Trustworthy learning, ethics, and generalization were the focus of the Modern Artificial Intelligence spring series

Joelle Pineau, Mutale Nkonde, Danielle Belgrave & Sham Kakade

Speakers in the Spring 2021 Modern AI seminar series, clockwise from upper left: Joelle Pineau, Sham Kakade, Mutale Nkonde, and Danielle Belgrave

This spring, the Modern Artificial Intelligence seminar series, now in its fifth year, ranged far beyond considerations of the cyber-physical “nuts and bolts” of learning to include an examination of the ethical considerations of AI — reflecting a sea change in attitudes about the role of AI in society.

The lineup for the series, hosted by NYU Tandon's Department of Electrical and Computer Engineering and organized by  Professor Anna Choromanska, comprised an elite group of experts representing thought leaders at the world’s top tech companies. But it also featured experts looking at broader implications of machine learning systems, and it’s role in reifying social mores and biases.

On April 13, Mutale Nkonde, Founding Director of AI for the People delved into the organization’s work in elections, online chatter and content moderation, with a focus on racially targeted disinformation on social media during the 2020 Election. Her talk explored the growth of disinformation from 2016 to 2020 with recommendations on how to increase the racial literacy of computer scientists working in industry settings.

This year AI for the People produced a film supporting a ban on facial recognition in New York State, in partnership with Amnesty International.

Addressing the opportunities and challenges of deployment of artificial intelligence systems for personalized healthcare were Danielle Belgrave and Niranjani Prasad of Microsoft Research Cambridge, whose March 10 presentation looked at how machine learning systems could make diagnostics and treatment individualized both for the patient and to identify new disease subtypes. The pair presented examples from mental health, respiratory disease and critical care settings.

The series kicked off on February 10 with Joelle Pineau of Facebook and McGill University, who lectured on building reproducible, reusable, and robust deep reinforcement learning systems. Noting that real-world applications are often hampered by the difficulty in producing reproducible results for state-of-the-art deep learning methods, she reviewed challenges that arise in experimental techniques and reporting procedures in deep learning, with a particular focus on reinforcement learning and applications to healthcare.

The series wrapped up on May 4 with Sham Kakade of the University of Washington, whose presentation, “Towards a Theory of Generalization in Reinforcement Learning” looked at a central question in reinforcement learning: what properties govern our ability to generalize and avoid the curse of dimensionality.  His presentation reviewed recent advances towards characterizing when generalization is possible in reinforcement learning.

Choromanska launched the Modern AI seminar series in Fall, 2017 to address the most important new research in the world of artificial intelligence with talks by researchers who have made fundamental contributions to the emerging technology.

She opened the seminar to high school students, who are able to watch the talks and meet with speakers. The seminar, a flagship event at NYU Tandon, promotes leading research to an audience that is ethnically and racially diverse and represents a plethora of different local institutions, as well as distant locations within and outside of US.

The speakers themselves come from diverse backgrounds. As the founder of the seminar, Prof. Choromanska says: “The most important goal of the seminar is to unify people around the highest quality machine learning research, regardless of their age, color, sex, social status, etc. We are all members of the research family.”