Qualifications: Ph.D., Computer Science and Automation; M.Sc., Engineering, Indian Institute of Science (IISc), Bangalore
Title: Senior Researcher
Affiliation: IBM India Research Lab, Bangalore
Contact Details: firstname.lastname@example.org, email@example.com
Short CV: Dinesh Garg is a Senior Researcher in the Analytics and Optimization group at IBM India Research Lab, Bangalore. His research interests lie at the intersection of data science and economics which include large-scale optimization, machine learning, data analytics, game theory, and mechanism design. To a large extent, his research work has been applied as well as interdisciplinary in nature—trying to blend data science and machine learning techniques into existing body of work on auction theory and incentive mechanism design. The motivation for such a research direction stems from various monetization problems that routinely arise in diverse application domains, including online ad auctions, social networks, crowdsourcing, crowdsensing, and human computation. Prior to joining IBM Research, he worked at Yahoo! Labs, Bangalore as a Scientist in the Computational Advertising group. Dinesh's research has led to awards from the Indian National Academy of Engineering, IEEE (Best Paper), IISc, and IBM Research.
Title of Talk 1: Quadratic Optimization in High Dimensions: A Tale of Two Methods
Synopsis: High dimensional unconstrained quadratic programs (UQPs) involving massive datasets are now common in application areas such as web, social networks, etc. Unless computational resources that match up to these datasets are available, solving such problems using classical UQP methods is very difficult. In this talk, we discuss alternatives. We first define high dimensional compliant (HDC) methods for UQPs—methods that can solve high dimensional UQPs by adapting to available computational resources. We then show that two old classical methods, namely Kaczmarz and Coordinate Descent, can easily be made HDC. We further discuss the Randomized and Block versions of each of two methods. Finally, we show that both the families of the methods naturally generalizes to what is known as Projection Based Methods. We conclude the talk with various pointers towards improving upon these methods.
Title of Talk 2: Mobile Payments—A Growing Threat and How Machine Learning Can Help
Synopsis: Mobile Payments is an emerging service which allows mobile phone users to transfer money in an instantaneous manner to other users/merchants merely by a press of button on their mobile phones. Typically, this service does not involve any bankers in the loop and hence is a perfect solution to boost trades and commerce activities in countries where banking infrastructure is very poor. The business of mobile payment services is growing at an unprecedented rate wordwide. However, the bad news is that this service is susceptible to various kinds of fraudulent attacks and is also an easy gateway for money laundering/terrorist financing activities. In this talk, we uncover some of these challenges faced in practice while designing such a service and possible ways to tackle them by leveraging machine learning techniques.
Title of Talk 3: Auction Design for Online Advertising
Synopsis: Online advertising is more than a 100 billion dollar market today. Display ads (aka banner ads) and sponsored search ads constitute two key forms of online advertising. The advertising space on the webpage is typically traded by means of innovative auction mechanisms which are conducted at the speed of milliseconds. These ad auctions are power horses behind this multi-billion dollar online advertising industry today. In this talk, we explain different building blocks of any online advertising problem and present a survey of various auction methods used for them.