Mark Hamilton is not only currently earning his PhD at CSAIL, but he has also been working for Microsoft since 2016. He is jointly affiliated with Microsoft and works on the Azure Synapse Analytics team as a research development engineer.
Prior to CSAIL and Microsoft, Mark Hamilton graduated from Yale with a B.Sc in math and physics.
“In my undergraduate years, I had a lot of different experiences all aimed at trying to chase after truth with a capital T,” Hamilton said. “First it was quantum field theory, then astronomy, then mathematical foundations, and finally to machine learning.”
Since working for Microsoft, Hamilton said he has come to love the expressiveness of software engineering. Before deciding on a PhD program, he wanted to work in the field to see how his interests evolved. Ultimately, Hamilton said he missed mathematics and felt compelled to return to academia which led him to CSAIL.
Now, Hamilton works alongside his advisor Professor Bill Freeman at CSAIL.
“Working with Bill and his research group has been one of the most enjoyable and formative experiences of my life,” Hamilton said. “In Bill’s group, everyone drives their own unique research direction, and this freedom and diversity of ideas helps everyone think outside the box. Learning about Bill’s and other group member’s work has been a constant inspiration.”
Hamilton said he values working on a diverse collection of projects and hopes to make a cross-disciplinary impact. His group’s latest work investigates new approaches to “explain” or “interpret” the behavior of black-box search engines and recommendation systems.
Along the way, Hamilton said they discovered this problem has a unique solution from an unlikely field: cooperative game theory. He said they hope the ML community can use this work to improve fairness and accountability in search engines and that the economics community can leverage their fast approximations to more efficiently examine the dynamics of multiplayer games.
Another example of his group’s cross disciplinary work is their recent project MosAIc [Paper] [Demo]. Hamilton said this project introduces a method to find “hidden visual connections” across art from vastly different time periods and cultural origin. Through experimentation and feedback from curators, historians, and artists, Hamilton said they have seen that this method can help uncover the movement of style across cultures and artistic media due to factors like global trade.
“Though we originally built the method to help people find hidden patterns and explore the world’s art, we found that it could also detect ‘Blind-spot’ in modern Generative Adversarial Networks (GANs) where they fail to create realistic samples,” Hamilton said. “For example, we found that recent GANs trained on human faces struggled to create faces with cowboy hats. We hope that our technique can help researchers find and diagnose problems with their GANs and drive diverse generative modelling with deep networks.”
Meanwhile at Microsoft, Hamilton leads an open-source library called Microsoft Machine Learning for Apache Spark. This project aims to create a machine learning ecosystem that scales to thousands of machines with a simple and composable API. At Microsoft, Hamilton said they use this system to integrate computer vision, language understanding, anomaly detection and other ML methods into a variety of databases. Microsoft ML for Apache Spark allows products such as Windows, Azure Security, and Xbox to build intelligent applications for their terabyte-scale datasets. At a broader scale, his team works on building the Azure Synapse Analytics Platform.
“What I enjoy most has been working with external nonprofit organizations to help solve problems in many different fields using our library,” Hamilton said. “For example, one project helped Snow Leopard Trust monitor snow leopard population dynamics by applying computer vision methods to millions of ‘camera-trap’ images from the mountains Kyrgyzstan and India.”
Hamilton said another example includes their project Gen Studio, a collaboration with the Metropolitan Museum of Art that introduced new algorithms to help museum visitors explore the museum’s million image collection in new ways. In particular, Gen Studio gives visitors interactive control of a Generative Adversarial Network (GAN) in order to create new works of art and find similar works in the museum. Many of the projects Hamilton and his team work on have helped non-profit institutions use machine learning to make an impact in their respective fields, according to Hamilton.
He said his research is driven by a desire to learn the fundamental mathematical structures that govern our world across all scales and domains.
“I am fascinated by the idea that we can create methods to ‘parse’ the myriad of data from our world into structured and interpretable components,” Hamilton said. “I feel that this distillation of signals into structure is at the heart of the scientific method and captures an important characteristic of our own intelligence. To me, working on these problems feels like the most impactful way I can help accelerate the pace of human knowledge discovery.”
Hamilton also said that it has also been a great excuse to learn more about domains outside of computer vision such as wildlife conservation, art, public health, and economics.
“There are few things better than learning about an entirely new field from your friends and collaborators,” Hamilton said. “Collaborating with others has deepened my reverence for other fields and kept me thinking long and hard about the intricate mathematical beauty present in all natural systems.”
Following his PhD at CSAIL, Hamilton is still deciding what is next for him. He said industry provides phenomenal resources and million-user impact, but academia provides an opportunity to build and lead a research group, and help the next generation of scientists discover the same love for the topic.
“Ideally, I would like to follow in Bill’s footsteps and pursue a joint appointment,” Hamilton said. “What I have found so far through my current joint appointment at Microsoft and MIT is that academic and corporate research share many of the same goals and can combine to magnify impact. This is especially true when work is shared openly with the community and global impact is the common goal.”