Announcement

About

I am currently a Faculty Fellow at the NYU Center for Data Science. Previously, I received my PhD in Electrical Engineering and Computer Science from MIT, where I was fortunate to be advised by Pablo Parrilo. Before that, I received my undergrad degree from Princeton.

My research interests are broadly at the interface of optimization, probability, and machine learning. One recent focus has been computational aspects of problems related to optimal transport.

I am interested in practical implementations as well as theory, and have previously interned in both tech (Apple Research, Google Research, Google) and Wall Street (DE Shaw, Tower Research).

In my free time, I like to ski, play tennis, and eat too many cookies. I also like chess and, in a past life, I obtained an International Master norm in Spain.

Publications

Preprints

• Flows, Scaling, and Entropy Revisited: a Unified Perspective via Optimizing Joint Distributions
  Jason Altschuler
  SIAM Group on Optimization's Views and News, invited expository article, to appear in 2022.
  

• Privacy of Noisy Stochastic Gradient Descent: more iterations without more privacy loss
  Jason Altschuler, Kunal Talwar.
  NeurIPS, to appear in 2022.
  

• Kernel approximation on algebraic varieties
  Jason Altschuler, Pablo Parrilo.
  SIAM Journal on Applied Algebra and Geometry (SIAGA), accepted with minor revisions, 2022.
  

Journal articles

• Polynomial-time algorithms for Multimarginal Optimal Transport problems with structure
  Jason Altschuler, Enric Boix-Adsera.
  Mathematical Programming (MAPR), 2022.
  

• Asymptotics for semi-discrete entropic optimal transport
  Jason Altschuler, Jonathan Niles-Weed, Austin Stromme.
  SIAM Journal on Mathematical Analysis (SIMA), 2022.
  

• Approximating Min-Mean-Cycle for low-diameter graphs in near-optimal time and memory
  Jason Altschuler, Pablo Parrilo.
  SIAM Journal on Optimization (SIOPT), 2022.
  

• Near-linear convergence of the Random Osborne algorithm for Matrix Balancing
  Jason Altschuler, Pablo Parrilo.
  Mathematical Programming (MAPR), 2022.
  

• Wasserstein barycenters are NP-hard to compute
  Jason Altschuler, Enric Boix-Adsera.
  SIAM Journal on Mathematics of Data Science (SIMODS), 2022.
  

• Wasserstein barycenters can be computed in polynomial time in fixed dimension [poster, talk, slides]
  Jason Altschuler, Enric Boix-Adsera.
  Journal of Machine Learning Research (JMLR), 2021.
  

• Hardness results for Multimarginal Optimal Transport problems
  Jason Altschuler, Enric Boix-Adsera.
  Discrete Optimization (DISOPT), 2021.
  

• Online learning over a finite action set with limited switching [poster, talk]
  Jason Altschuler, Kunal Talwar.
  Mathematics of Operations Research (MOR), 2021. (Preliminary conference version in COLT)

• Lyapunov Exponent of Rank One Matrices: Ergodic Formula and Inapproximability of the Optimal Distribution [slides]
  Jason Altschuler, Pablo Parrilo.
  SIAM Journal on Control and Optimization (SICON), 2020. (Preliminary conference version in CDC)

• Best arm identification for contaminated bandits
  Jason Altschuler, Victor-Emmanuel Brunel, Alan Malek.
  Journal of Machine Learning Research (JMLR), 2019.

• Inclusion of forbidden minors in random representable matroids
  Jason Altschuler, Elizabeth Yang.
  Discrete Mathematics, 2017.

• Rapid analysis and exploration of fluorescence microscopy images
  Benjamin Pavie, Satwik Rajaram, Austin Ouyang, Jason Altschuler, Robert Steininger, Lani Wu, Steven Altschuler.
  Journal of Visualized Experiments (JoVE), 2014.

Conference proceedings

• Averaging on the Bures-Wasserstein manifold: dimension-free convergence of gradient descent
  Jason Altschuler, Sinho Chewi, Patrik Gerber, Austin Stromme.
  Neural Information Processing Systems (NeurIPS), 2021. (Selected for Spotlight presentation)
  

• Lyapunov Exponent of Rank One Matrices: Ergodic Formula and Inapproximability of the Optimal Distribution [slides]
  Jason Altschuler, Pablo Parrilo.
  Conference on Decision and Control (CDC), 2019.

• Massively scalable Sinkhorn distances via the Nyström method
  Jason Altschuler, Francis Bach, Alessandro Rudi, Jonathan Niles-Weed.
  Neural Information Processing Systems (NeurIPS), 2019.
  

• Online learning over a finite action set with limited switching [poster, talk]
  Jason Altschuler, Kunal Talwar.
  Conference on Learning Theory (COLT), 2018.

• Near-linear time approximation algorithms for optimal transport via Sinkhorn iteration [poster]
  Jason Altschuler, Jonathan Weed, Philippe Rigollet.
  Neural Information Processing Systems (NeurIPS), 2017. (Selected for Spotlight presentation)
  

• Greedy column subset selection: new bounds and distributed algorithms [poster, slides]
  Jason Altschuler, Aditya Bhaskara, Gang Fu, Vahab Mirrokni, Afshin Rostamizadeh, Morteza Zadimoghaddam.
  International Conference on Machine Learning (ICML), 2016.
  

Theses

• Transport and Beyond: Efficient Optimization over Probability Distributions
  PhD Thesis, MIT, 2022. Advised by Pablo Parrilo.

• Minimax rates for online learning with limited decision changes (superseded by this paper in MOR/COLT)
  Undergrad thesis, Princeton CS department, 2016. Advised by Elad Hazan.
  

• Probabilistic linear Boolean classification
  Undergrad thesis, Princeton math department, 2016. Advised by Emmanuel Abbe.
  

Teaching

• COS 511: Theoretical machine learning (grad course) -- Spring 2016
• MAT 340: Applied algebra -- Fall 2015
• MAT 216: Honors real analysis -- Fall 2014
• MAT 217: Honors linear algebra -- Spring 2014
• MAT 215: Honors real analysis -- Fall 2013

Disambiguation