Omid Madani
Born
Dehkuyeh village
CitizenshipUnited states
Alma materUniversity of Houston
University of Washington
Known forPrediction Games
WebsiteOmadani.net

Omid Madani is an Iranian-American distinguished computer scientist specializing in artificial intelligence (AI) and machine learning. He is known for his contributions to developing and applying learning algorithms to diverse problems spanning information retrieval and network security, as well as the analysis of the computational complexity of planning and control problems in AI.[1]

Early life and education

edit

Madani’s family originate from the Dehkuyeh village in Larestan County and the city of Shiraz, in south central Iran. He grew up in Bandar Abbas, Iran, and Dubai, UAE (one year after the onset of Iran-Iraq war).[2]

He moved to the United States in 1989 to pursue higher education. When attending Saddleback Community College in Mission Viejo, CA, he presented Roger Penrose’s The Emperor's New Mind for a project in a physics course, which led to his earliest exposure to Turing Machines as well as artificial intelligence concepts. He earned his Bachelor of Science degree in Computer Science with a minor in Mathematics from the University of Houston, in 1993. He then pursued graduate studies at the University of Washington, where he received his Masters in Computer Science and Engineering in June 1996. His master's thesis focused on fast algorithms for restriction site mapping (computational biology), under the advisement of Larry Ruzzo and Richard Karp. Madani continued his studies at the University of Washington, where he completed his Ph.D. in Computer Science and Engineering in August 2000. His doctoral research, titled "Complexity Results for Infinite-Horizon Markov Decision Processes," was conducted under the guidance of Steve Hanks and Richard Anderson. After a short stint in industry, he went back to academia and began a postdoctoral fellowship at the University of Alberta from 2001 to 2003, where he worked with Russell Greiner.[3]

Academic career

edit

Madani has published extensively in leading journals and conferences over the past decades, contributing to the understanding of machine learning and artificial intelligence, and developing applications. Notable recent publications include his 2023 paper on hierarchical concept learning in Frontiers in Computational Neuroscience and his earlier works on many-class online learning, models of active learning, and the computational complexity of decision making.[4]

Much of Madani's early work in machine learning was motivated by the desire to lower and eliminate the (manual) costs of building models. For example, supervised learning techniques, while highly successful, in practice often assume availability of explicit and time-consuming labels. Considerations of development of the mind (and research in allied fields such as cognitive psychology), and in particular how humans learn so much in their early years about their physical and social worlds, in effect resolving the semantic gap between their low-level sensory inputs and high-level categories and thoughts, and without explicit feedback, also motivated new techniques and approaches. In 2020, Madani resumed his pursuit of the ideas of Prediction Games, a framework he first proposed in the mid-2000s, led by his work on efficient large-scale online multiclass learning. In Prediction Games, the system develops its own many concepts: a hierarchical web of spatiotemporal patterns, or structured percepts, is self-grown and self-developed primarily based on the unifying goal of improving upon one’s prediction of, and the structuring of, the input stream. There is a symmetry in this approach in that concepts serve as both the predictors and the predictands (prediction targets), as well as building blocks of higher-level concepts. A major challenge is that, somewhat akin to a house of cards, new higher concepts are built on uncertain (possibly infirm and shaky) foundations. His goal is to understand and advance such learning systems and apply the ideas to domains such as computer vision and robotics.[5]

Industrial experience

edit

From 2003 to 2008, he was a Senior Research Scientist at Yahoo! Research, where he worked on large-scale learning and information retrieval. He then joined the AI Center of SRI International, where he further advanced algorithms for efficient prediction under many classes and led projects supported by DARPA and other funding agencies.[6]

From 2011 to 2014, Madani worked at Google Research's Perception Group, focusing on video classification, object detection, and topic mining, with applications to YouTube and other Google services. He later joined Cisco's Tetration Analytics Group as a founding member and Principal Engineer, where he contributed to the development of a platform for analyzing network communications for data center security and visibility.[7]

Patents and awards

edit

Madani holds over 20 patents in various areas of machine learning, including unsupervised concept induction, large-scale classification, and network security. His work has been recognized with several awards, including the Above & Beyond award from SRI International in 2009 and the Alberta Ingenuity postdoctoral fellowship in 2002.[8]

Professional memberships and community service

edit

Madani is a lifetime member of the Association for the Advancement of Artificial Intelligence (AAAI) and a member of the Association for Computing Machinery (ACM). He has served on the program committees of major conferences such as IJCAI, AAAI, ICML, ICLR, UAI, and NeuroIPS, and has reviewed journals including the Journal of the ACM, AI Journal, and Journal of Machine Learning Research.[1][3]

Selected publications

edit

Prediction games (open-ended unsupervised learning for perception)

edit
  • Madani, O. (2023). "An Information Theoretic Score for Learning Hierarchical Concepts." Frontiers in Computational Neuroscience.
  • Madani, O. (2024). “Tracking Changing Probabilities via Dynamic Learners.” Arxiv.
  • Madani, O. (2007). “Prediction Games in Infinitely Rich Worlds.” Yahoo! Research Technical Report (earlier in KDD UBDM workshop and AAAI Fall symposia).

Other topics in machine learning (online learning, active learning, …)

edit
  • Jeyakumar, V., Madani, O., ParandehGheibi, A., Yadav, N. (2016) “Data driven data center network security.” ACM International Workshop on Security And Privacy Analytics.
  • Madani, O., Connor, M., Greiner, W. (2009). “Learning When Concepts Abound.”, Journal of Machine Learning Research.
  • Raghavan, H., Madani, O., Jones, R. (2006). “Active Learning with Feedback on Features and Instances. Journal of Machine Learning Research.
  • Madani, O., Pennock, D., Flake, G. (2004). “Co-validation: Using model disagreement on unlabeled data to validate classification algorithms.”, Neural Information Processing Systems.
  • Madani, O., Lizotte, D., Greiner, R. (2004). “Active Model Selection.”, Uncertainty in AI.

Algorithm design and computational complexity of dynamic decision making

edit
  • Madani, O., Hanks, S., & Condon, A. (2003). "On the Undecidability of Probabilistic Planning and Related Stochastic Optimization Problems." Artificial Intelligence Journal.
  • Madani, O., Thorup, M., & Zwick, U. (2010). "Discounted Deterministic Markov Decision Processes and Discounted All-Pairs Shortest Paths." ACM Transactions on Algorithms.
  • Madani, O. (2002). "On Policy Iteration as a Newton's Method and Polynomial Policy Iteration Algorithms." National conference on Artificial Intelligence.
  • Etzioni, O., Hanks, S., Jiang, T., Karp, R.M., Madani, O., and Waarts, O. (1996)“Efficient Information Gathering on the Internet.” IEEE FOCS.

References

edit
  1. ^ a b "Omid Madani". Association for Computing Machinery.
  2. ^ "Omid Madani - The Mathematics Genealogy Project". Mathematics Genealogy Project.
  3. ^ a b "OMID MADANI". CiteSeerX.
  4. ^ "Omid Madani". www.omadani.net.
  5. ^ "Omid Madani Papers". aimodels.fyi.
  6. ^ "AI Seminar Intranet - 2024". University of Alberta.
  7. ^ "dblp: Omid Madani". DBLP.
  8. ^ "Omid Madani Inventions, Patents and Patent Applications - Justia Patents Search". Justia.