performing (``FPTAS'') approximation algorithms, when a bidder can invest before participating, its investment incentives may be so distorted that the net welfare performance is arbitrarily bad. An algorithm's worst-case allocation and investment performance coincide if and only if a particular kind of externality is sufficiently small. We introduce a new FPTAS for the knapsack problem that has no such negative externalities, so it is high-performing with and without investments.

Bio:

Paul Milgrom is the Ely Professor of Humanities and Sciences in the Department of Economics at Stanford University and the recipient of numerous awards, including the 2020 Sveriges Riksbank Prize in Memory of Alfred Nobel, for “improvements to auction theory and invention of new auction methods.” Paul is the author of two books about auction design and his scholarly publications have more than 100,000 Google Scholar citations. He co-invented the two auction formats most commonly used for selling radio spectrum licenses in North America, Europe, Asia and Australia and the Auctionomics team that designed the U.S. Incentive Auction process which reallocated UHF-TV channels for use in mobile broadband.

"Using Deep Networks and Transfer Learning for Player Experience Management in Online Games"

Presentation by Nishith Pathak
Data Scientist
Wargaming, Austin

3:30 pm - Seminar
4:30 pm - Reception, snacks and beverages

About the seminar:

Online games are a rich eco-system of many users interacting and participating in goal oriented activities. Developing and maintaining such systems is a non-trivial process. User satisfaction in online games is an important and complex concept which has many factors feeding into it, such as attaining goals in the game, the social experience of interacting with other players as well as the UI and UX aspects of the game world. Traditionally the gaming industry has relied on game designers’ experience and user surveys along with AB testing for informing various decisions related to these topics. However, with recent advances in machine learning, particularly the ability of deep networks to deal with unstructured data, there are now various possibilities for designing automated systems for more agile, scalable and effective user experience management. This talk will focus on how how deep networks and transfer learning can be used for some of the challenges faced in maintaining a positive user experience. We specifically look at models for analyzing text data and focus on two problems. The first one is related to identifying toxic behavior among users. The second problem is extracting sentiment and relevant topics from user feedback.

Bio:

Dr. Nishith Pathak is a senior Data Scientist with Wargaming.net, developer of some of the most popular online games such as World of Tanks and World of Warships. He completed his PhD in Computer Science from the University of Minnesota Twin Cities, Minneapolis, MN and his Bachelors in Technology from the Indian Institute of Technology, New Delhi. His research interests lie in the area of machine learning and its application in the gaming domain. He has collaborated actively with industry leaders and has over thirteen years of experience researching, developing and implementing systems for facilitating processes in various aspects of a game’s life cycle involving development as well as maintainence. He has also published various peer-reviewed research articles in leading IEEE and ACM conferences as well as journals. For the last four years he has been working at the Prague and Austin locations for Wargaming.net and has been involved in developing machine learning based systems related to customer satisfaction, game balance and user skill estimation.

Watch Seminar:

"Time series forecasting in logistics"

Presentation by Kaisa Taipale
Director of Data Science
C.H. Robinson, Twin Cities

3:30 pm - Seminar
4:30 pm - Reception, snacks and beverages

About the seminar:

While the time series beloved by venture capitalists go up and to the right, many other time series, including those in freight, exhibit multi-season cyclicality or periodicity. The business cycle, agricultural cycles, holidays, and more all influence freight time series. This talk will delve into time series that show layers of cyclicality and highlight different methods for attempting to understand the complex system of freight logistics in the US. I will also focus heavily on the business trade-offs inherent in modeling choices, as this is not a theoretical exercise but a practical one.

Bio:

Dr. Kaisa Taipale is Director of Data Science at C.H. Robinson, the largest third-party logistics firm in the United States, and was formerly with the School of Mathematics at the University of Minnesota, working with the Minnesota Center for Financial and Actuarial Mathematics. Her group at C.H. Robinson is responsible for long-term time series models for North American truckload freight, constructing and monitoring appropriate feedback loops to anticipate and react to economic and geopolitical events that affect the supply chain. Kaisa holds a PhD in algebraic geometry from the University of Minnesota. Prior to joining C.H. Robinson and her time on the faculty at the University of Minnesota, she held postdoctoral positions at MSRI (Mathematical Sciences Research Institute), Cornell, and St. Olaf College.