Large Scale Analytics
Good solutions for large scale analytics problems are built on the ability to couple amounts of data, with appropriate models, and scalable optimization techniques -- and somewhere in the mix is also good visualization. The key in developing algorithms is to identify features of the problem that can be exploited. We are forever on the quest for models and methods that can have significant practical impact.
We have developed algorithms that exploit the sparsity of CT scans to reduce the radiation impact of patients by a factor of 8 or more. The very same class of algorithms also solve very large scale distributed machine learning problems while ensuring the privacy of the data. We are working on extending these algorithms on inferring the structure of a network of neurons from neuron firing data.
Abundant data and cheap computing are revolutionizing supply chain management. Traditional supply chain control algorithms rely on closed-form models that are easy to compute, but that do not accurately reflect the complexities of modern supply chain systems. We are developing on a new class of inventory planning algorithms that rely on fast computations and abundant real-time data to provide near-optimal policies that are always up to date - and on associated visualization tools to help managers more effectively use these models to make decisions.