Contact information:

Sammy Khalife
Cornell Tech
Bloomberg Building
2 W Loop Rd, New York City
New York City, NY 10044
Email: khalife.sammy[AT]cornell[DOT]edu

I'm a postdoctoral fellow at Cornell Tech, mentored by Andrea Lodi. Previously, I was a postdoctoral fellow in the Department of Applied Mathematics and Statistics at Johns Hopkins University, mentored by Amitabh Basu. My current research interests are a mix of discrete optimization, theoretical deep learning, and data science. More specifically, I am interested in:

  • Formal expressivity of Graph Neural Networks and role of the activation function
  • Algorithm selection in optimization: sample complexity of learning branch-and-cut rules in integer programming (and beyond)
  • Computational complexity aspects of Neural Networks: what are the precise numbers of layers and neurons required to compute a target function?
  • Applications of these tools in natural language processing (language models), bioinformatics (determining 3D-structure of proteins) and network science.

Teaching

  • Johns Hopkins University

    • Discrete Mathematics 553.171 (Fall 2021, 2022, Spring 2023)

    • Numerical Linear Algebra 553.385 (Spring 2022) Course link

  • Ecole Polytechnique, France

    • Data Science Starter Program (2017-2019, Formation Big Data, EXED)

    • Machine learning INF554 (2019, Teaching Assistant)

  • Saint-Joseph University, Beirut

    • Master Data Science: Graph Theory, Probabilistic Graphical Models (Spring 2019, Instructor)

Research Papers

Preprints/Papers Under Review

  • Is uniform expressivity too restrictive? Towards efficient expressivity of graphs neural networks, (with J. Tonelli-Cueto)
    [arXiv]

  • The logic of rational graph neural networks.
    [arXiv]

  • On the power of graph neural networks and the role of the activation function, (with A. Basu)
    [arXiv]

  • Sequence graphs realizations and ambiguity in language models (with Y. Ponty and L. Bulteau)
    [arXiv]

Refereed Publications

  • Sample complexity of algorithm selection using neural networks and its applications to branch-and-cut (with H. Cheng, B. Fiedorowicz and A. Basu), to appear in the Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2024
    [arXiv]

  • Neural networks with linear threshold activations: structure and algorithms (with H. Cheng, H., A. Basu (2023), Mathematical Programming, 2023 (Extended version)
    [arXiv]

  • Neural networks with linear threshold activations: structure and algorithms (with A. Basu), In Integer Programming and Combinatorial Optimization (IPCO) 2022.
    [arXiv]

  • Further results on latent discourse models and word embeddings (with D. Gonçalves, D., Y. Allouah, and L. Liberti) Journal of Machine Learning Research, 22, pp. 1-36, November 2021.
    [Link]

  • Distance geometry for word embeddings and applications (with D. Gonçalves and L. Liberti). Journal of Computational Mathematics and Data Science, 2021.
    [Link]

  • Sequence graphs realizations and ambiguity in language models, (with Y. Ponty, and L. Bulteau L), In International Computing and Combinatorics Conference (COCOON), 2021
    [Link]

  • Secondary structure assignment of proteins in the absence of sequence information, (with T. Malliavin, and L. Liberti), Bioinformatics Advances, Volume 1, Issue 1, 2021.
    [Link]

  • Sequence graphs: characterization and counting of admissible elements. Cologne-Twente Workshop on Graphs and Combinatorial Optimization, 2020.
    [Slides]

  • Structure and influence analysis of worldwide capitalistic ownership, (with S. Khalife, J. Read, and M. Vazirgiannis), Journal of Applied Network Science (2020).
    [Link]

  • Geometry and analogies: a study and propagation method for word representations, (L. Liberti and M. Vazirgiannis) In International Conference on Statistical Language and Speech Processing, 2019, pages 100–111. Springer
    [Link]

  • Scalable graph-based method for individual named entity identification, (with M. Vazirgiannis),In Proceedings of the Thirteenth Workshop on Graph-Based Methods for Natural Language Processing (TextGraphs-13), 2019, pages 17–25,
    [Link]

  • Empirical analysis of a global capital-ownership network, (with J. Read, and M. Vazirgiannis), In International Conference on Complex Networks and Their Applications, 2019, pages 656–667. Springer
    [Link]

  • Detection of sleep spindles in NREM 2 sleep stages: Preliminary study & benchmarking of algorithms, (with O. Pallanca and J. Read). In Zheng, H. J., Callejas, Z., Griol, D., Wang, H., Hu, X., Schmidt, H. H. H. W., Baumbach, J., Dickerson, J., and Zhang, L., editors, IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2018, Madrid, Spain, December 3-6, 2018, pages 2652–2655. IEEE Computer Society
    [Link]

Other expository notes

  • Khalife, S (2024). Notes about algorithm design for cross-matching
    [Notes]

Dissertations

  • Thèse (Ecole Polytechnique): Graphs, Geometry and Representations for Language Models and Networks of Entities
    Advisor: Michalis Vazirgiannis