Comanici Gheorghe

  School of Computer Science, McGill University

   3480 University, Montréal, Québec

   H3A 2A7

   e-mail: gcoman [at] cs [dot] mcgill [dot] ca

About me:

My name is Gheorghe Comanici, and many know me as Ghitza. I am a PhD candidate in the School of Computer Science at McGill University, Montréal, Québec. I am a proud citizen of Canada since 2006, but also fortunate to be raised in the midst of nature's beauty: Bistriţa, Romania. I've lived there until 2002, when I was plunged in a culture amalgam called Montréal. 

I have always loved doing and teaching math more than math itself. Having a degree in Mathematics and Computer Science lead me to graduate studies in Artificial Intelligence, in particular I have a great interest in Reinforcement Learning algorithms and I study the mathematical properties that make such approaches successful and necessary.

I have been a proud member of the Reasoning and Learning Lab for the last few years. If you are interested to discuss with me and my other friendly colleagues about the great research topics we are studying, about the weather or world peace, you can knock on our door in McConnell 111. If our discussion on weather is positive (I doubt it'll be the case for world peace) we can always enjoy the great Downtown McGill Campus, downstairs from my office.

My teaching duties at McGill University include:
  • COMP322Winter 2015: undergraduate introductory course on C++ 
  • COMP322, Winter 2014
  • COMP 251Fall 2013: undergraduate course on data structures and algorithms.
  • COMP 321, Fall 2011: undergraduate course in preparation for the ACM ICPC programming competition.  

My research interests revolve around Probabilistic Systems for Artificial Intelligence Systems; lately I have been mostly working on computational aspects of bisimulation metrics for state space abstraction for Markov Decision Process. I also find myself working on Hierarchical Reinforcement Learning, Temporal Abstraction, Spectral Learning and other Machine Learning topics. In my spare time I will be solving interesting problems on Data Structures, Statistics, Probability or Measure Theory. My ultimate goal is to contribute to the advancement of lifelong learning machines.

Find below a poster I have prepared to showcase my work and that of fellow members of the RL Lab. 


  1. • Comanici, Gheorghe; Precup, Doina and Panangaden, Prakash. Basis refinement strategies for linear value function approximation. NIPS 2015 [to appear]

  2. • Ruan, Sherry Shanshan; Comanici, Gheorghe; Panangaden, Prakash and Precup, Doina. Representation Discovery for MDPs using Bisimulation Metrics. 29th AAAI Conference, 2015 [paper]

  1. • Comanici, Gheorghe; Panangaden, Prakash and Precup, Doina. On-the-Fly Algorithms for Bisimulation Metrics. 9th International Conference on Quantitative Evaluation of SysTems (QEST), 2012 [paper] 

  2. • Paduraru, Cosmin; Precup, Doina; Pineau, Joelle and Comanici, Gheorghe. A Study of Off-Policy Learning in Computational Sustainability. Proceedings of the 9th European Workshop on Reinforcement Learning, 2012 [paper] 

  3. • Comanici, Gheorghe and Precup, Doina. Basis Function Construction using Spectral Clustering and Bisimulation Metrics. Lecture Notes in Computer Science, Volume 7113, 2012 [paper] 

  4. • Comanici, Gheorghe and Precup, Doina. Basis Function Construction using Spectral Clustering and Bisimulation Metrics. The 25th AAAI Conference, 2011 [paper]

  5. • Comanici, Gheorghe and Precup, Doina. Basis Function Construction using Spectral Clustering and Bisimulation Metrics (Extended Abstract). The 10th International Conference on Autonomous Agents and Multiagent Systems, 2011 [paper][poster] (presentation on the topic: [slides])

  6. • Comanici, Gheorghe. Optimal Time Scales for Reinforcement Learning Behaviour Strategies. Master Thesis, McGill University, 2010 [paper] (presentations on the topic: [slides1] [slides2] [poster])

  7. • Comanici, Gheorghe and Precup, Doina. Optimal Policy Switching Algorithms for Reinforcement Learning. The 9th International Conference on Autonomous Agents and Multiagent Systems, 2010 [paper][poster][slides]

  1. Photography is one of my passions. A 2D grid of pixels is just enough information for the human mind to both capture a moment of “open shutter” and ignite imagination on a lifetime span. On the side you will enjoy some pictures that I am mostly proud of.