Events

Filter options Close

Event Type

Impact Area

Research Area

Seminar Series

Reset Filters
Search options

April 03, 2020

April 06, 2020

Private Event

EECS Special Seminar: Dylan Hadfield-Menell "The Value Alignment Problem in Artificial Intelligence"

Dylan Hadfield-Menell
UC Berkeley
4:00P
- 5:00P

Location

online
Add to Calendar 2020-04-06 16:00:00 2020-04-06 17:00:00 America/New_York EECS Special Seminar: Dylan Hadfield-Menell "The Value Alignment Problem in Artificial Intelligence" Much of our success in artificial intelligence stems from the adoption of a simple paradigm: specify an objective or goal, and then use optimization algorithms to identify a behavior (or predictor) that optimally achieves this goal. This has been true since the early days of AI (e.g., search algorithms such as A* that aim to find the optimal path to a goal state), and this paradigm is common to AI, statistics, control theory, operations research, and economics. Loosely speaking, the field has evaluated the intelligence of an AI system by how efficiently and effectively it optimizes for its objective. This talk will provide an overview of my thesis work, which proposes and explores the consequences of a simple, but consequential, shift in perspective: we should measure the intelligence of an AI system by its ability to optimize for our objectives.In an ideal world, these measurements would be the same -- all we have to do is write down the correct objective! This is easier said than done: misalignment between the behavior a system designer actually wants and the behavior incentivized by the reward or loss functions they specify is routine, it is commonly observed in a wide variety of practical applications, and fundamental, as a consequence of limited human cognitive capacity. This talk will build up a formal model of this value alignment problem as a cooperative human-robot interaction: an assistance game of partial information between a human principal and an autonomous agent. It will begin with a discussion of a simple instantiation of this game where the human designer takes one action, write down a proxy objective, and the robot attempts to optimize for the true objective by treating the observed proxy as evidence about the intended goal. Next, I will generalize this model to introduce Cooperative Inverse Reinforcement Learning, a general and formal model of this assistance game, and discuss the design of efficient algorithms to solve it. The talk will conclude with a discussion of directions for further research including applications to content recommendation and home robotics, the development of reliable and robust design environments for AI objectives, and the theoretical study of AI regulation by society as a value alignment problem with multiple human principals.Bio: Dylan is a final year Ph.D. student at UC Berkeley, advised by Anca Dragan, Pieter Abbeel, and Stuart Russell. His research focuses on the value alignment problem in artificial intelligence. His goal is to design algorithms that learn about and pursue the intended goal of their users, designers, and society in general. His recent work has focused on algorithms for human-robot interaction with unknown preferences and reliability engineering for learning systems. online Belfer sarah_donahue@hks.harvard.edu

April 07, 2020

Private Event

Telehealth and Coronavirus: impact on each other and opportunity to surmount Machiavellian forces

Amar Gupta

Part Of

2:00P
- 3:00P

Location

virtual via Zoom
Add to Calendar 2020-04-07 14:00:00 2020-04-07 15:00:00 America/New_York Telehealth and Coronavirus: impact on each other and opportunity to surmount Machiavellian forces Join Zoom Meeting: 
https://mit.zoom.us/j/277582146

 One tap mobile:
+16465588656,,277582146# (New York) or 
+16699006833,,277582146# (San Jose)

Abstract:Healthcare is increasingly adopting a three pronged approach that involves: (i) personnel in proximity to the patient; (ii) domain experts who may be located in another city, state, or country; and (iii) computer-based analytics and other IT-based processes. Prudent application of such approaches can offer better quality of healthcare, faster access to proper attention and advice, and lower costs too. In the context of the evolving telehealth ecosystem, the discussion will include coverage of the aspects of knowledge acquisition, knowledge discovery, knowledge management, and knowledge dissemination, with greater focus on interoperability and its critical role in the use of telehealth in pandemic and other situations. In the US, 911 led to a major reform of the banking industry and the immediate adoption of automated processes for reading and processing of bank checks. Can coronavirus lead to a similar change in the healthcare industry and catalyze the widespread adoption of telehealth practices? What is the role of computer science departments and other departments at universities and what are the critical success factors to attain a better world? Examples of successful and unsuccessful strategies from US (Massachusetts, New York, Arizona, and California) and foreign countries will be used.Bio:Dr. Amar Gupta has spent most of his career fostering innovative technologies and processes, and then helping with their deployment on a global basis. He worked earlier at MIT as Principal Research Scientist and Senior Research Scientist. Then he moved from Boston to serve as Endowed and Tenured Professor at two universities. At the first university, he held several concurrent appointments including Professor of Computer Science, Professor of Public Health, and Professor of Law. At the second one, he served concurrently as Dean of Computer Science and Information Systems. In 2015, he rejoined MIT initially as Visiting Professor. Dr. Gupta was a founding Associate Editor of IEEE Transactions on Internet Technology and an Associate Editor of IEEE Transactions on Industrial Electronics. He has published over 70 papers with the largest subset in CS journals. Dr. Gupta has served as Advisor to World Health Organization and other leading international and national organizations, as well as major corporations. At MIT, he developed and taught the course Telehealth and Telemedicine for Global Health; it was co-listed as 6.884 in Spring of 2018 and its details are available at the MIT Stellar site. More details about Dr. Gupta and his activities are available at: news.mit.edu/2018/removing-health-care-barriers-and-boundaries-amar-gupta-telemedicine-0522 Belfer sarah_donahue@hks.harvard.edu

Private Event

EECS Special Seminar: Andre Wibisono "A Variational Perspective on Optimization, Sampling, and Game Dynamics for Machine Learning"

Andre Wibisono
Georgia Institute of Technology
3:00P
- 4:00P

Location

online
Add to Calendar 2020-04-07 15:00:00 2020-04-07 16:00:00 America/New_York EECS Special Seminar: Andre Wibisono "A Variational Perspective on Optimization, Sampling, and Game Dynamics for Machine Learning" Machine learning has been successful and is prevalent in everyday life, shaping many aspects of modern society. Nevertheless, many fundamental questions remain, and it is important to develop a proper theoretical understanding of machine learning to guide its future development. In this talk I will discuss the fundamental properties of optimization, sampling, and game dynamics for machine learning. In optimization, I will present a variational perspective on accelerated methods via the principle of least action in continuous time, and derive new families of accelerated methods which achieve faster convergence under refined smoothness conditions. In sampling, I will present a study of sampling as optimization in the space of measures, and show fast convergence of Langevin algorithms under isoperimetry conditions which extend classical log-concavity results. In game dynamics, I will present an analysis of minimax games as skew-optimization in the space of joint configurations, and show fast convergence of the classical fictitious play algorithm and its optimistic variant under smoothness.Bio: Andre Wibisono is a postdoctoral researcher in the School of Computer Science at the Georgia Institute of Technology. He received his SB in Mathematics and in Computer Science in 2009 and his MEng in Computer Science in 2010 from MIT. He received his MA in Statistics in 2013 and his PhD in Computer Science in 2016 from the University of California, Berkeley, advised by Michael I. Jordan. Andre was a postdoctoral researcher at the University of Wisconsin, Madison from 2016 to 2018, before he came to Georgia Tech. His research interests are in the theoretical and algorithmic aspects of machine learning, including for problems in optimization, sampling, and game dynamics. online Belfer sarah_donahue@hks.harvard.edu

April 14, 2020

Lessons from Large-Scale Cloud Software at Databricks

Matei Zaharia
Stanford
4:00P
- 5:00P

Location

32-D463 (Star)
32-D463 (Star)
Add to Calendar 2020-04-14 16:00:00 2020-04-14 17:00:00 America/New_York Lessons from Large-Scale Cloud Software at Databricks The cloud has become the most common way to deliver commercial software, but it requires building products in a very different way from traditional software, which has not been heavily studied in research. I will explain some of these challenges based on my experience at Databricks, a startup that provides a data analytics platform as a service on AWS and Azure. Databricks manages millions of VMs per day to run data engineering and machine learning workloads using Apache Spark, TensorFlow, Python and other software for thousands of customers. Two main challenges arise in this context: (1) building a scalable and reliable control plane that can manage millions of VMs and (2) adapting the data processing software itself (e.g. Apache Spark) for an elastic cloud environment — for instance, autoscaling instead of assuming static clusters. These challenges are especially significant for data analytics workloads whose users constantly push boundaries in terms of scale (number of VMs used, data size, metadata size, number of concurrent users, etc). I’ll describe some of the common challenges that our services faced and some of the main ways that Databricks has extended open source analytics software for the cloud environment (e.g., implementing autoscaling for Apache Spark and designing an ACID storage layer on top of S3 in the Delta Lake project).Bio:Matei Zaharia is an Assistant Professor of Computer Science at Stanford University and Chief Technologist at Databricks. He started the Apache Spark project during his PhD at UC Berkeley in 2009, and has worked broadly on other cluster computing and analytics software, including Apache Mesos, Apache Hadoop and MLflow. Today, Matei is a PI in the Stanford DAWN Lab doing research on infrastructure for machine learning, and continues to work on data analytics systems at Databricks. Matei’s research was recognized through the 2014 ACM Doctoral Dissertation Award, an NSF CAREER Award, and the US Presidential Early Career Award for Scientists and Engineers (PECASE). 32-D463 (Star) Belfer sarah_donahue@hks.harvard.edu

April 21, 2020

November 18, 2020

Save the Date: AI and the Work of the Future Congress 2020

MIT CSAIL, MIT Work of the Future Task Force, MIT IDE
8:00A
- 5:00P

Location

Building W16: Kresge Auditorium
Add to Calendar 2020-11-18 8:00:00 2020-11-18 17:00:00 America/New_York Save the Date: AI and the Work of the Future Congress 2020 The MIT Computer Science and Artificial Intelligence Lab (CSAIL), MIT Work of the Future (WOTF), MIT Initiative on the Digital Economy (IDE) present the MIT AI & the Work of the Future Congress.The AI age is here, transforming work and the skills required for humans to thrive. Join us as we gather leading academics, business leaders, entrepreneurs, education and labor leaders, and policy makersto explore the impact and future trends of technological disruption, rethink the nature of work, and chart a course for catalyzing innovation for inclusion and economic opportunity for all.Join MIT on November 18 to explore, rethink, and chart a course for the work of the future.The remarkable progression of technological innovation is ushering in a new era of unprecedented health, convenience, and prosperity. But the machines imbued with human and superhuman capabilities also bring a flood of collective uncertainty and anxiety about the impact on work and economies. How can we create a future of work that complements and augments human potential while contributing to shared economic prosperity?To help us understand and meet this grand challenge, three of MIT’s most innovative, future-focused initiatives have joined forces to present the MIT AI & the Work of the Future Congress. This year’s Congress will convene the leaders and visionaries on the front line that are: Rigorously measuring technological impact on enterprises, governments, and societies Generating solutions for organizations and policymakers that are working today Crafting strategies and guidelines for the futureMIT believes the challenge of building an equitable and prosperous future of work is an urgent imperative. The question we should be asking ourselves at this historic moment isn’t “what is technology going to do to our economy and society,” but rather “what will we do with technology?” Audience participation and break out sessions will be available.Don’t miss the opportunity to meet the leaders reinventing the work of the future and to engage in this important conversation.Registration information will be available soon. Members of the MIT community can reach out to alliances@csail.mit.edu for ticket information. Building W16: Kresge Auditorium Belfer sarah_donahue@hks.harvard.edu