Mario Diaz

Mario Diaz

Research Associate
Instituto de Investigaciones en Matemáticas Aplicadas y en Sistemas (IIMAS)
Universidad Nacional Autónoma de México (UNAM)
Curriculum Vitae


Interests

My current research mainly focuses on theoretical machine learning, data privacy and information theory.
I also keep a research stream on random matrices, free probability theory and their applications.


Publications and Preprints

Journal Papers

  1. On the global fluctuations of block Gaussian matrices
    M. Diaz, J. Mingo and S. Belinschi
    Probability Theory and Related Fields, 2020
    DOI:10.1007/s00440-019-00925-1
     
  2. On the robustness of information-theoretic privacy measures and mechanisms
    M. Diaz, H. Wang, F. Calmon and L. Sankar
    IEEE Transactions on Information Theory, 2019
    DOI:10.1109/TIT.2019.2939472
     
  3. Estimation efficiency under privacy constraints
    S. Asoodeh, M. Diaz, F. Alajaji and T. Linder
    IEEE Transactions on Information Theory, 2019
    DOI:10.1109/TIT.2018.2865558
     
  4. On the capacity of block multiantenna channels
    M. Diaz and V. Pérez-Abreu
    IEEE Transactions on Information Theory, 2017
    DOI:10.1109/TIT.2017.2712711
     
  5. Information extraction under privacy constraints
    S. Asoodeh, M. Diaz, F. Alajaji and T. Linder
    Information, 2016
    DOI:10.3390/info7010015

Conference Papers

  1. A tunable loss function for binary classification
    T. Sypherd, M. Diaz, L. Sankar and P. Kairouz
    IEEE International Symposium on Information Theory (ISIT), 2019
    DOI:10.1109/ISIT.2019.8849796
     
  2. Generalization bounds via Wasserstein distance
    H. Wang, M. Diaz, JCS Santos Filho and F. Calmon
    IEEE International Symposium on Information Theory (ISIT), 2019
    DOI:10.1109/ISIT.2019.8849359
     
  3. The utility cost of robust privacy guarantees
    H. Wang, M. Diaz, F. Calmon and L. Sankar
    IEEE International Symposium on Information Theory (ISIT), 2018
    DOI:10.1109/ISIT.2018.8437735
     
  4. Privacy-aware guessing efficiency
    S. Asoodeh, M. Diaz, F. Alajaji and T. Linder
    IEEE International Symposium on Information Theory (ISIT), 2017
    DOI:10.1109/ISIT.2017.8006629
     
  5. On the symmetries and the capacity achieving input covariance matrices of multiantenna channels
    M. Diaz
    IEEE International Symposium on Information Theory (ISIT), 2016
    DOI:10.1109/ISIT.2016.7541464

Preprints

  1. To Split or Not to Split: The Impact of Disparate Treatment in Classification
    H. Wang, H. Hsu, M. Diaz and F. Calmon
    arXiv:2002.04788
     
  2. Privacy amplification of iterative algorithms via contraction coefficients
    S. Asoodeh, M. Diaz and F. Calmon
    arXiv:2001.06546
     
  3. Fluctuations for matrix-valued Gaussian processes
    M. Diaz, A. Jaramillo and J.C. Pardo
    arXiv:2001.03718
     
  4. Theoretical guarantees for model auditing with finite adversaries
    M. Diaz,P. Kairouz, J. Liao and L. Sankar
    arXiv:1911.03405
     
  5. A class of parameterized loss functions for classification: optimization tradeoffs and robustness characteristics
    T. Sypherd, M. Diaz, H. Laddha, L. Sankar, P. Kairouz and G. Dasarathy
    arXiv:1906.02314


Teaching


Miscellaneous Academic Activities


News

Our paper, "On the alpha-loss Landscape in the Logistic Model", has been accepted to ISIT 2020! This is joint work with @MDMarioDiaz, Lalitha Sankar, and Gautam Dasarathy. See https://t.co/OCnhelCEp1 for all accepted papers to ISIT 2020.

— Tyler Sypherd (@SypherdTyler) March 31, 2020

Our paper "Privacy Amplification of Iterative Algorithms via Contraction Coefficients" has been accepted for presentation at ISIT 2020!https://t.co/j94Pqr2LiQ

— Mario Diaz (@MDMarioDiaz) March 31, 2020

Our paper has been accepted by FORC 2020! If disparate treatment is allowed, is it always "beneficial" to train and deploy separate models on different populations? Check here: https://t.co/kWbPLRmdZe for an answer! This is a joint work with @MDMarioDiaz Hsiang Hsu @FlavioCalmon https://t.co/W7njfLmbv3

— Hao Wang (@HW_HaoWang) March 29, 2020

Interested in quantifying the effect of disparate treatment on prediction accuracy? Our latest (information theoretic) take on this problem might be of interest for you!https://t.co/IUyWz5iPo6

— Mario Diaz (@MDMarioDiaz) February 13, 2020

In this work with Shahab Asoodeh and @FlavioCalmon, we analyzed the differential privacy guarantees of iterative algorithms from an information theoretic perspective, relying on the contractivity of certain Markov kernels under a specific f-divergence.https://t.co/SfCmA47tWG

— Mario Diaz (@MDMarioDiaz) January 22, 2020

About two years ago this project started as a Sunday reading group over Skype between Arturo and I. So happy that this preprint is finally out! More about fluctuations of matricial processes to come...https://t.co/UwFc3d1ldk

— Mario Diaz (@MDMarioDiaz) January 14, 2020

Our contributed talk got accepted at SIAM Conference on Mathematics of Data Science #SIAMMDS20 A big shout-out to @SypherdTyler, great job! https://t.co/164ZwVPTcj

— Mario Diaz (@MDMarioDiaz) January 13, 2020

Last update: March 2020