Courtesy of Boeing's Dragos Margineantu below is a transcript of his speaking session on 'Robust AI Systems' to Build a Thriving Enterprise that took place at Artificial Intelligence World Live - A Virtual Conference.
Session Information:
Robust AI Systems
Practical decision systems require much more than end-to-end learned models. This talk will focus on research and engineering questions on machine learning robustness that executives should be aware of.
- Usable/practical decision systems (especially for high-stakes decisions) have to be both
- Accurate
- Robust
- The major practical challenge in designing and implementing robust decision systems: it is easy to describe the desired correct behavior, but specifying correctly all consequences for failing to meet the terms of the contract, is extremely hard to specify.
- Practical decision systems are required to decide and act in the face of unknowns
- To do that, ALL learned models are required to output not only a prediction/estimate but also a self-competence estimate
- Other methods for achieving robustness: multifaceted understanding, and causal models.
Session Video:
To view all Artificial Intelligence World Live Recording Sessions Please click below:
-2.png?width=600&name=Event%20Email%20Graphic%20Virtual%20Conferences%20(8)-2.png)
To View All Events Recorded Sessions Please Click Below:
.png?width=600&name=Event%20Email%20Graphic%20Virtual%20Conferences%20(46).png)
About the Speaker:
Dragos Margineantu,
Artificial Intelligence Chief Technologist,
Boeing.
Dragos Margineantu is the AI Chief Technologist and a Senior Technical Fellow with BoeingResearch & Technology. His research interests include Computational Decision Systems, Probabilistic Models for uncertainty, Machine Learning, Artificial Intelligence, Anomaly Detection, Robust MachineLearning, Human-in-the-loop Systems, Inverse Reinforcement Learning, Tractability in Intelligent Systems, Cost-sensitive Learning, Active, and Ensemble Learning. Dragos was one of the research pioneers in ensemble learning and cost-sensitive learning since the 1990s.
At Boeing, he designed and developed computational solutions for airplane maintenance, autonomous systems, surveillance, manufacturing optimization, and design. Dragos is the technical lead of the team developing the decision system, perception localization, and mapping for the Autonomous Caravan, developed by BCA PD. He served as PI and technical lead of DARPA programs ranging from "Learning Applied to Ground Robots” (that pioneered autonomous driving) to “Bootstrapped Learning”, and from“Personal Assistant that Learns (PAL)” to “Assured Autonomy”–a program that sets the foundations of robust machine learning to engineer for autonomous systems. Dragos Margineantu serves as the Editor of the Springer book series on “Applied MachineLearning” and as the Action Editor for Special Issues for the Machine Learning Journal (MLJ). He serves on the editorial board of both major machine learning journals (MLJ and JMLR), and served as senior program committee member of all major machine learning and AI research conferences. In 2015 he was the program chair of the premier applied data science conference, the KDD-Applied Science Track. In his free time, Dragos coaches middle schoolers for Mathematics Competitions or takes his camera out to photograph unique corners of nature.
Proqis Digital Virtual Conference Series
View our schedule of industry leading free to attend virtual conferences. Each a premier gathering of industry thought leaders and experts sharing key solutions to current challenges.
View Schedule of Events
.png)
-1.png)
-2.png)
-2.png)
%20(1)%20(1).png?width=1410&name=Add%20a%20heading%20(8)%20(1)%20(1).png "Add a heading (8) (1) (1)")
.png "Access all 75 Award Finalist Entires")
-1.png?width=300&name=ATTENDEE%20-%20Proqis%20Digital%20Event%20Graphics%20(2)-1.png "ATTENDEE - Proqis Digital Event Graphics (2)-1")
-1.png?width=300&name=ATTENDEE%20-%20Proqis%20Digital%20Event%20Graphics%20(1)-1.png "ATTENDEE - Proqis Digital Event Graphics (1)-1")
