Machine Learning and Statistics

Advances in machine learning (ML) have had a profound impact on a vast variety of applications across diverse fields. At Microsoft Research (MSR) New England, we are dedicated to advancing the state of the art of ML and actively pursue research across a wide variety of ML disciplines. These include using ML techniques to drive discovery in new domains, pioneering automation methods that allow non-experts to leverage the power of ML, and exploring the ability of ML to not only reveal correlations within data but also identify the causal mechanisms that drive those correlations.

While our lab pursues a broad and diverse research agenda, many of our projects fall into the following categories:

• Novel applications of ML to challenging and impactful problems: ML has shown itself to be a powerful tool for addressing problems that are challenging using classical techniques. From sub-seasonal climate forecasting to analyzing the efficacy of cancer immunotherapy to program synthesis, the members of our lab are actively exploring how statistical and ML techniques can yield new and impactful results.
• Automated ML: While ML continues to demonstrate its utility across many domains, successfully applying ML techniques requires significant expertise and development time. The AutoML team works on developing techniques that can automate much of the development of ML pipelines, allowing non-experts to leverage the power of ML techniques, and freeing experts from much of the tedious and time consuming tasks often required to develop and deploy a ML pipeline.
• Causal Inference: Traditional ML primarily is concerned with recognizing correlations within data, but not attempting to understand the causal mechanisms that drive those correlations. We are exploring new techniques that can identify the causal relationships in the data, and exploring how these techniques can be applied to significant problems in economics.
• Generative Modeling: Generative models have emerged as powerful tools for learning complex data distributions, with applications ranging from text and image generation to scientific discovery. We are developing new methods for generative modeling, sampling, and fine-tuning, drawing on connections to stochastic optimal control, statistical inference, and applied mathematics. Our goal is to improve the efficiency, reliability, and controllability of generative models.
• AI for Chemistry and Materials Science: Understanding and designing molecules and materials often requires computationally expensive simulations and experiments. We are exploring how ML can accelerate scientific discovery through more efficient molecular simulations, the modeling of rare events and transitions, and the prediction and generation of novel materials. Our research combines advances in generative modeling and scientific ML with insights from physics and chemistry to address challenging scientific problems.

We are excited about the potential of ML as a powerful tool to drive discovery, and are passionate about contributing new, novel, and meaningful results across wide ML domains and applications.