Karim Abou-Moustafa, Ph.D.

sr. applied ai research scientist
intel       chandler, az


Research |  Updates |  Publications |  Professional Services |  Contact 

[Research]

 Google Scholar profile    Semantic Scholar profile    DBLP profile    ORCID profile    LinkedIn profile    IEEE Explore profile

As of July 2021, I'm a Sr. Applied AI Research Scientist at Intel's R&D division known as Intel TD (Technology Development).

I work on various aspects of statistical learning algorithms and their confluence with problems in high-dimensional data analysis, statistical pattern recognition, and computer vision. More broadly, I am interested in computational, statistical, and information(al) aspects related to learning algorithms and learning from high-dimensional data.

Topics I've worked on include (see publications below):

  • Robust estimation of high-dimensional covariance matrices for applications such as anomaly detection and out-of-distribution detection.
  • Stability of learning algorithms for deriving strong guarantees on the generalization performance of learning algorithms. In particular, I've worked on the concentration of risk estimates, developed in terms of LOOCV and KFCV, around the expected risk of a learning algorithm under different notions of algorithmic stability.
  • Dimensionality reduction and low-dimensional embedding; this includes supervised and unsupervised algorithms for linear/nonlinear dimensionality reduction using models and techniques from discriminant analysis, kernel methods, manifold learning algorithms, and graph embedding algorithms. This line of work was complemented with the development of scalable and distributed algorithms for feature selection and interaction detection between predictor variables for regression and classification problems during my time at SAS Inc.
  • Generative-Discriminative models for time-series data classification.

At Intel, I lead projects on (i) anomaly detection and out-of-distribution detection for image data and high-dimensional data in general, and (ii) active learning from severely imbalanced datasets.

[Updates]

  • Area Chair for NeurIPS 2024

  • Preliminary work on robust estimation of high-dimensional covariance matrices using Tyler's M-Estimator was accepted for publication in IEEE Information Theory Workshop in Saint-Malo, France (April 2023).

[Refereed Publications]

  • Karim Abou-Moustafa
    "Shrinkage Coefficient Estimation for Regularized Tyler's M-Estimators: A Leave-One-Out Approach",
    Proceedings of IEEE Information Theory Workshop (ITW), Saint-Malo, France, pp. 335-340, 2023.
    [Online Link] [BibTeX] [PDF]

  • Karim Abou-Moustafa and Csaba Szepesvári
    "An Exponential Efron-Stein Inequality for Lq Stable Learning Rules",
    Algorithmic Learning Theory (ALT), Proceedings of Machine Learning Research, Vol. 98, pp. 31-63, 2019.
    [Online Link] [BibTeX] [PDF]

  • Karim Abou-Moustafa and Csaba Szepesvári
    "An Exponential Tail bound for the Deleted Estimate",
    Thirty-Third AAAI Conference on Artificial Intelligence, 2019.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa and Frank Ferrie
    "Local Generalized Quadratic Distance Metrics: Application to the k-nearest neighbors classifier",
    Journal of Advances in Data Analysis and Classification (ADAC), Vol. 12, No. 2, pp. 341-363, 2018.
    [Online Link] [BibTeX] [PDF]

  • Karim Abou-Moustafa and Dale Schuurmans
    "Generalization in Unsupervised Learning",
    European Conference on Machine Learning (ECML), 2015.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Fernando De La Torre, and Frank Ferrie
    "Pareto Models for Multiclass Discriminative Linear Dimensionality Reduction",
    Pattern Recognition, Vol. 48, No. 5, pp. 1863-1877, 2015.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Dale Schuurmans, and Frank Ferrie
    "Learning a Metric Space for Neighbourhood Topology Estimation",
    Asian Conf. on Machine Learning, JMLR W&CP 29: pp. 341-356, 2013.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Frank Ferrie and Dale Schuurmans
    "Divergence Based Graph Estimation for Manifold Learning",
    IEEE Global Conf. on Signal and Information Processing, Austin, TX, Dec. 2013.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa and Frank Ferrie,
    "A Note on Metric Properties for Some Divergence Measures: The Gaussian Case",
    Asian Conf. on Machine Learning, JMLR W&CP 25: pp. 1-15, 2012.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa and Frank Ferrie,
    "Modified Divergence Measures for Gaussian Densities",
    LNCS 7626, Proc. of the IAPR Int. Workshop on Structural, Syntactic, Statistical Pattern Recognition (S+SSPR), pp. 426-436, Springer 2012.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa and Frank Ferrie,
    "A Framework for Hypothesis Learning Over Sets of Vectors",
    Proc. of ACM's SIGKDD 9th Workshop on Mining and Learning with Graphs, 2011.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Mohak Shah, Fernando De La Torre, and Frank Ferrie,
    "Relaxed Exponential Kernels for Unsupervised Learning",
    LNCS 6835, Pattern Recognition, Proc. of the 33rd DAGM Symposium, pp. 184-195, Springer, 2011.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Fernando De La Torre, and Frank Ferrie,
    "Designing a Metric for the Difference Between Two Gaussian Densities",
    Advances in Intelligent and Soft Computing; J. Angeles, B. Boulet, J. Clark, J. Kovecses and K. Siddiqi (Eds.), Vol. 83, pp. 57 - 70, Springer, Dec. 2010.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Fernando De La Torre, and Frank Ferrie,
    "Pareto Discriminant Analysis",
    IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pp. 3602 - 3609, 2010.
    [BibTeX] [PDF] [and here]

  • Karim Abou-Moustafa and Frank Ferrie,
    "Local Metric Learning on Manifolds with Applications to Query-based Operations",
    LNCS 5342, Proc. of the IAPR Int. Workshop on Structural, Syntactic, Statistical Pattern Recognition (S+SSPR), pp. 872 - 838, Springer, 2008.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa and Frank Ferrie,
    "Fast and Regularized Local Metric for Query-based Operations",
    IEEE Proc. of the 19th Int. Conf. on Pattern Recognition (ICPR), 2008.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa and Frank Ferrie,
    "The Minimum Volume Ellipsoid Metric",
    LNCS 4713, Pattern Recognition, Proc. of the 29th DAGM Symposium, pp. 335 - 344, Springer, 2007.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Mohamed Cheriet, and Ching Suen,
    "Classification of Time-Series Data Using a Generative/Discriminative Hybrid",
    IEEE Proc. of the 9th Int. Workshop on Frontiers in Handwriting Recognition (IWFHR), pp. 51 - 56, 2004.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Mohamed Cheriet, and Ching Suen,
    "A Generative-Discriminative Hybrid for Sequential Data Classification",
    IEEE Proc. of the Int. Conf. on Acoustics, Speech & Signal Processing (ICASSP), Vol. 5, pp. 805 - 808, 2004.
    [BibTeX] [PDF]

  • Karim Abou-Moustafa, Mohamed Cheriet, and Ching Suen,
    "On The Structure of Hidden Markov Models",
    Pattern Recognition Letters, Vol. 25, pp. 923 - 931, June 2004.
    [BibTeX] [PDF]

[Non-Refereed Publications]

  • Karim Abou-Moustafa and Csaba Szepesvári
    "An a Priori Exponential Tail Bound for k-Folds Cross-Validation",
    [arXiv:1706.05801], 2017.

  • Karim Abou-Moustafa,
    "What is The Distance Between Objects in a Data Set?",
    IEEE Pulse Magazine, Vol. 7, No. 2, pp. 41-47, March-April, 2016.
    [BibTeX] [link]

  • Karim Abou-Moustafa,
    "On Derivatives of Eigenvalues and Eigenvectors of the Generalized Eigenvalue Problem",
    McGill Tech. Report No. TR-CIM-10-09, 2009.
    [BibTeX] [PDF]

[Theses]

  • Karim T. Abou-Moustafa, "Metric Learning Revisited. New Approaches for Supervised and Unsupervised Metric Learning with Analysis and Algorithms", Ph.D. Thesis, McGill University, Montréal, QC, Canada, 2011.

  • Karim T. Abou-Moustafa, "A Generative-Discriminative Framework for Time-Series Data Classification", Masters Thesis, Concordia University, Montréal, QC, Canada, 2004.

[Professional Services]

[Contact Information]