Mahmoud Assran, ("Mido")
PhD Student, McGill University
Mila – Quebec AI Institute
mahmoud.assran@mail.mcgill.ca
[CV], [Scholar], [Overview], [Facebook AI]
Hi! I'm Mahmoud Assran, people call me Mido. I am a PhD student and a Vadasz Doctoral Fellow in engineering at McGill University and Mila – Quebec AI Institute, advised by Michael Rabbat. I am also a Visiting Researcher at Facebook AI Research in the Montreal lab. I grew up in the land of living skies, but have been pursuing my studies (and practicing my french) out east. My doctoral research is graciously supported through a Vanier Canada Graduate Scholarship.

My interests include numerical optimization, distributed computing, representation learning, and self-/semi-/weakly-supervised learning. The goal is to develop algorithms that learn efficiently (with less time, energy, and data than current approaches), and that derive robust and generalizable representations related to semantic concepts in data. My previous work has spanned large-scale empirical studies and theoretical analyses.

Feel free to reach me at mahmoud.assran@mail.mcgill.ca.



Open-Sourced Code

facebookresearch/gala

Parallelizing multi-simulator Reinforcement Learning using asynchronous gossip.

NuerIPS 2019
facebookresearch/sgp

Parallelizing large-scale Supervised Learning using Push-Sum gossip with delays.

ICML 2019
MidoAssran/maopy

Multi-Agent Optimization for Python: decentralized optimization implemented with mpi4py.

IEEE Transactions on Automatic Control



Publications

Semi-Supervised Learning
Recovering Petaflops in Contrastive Semi-Supervised Learning of Visual Representations
preprint 2020

We find that access to a few labeled samples during self-supervised pre-training can save a lot of compute. We propose one such strategy for leveraging labeled samples.

[paper]
Distributed Optimization
Advances in Asynchronous Parallel and Distributed Optimization
preprint 2020

We survey centralized and decentralized asynchronous optimization methods (and proof-tenchinques) for large-scale machine-learning.

[paper]
Optimization
On the Convergence of Nesterov's Accelerated Gradient Method in Stochastic Settings
ICML 2020

Nesterov momentum + SGD can provably diverge for a broad class of smooth strongly-convex finite-sums, even under interpolation. Watch out for inconsistent mini-batches!

[paper], [video]
Deep Reinforcement Learning
Gossip-based Actor-Learner Architectures for Deep Reinforcement Learning
NeurIPS 2019

We use asynchronous gossip to parallelize Deep RL algorithms; we improve computational efficiency while maintaing sample efficiency.

[paper], [code], [poster], [video]
Distributed Optimization
Stochastic Gradient Push for Distributed Deep Learning
ICML 2019

We parallelize Supervised Learning using stochastic gradients and Push-Sum gossip. We also prove that the method converges at the same rate as SGD for smooth non-convex objectives.

[paper], [code], [poster]
Distributed Optimization
Asynchronous Gradient Push
IEEE Transactions on Automatic Control 2018*

Asynchronous agents running gradient descent on their local objectives while asynchronously gossiping with their peers minimize a weighted average of their objectives.

[paper], [code]
Distributed Optimization
An Empirical Comparison of Multi-Agent Optimization Algorithms
IEEE GlobalSIP 2017

We propose Asynchronous Gradient-Push for decentralized learning over directed graphs, and show that this primitive asynchronous method can outeperform synchronous state-of-the-art methods in practice.

[paper], [code], [poster]



Presentations & Talks

  • "Gossip-based Actor-Learner Architectures for Deep Reinforcement Learning"
    Aggregate Intellect (ai.science)
    (Remote Talk) Montreal, Canada(2020)
  • "On the Convergence of Nesterov's Accelerated Gradient Method in Stochastic Settings"
    International Conference on Machine Learning (ICML)
    (Remote Talk) Vienna, Austria(2020)
  • "On the Convergence of Nesterov's Accelerated Gradient Method in Stochastic Settings"
    Montreal Machine Learning and Optimization Research Group
    Montreal, QC, Canada(2020)
  • "Youth Engagement through Project Based Learning"
    English Montreal School Board, Teachers' Professional Development
    Montreal, QC, Canada(2020)
  • "Artificial Intelligence: Learning over Networks"
    Indigenous Services Canada, National Indigenous Information Technology Alliance
    Winnipeg, MB, Canada(2019)
  • "Convergence of Asynchronous Subgradient Push under Arbitrary Bounded Delay"
    IEEE Asilomar Conference on Signals, Systems, and Computing
    Pacific Grove, CA, United States(2019)
  • "Gossip-based Actor-Learner Architectures for Deep Reinforcement Learning,"
    Montreal AI Symposium
    Montreal, QC, Canada(2019)
  • "Stochastic Gradient Push for Distributed Deep Learning,"
    International Conference on Machine Learning
    Long Beach, CA, United States(2019)
  • "Stochastic Gradient Push for Distributed Deep Learning,"
    Systems for Machine Learning Workshop, Neural Information Processing Systems
    Montreal, QC, Canada(2018)
  • "Convergence of Asynchronous Subgradient Push under Arbitrary Bounded Delay,"
    European Control Conference
    Limassol, Cyprus(2018)
  • "An Empirical Comparison of Multi-Agent Optimization Algorithms,"
    IEEE Global Conference on Signal and Information Processing(2017)
    Montreal, QC, Canada



Academic Reviewing

  • Automatica(Journal)
  • Neural Information Processing Systems (NeurIPS)(Conference)
  • International Conference on Machine Learning (ICML)(Conference)
  • Association for the Advancement of Artificial Intelligence (AAAI)(Conference)
  • ReScience: ICLR Reproduciblity Challenge(Journal)
  • IEEE Transactions on Signal Processing(Journal)
  • IEEE Transactions on Control of Network Systems(Journal)
  • IEEE Transactions on Signal and Information Processing over Networks(Journal)
  • IEEE Signal Processing Magazine(Journal)
  • International Federation of Automatic Control World Congress(Conference)



Honours & Awards

  • Vanier Canada Graduate Scholarship
    Natural Sciences and Engineering Research Council of Canada
    Awarded to only 12 doctoral students in the Computing Sciences over the past 5 years; the Vanier CGS plays an important role in fulfilling the Government of Canada's Science and Technology strategy to promote the devleopment and application of leading-edge knowledge, support the development of a world-class workforce, and attract and retain the world's top graduate students. Awarded to doctoral students who have demonstrated strong leadership skills and a high standard of scholarly achievement in graduate studies in the social sciences and humanities, natural sciences and engineering, and health.
  • Alexander Graham Bell Canada Graduate Scholarship
    Natural Sciences and Engineering Research Council of Canada
    (Declined) Ranked 16th out of 1878 doctoral applicants in Canada pursuing research in the natural sciences and engineering. Awarded to the top-ranked doctoral students in Canada, to ensure a reliable supply of qualified personnel to meet the needs of Canada's knowledge economy.
  • One of 12 Outstanding Program Committee Members
    Association for the Advancement of Artificial Intelligence (AAAI-20)
    Awarded in recognition of outstanding service on the program committee, characterized by exceptional care, thoroughness, and thoughtfulness in the reviews and discussions of the papers.
  • Graduate Mobility Award
    McGill University
    To support student engagement in international research experiences outside of McGill, the Graduate Mobility Award (GMA) helps cover expenses associated with a visit/stay at a research lab, field work abroad, and/or visit with a research collaborator.
  • Vadasz Doctoral Fellowship
    McGill University
    Les Vadasz is the cofounder of Intel. This fellowship was established in 2006 by the Vadasz Family Foundation to recruit outstanding students into the Faculty of Engineering's doctoral degree program.
  • Engineering Doctoral Award
    McGill University
    This award aims to recruit the best and brightest new doctoral students from all over the world. Selection for this award is based on the excellence of a student’s academic and research record, including publications, presentations and his or her potential to make a major impact on the quality of research in the Faculty of Engineering at McGill University and in the field of engineering.
  • Graduate Excellence Fellowship
    McGill University
    Nomination for this award is at the discretion of the Department or School.
  • Engineering Undergraduate Student Masters Award
    McGill University
    Competitive recruitment funding to assist in attracting and retaining high caliber undergraduate students with demonstrated academic excellence, as well as applied research experience, into the Faculty’s research Graduate Programs.
  • Rhodes Scholar Finalist
    University of Oxford
    One of 12 finalists selected for the Rhodes Scholarship, an international postgraduate award for students to study at the University of Oxford.
  • Deans Honour List (Designation for Graduating Students)
    McGill University
    Awarded to students in the top 10% of the faculty's graduating class. This calculation is based on the CGPA.
  • Ian McLachlin Prize for Entrepreneurship in Engineering
    McGill University
    Established in 1998 by Ian McLachlin, B.Eng. 1960, to encourage Engineering students to undertake Entrepreneurial Studies. Awarded by the Faculty of Engineering.
  • Les Vadasz Award in Engineering
    McGill University
    Les Vadasz is the cofounder of Intel. Awarded annually, by the Faculty of Engineering, to a student based on high academic standing with a preference given to an interest in, and contribution towards, engineering sustainability and/or design.
  • Faculty of Engineering Scholarship
    McGill University
    (Awarded to Top 5%) Established in 1992 by the University to provide awards based on academic achievement to students in the top 5% of the Faculty. Granted by the Faculty of Engineering Scholarships Committee based on good academic standing.
  • Accenture Prize in Engineering and Science
    McGill University
    Awarded by the Faculty Scholarships Committees, upon recommendation of the respective departments, on the basis of academic excellence and demonstrated leadership qualities.
  • NSERC Undergraduate Student Research Award
    McGill University
    Undergraduate student research award, granted by the university and the Natural Sciences and Engineering Research Council of Canada, based on the student’s academic record, research aptitude, and merit.
  • IEEE Signal Processing Society Winter School Scholarship
    Institute for Electrical and Electronics Engineers
    Awarded by the IEEE Signal Processing Society on the basis of merit to engineering students.
  • John Howard Ambrose Award
    McGill University
    Awarded by the Faculty of Engineering Scholarships Committee on the basis of merit to engineering students.
  • John Green Memorial Award
    McGill University
    For the student with highest standing in the penultimate year of Honours Electrical Engineering. Awarded on the recommendation of the Department.
  • 1st Place Google/PCH Hardware Hackathon
    Google
    (Awarded $5000 cash scholarship and $50,000 in product services; e.g., manufacturing, incorporation fees, etc.) Developed a solution and a prototype to provide Internet access to developing communities by utilizing a distributed data mesh with an offline search engine.
  • McCaig Family Scholarship in Engineering
    McGill University
    Awarded by the Faculty of Engineering to undergraduate students who have completed at least one year of the B.Eng. program on the basis of high academic standing.
  • Christie Steinmetz Award
    McGill University
    Awarded to a second year student in Electrical Engineering on the basis of academic performance and demonstration of exceptional promise as an engineer. Awarded on the recommendation of the Department.


Copyright © Mido Assran 2015

This page was made from scratch (icons included), so if you want to use any of it, I would appreciate it if you would just ask first.