Hanie Sedghi   UCI Wordmark

Research Scientist

Allen Institute for Artificial Intelligence (AI2)


Link to my CV.



Email: hsedghi@usc.edu


 About me

I work on large-scale machine learning, especially latent variable probabilistic models. My approach is to bond theory and practice in machine learning by designing algorithms with theoretical guarantees that also work efficiently in practice and lead the state of the art. In particular, I am interested in modeling machine learning problems as optimization frameworks and designing algorithms that have efficient guarantees in high dimensions while outperforming existing methods.
I got my Ph.D. from department of Electrical Engineering at University of Southern California with a minor in Mathematics in August 2015. My Ph.D. advisers are Professor Edmond Jonckheere and Professor Anima Anandkumar, UC Irvine. Before joining USC, I received my B.Sc. and M.Sc. in Electrical Engineering from Sharif University of Technology, Tehran, Iran in 2007 and 2009 respectively.


Research Interests

  • Theoretical and Applied Machine Learning and Statistics
  • Deep Learning
  • High-Dimensional Statistics and Large-Scale Machine Learning
  • Inference and Learning in Graphical Models and Latent Variable Models
  • Stochastic Optimization and Sparse Statistical Recovery


Honors and Awards

  • Certificate of Excellence in PhD studies, Association of Professors and Scholars of Iranian Heritage, 2015
  • USC Graduate School Research Fellowship, University of Southern California, Summer 2012
  • USC Provost Fellowship,University of Southern California, 2010-2014
  • Best Poster Award, MHI Research Festival, University of Southern California, Spring 2013
  • Financial Research Award, Iran Telecommunication Research Center, 2008
  • Recognized as a national elite by National Elite Foundation, Iran, 2007
  • Ranked 36th in the nationwide entrance exam for graduate degree in Electrical Engineering, Iran, 2007
  • Ranked 27th among 450000 participants in the universities entrance exam in Mathematics and Physics field, Iran, 2003



  • How Good Are My Predictions? Efficiently Approximating Precision-Recall Curves for Massive Datasets, by A. Sabharwal* and H. Sedghi*, accepted for plenary presentation at Uncertainty in Artificial Intelligence (UAI), 2017
    Download: PDF, Code.
  • Patent Pending
  • Knowledge Completion for Generics using Guided Tensor Factorization, by H. Sedghi and A. Sabharwal, accepted for publication in Transactions of the Association for Computational Linguistics (TACL), 2017
    Download: PDF, Code.
  • Training Input-Output Recurrent Neural Networks through Spectral Methods, by H. Sedghi and A. Anandkumar, March 2016
    Download: PDF.
  • Provable Tensor Methods for Learning Mixtures of Generalized Linear Models, by H. Sedghi, M. Janzamin and A. Anandkumar, accepted in Artificial Intelligence and Statistics (AISTATS) 2016.
    Download: PDF.
  • FEAST at Play: Feature ExtrAction using Score function Tensors, by M. Janzamin*, H. Sedghi*, UN Niranjan*, A. Anandkumar, In NIPS Feature Extraction: Modern Questions and Challenges, Montreal, Canada, December 2015.
    Download: PDF.
  • Beating the Perils of Non-Convexity: Guaranteed Training of Neural Networks using Tensor Methods, by M. Janzamin, H. Sedghi and A. Anandkumar, June. 2015.
    Download: PDF, video.
  • Learning Mixed Membership Community Models in Social Tagging Networks through Tensor Methods, by A. Anandkumar and H. Sedghi, April 2015.
    Download: PDF.
  • Score Function Features for Discriminative Learning: Matrix and Tensor Framework, by M. Janzamin, H. Sedghi and A. Anandkumar, Dec. 2014.
    Download: PDF.
  • Provable Methods for Training Neural Networks with Sparse Connectivity, by H. Sedghi and A. Anandkumar, accepted for presentation in Neural Information Processing Systems (NIPS) Deep Learning Workshop, Montreal, 2014. and in International Conference on Learning Representation (ICLR), May, 2015.
    Download: PDF.
  • Multi-Step Stochastic ADMM in High Dimensions: Applications in Sparse Optimization and Noisy Matrix Decomposition, by H. Sedghi, A. Anandkumar, E. Jonckheere. Neural Information Processing Systems (NIPS), Montreal, 2014.
    Download: NIPS version,PDF, video.
  • Statistical Structure Learning to Ensure Data Integrity in Smart Grid, by H. Sedghi and E. Jonckheere. Accepted for publication in IEEE Transactions on Smart Grid.
    Download: PDF.
  • Statistical Structure Learning of Smart Grid for Detection of False Data Injection, H. Sedghi and E. Jonckheere, IEEE Power and Energy Society General Meeting2013.
    Download: PDF.
  • On Conditional Mutual Information in Gaussian-Markov Structured Grids, H. Sedghi and E. Jonckheere, Information and Control in Networks, G. Como, B. Bernhardson, and A. Rantzer, vol. 450, pp 277-297, Springer.
    Download: PDF.
  • A Misbehavior-Tolerant Multipath Routing Protocol for Wireless Ad hoc Network, H. Sedghi, M.R. Pakravan and M. R. Aref, International Journal of Research in Wireless Systems, vol. 2, issue 2.
    Download: PDF.
  • A Game-Theoretic Approach for Power Allocation in Bidirectional Cooperative Communication, M.Janzamin, M. R. Pakravan and H. Sedghi, IEEE Wireless Communication and Networking Conference (WCNC) Sydney, 2010.
    Download: PDF.