2020 Frontiers of AI:ML - Rahul-Mazumder

Conference Video|Duration: 40:12
July 14, 2020
Please login to view this video.
  • Video details
    Two key ingredients in an analytics pipeline are the predictive and prescriptive components. Usually, the former involves obtaining high-quality predictions from a machine learning algorithm, and the latter takes the resulting predictions as inputs to an optimization problem for downstream decision-making. Key challenges remain in addressing scalability concerns for each of these tasks.  I will discuss some recent developments for creating large-scale algorithms for Gradient Boosting Machines and relatives, that also allow for interpretability and  model compactness. For the prescriptive part, I will present new algorithms to solve extreme scale linear programs involving trillions of decision variables, that arise in several web-applications such as email volume optimization, matching, promotion optimization, ranking and recommender systems.  

    This presents joint research with members of my research group, and collaborators at Google and LinkedIn. 

Locked Interactive transcript
Please login to view this video.
  • Video details
    Two key ingredients in an analytics pipeline are the predictive and prescriptive components. Usually, the former involves obtaining high-quality predictions from a machine learning algorithm, and the latter takes the resulting predictions as inputs to an optimization problem for downstream decision-making. Key challenges remain in addressing scalability concerns for each of these tasks.  I will discuss some recent developments for creating large-scale algorithms for Gradient Boosting Machines and relatives, that also allow for interpretability and  model compactness. For the prescriptive part, I will present new algorithms to solve extreme scale linear programs involving trillions of decision variables, that arise in several web-applications such as email volume optimization, matching, promotion optimization, ranking and recommender systems.  

    This presents joint research with members of my research group, and collaborators at Google and LinkedIn. 

Locked Interactive transcript