Theofanis Karaletsos

Machine Learning Researcher
theofanis.karaletsos [@] gmail.com

I am a senior research scientist at Uber AI Labs in San Francisco CA, working on machine intelligence. My area of interest is probabilistic machine learning, deep learning, probabilistic programming (see pyro.ai), and their applications in fields as diverse as simulation, reinforcement learning, healthcare, biology, spatiotemporal modeling, vision, language, and large-scale economics (such as spatiotemporal markets, two-sided marketplaces, agent models, and agent behavior). An important focus of my work has been building systems that both quantify and consume uncertainty, and also are imbued with useful inductive biases. I seek to build robust, data-efficient models of complex systems that allow us to understand and control the world around us. 
In a previous life, I was a researcher at AI-startup Geometric Intelligence (which was acquired by Uber to form Uber AI Labs), a researcher at the Sloan Kettering Institute in New York, and a PhD student at the Max Planck Institute For Intelligent Systems.

Papers


Preprints

Hierarchical Gaussian Process Priors for Bayesian Neural Network Weights
Theofanis Karaletsos, Thang D. Bui.
full version of AABI 2019 paper, 2020
arXiv

Probabilistic Meta-Representations For Neural Networks
Theofanis Karaletsos, Peter Dayan, Zoubin Ghahramani.
Uncertainty in Artificial Intelligence: Uncertainty In Deep Learning Workshop 2018.
arXiv

Automatic Relevance Determination For Deep Generative Models
Theofanis Karaletsos, Gunnar Rätsch.
Pre-print 2015
arXiv

Conference Publications

Tyche: Bayesian Neural Networks in 5 lines of code
Hippolyt Ritter, Theofanis Karaletsos.
International Conference on Probabilistic Programming (PROBPROG) 2020.

Generalized Hidden Parameter MDPs: Transferable Model-Based RL in a handful of trials
Christian Perez, Felipe Such, Theofanis Karaletsos.
Proceedings Of the AAAI Conference On Artificial Intelligence (AAAI) 2020 (full oral).
arXiv

Likelihood-Free Inference with emulator networks
Jan-Matthis Lückmann, Giacomo Bassetto, Theofanis Karaletsos, Jakob H. Macke.
Proceedings of The 1st Symposium on Advances in Approximate Bayesian Inference (AABI), PMLR 96:32-53,2019, 2019.
PMLR

Pathwise Derivatives For Multivariate Distributions
Martin Jankowiak, Theofanis Karaletsos.
Proceedings Of the International Conference on Artificial Intelligence and Statistics (AISTATS), 2018.
Conditional Similarity Networks
Andreas Veit, Serge Belongie, Theofanis Karaletsos.
Computer Vision and Pattern Recognition (CVPR), 2017.
arXiv

Adversarial Message Passing For Graphical Models
Theofanis Karaletsos.
NIPS Advances In Approximate Bayesian Inference (AABI), 2016.
arXiv

Bayesian unsupervised representation learning with oracle constraints
Theofanis Karaletsos, Serge Belongie, Gunnar Rätsch.
International Conference on Learning Representations (ICLR), 2016.
arXiv

A Generative Model Of Words and Relationships from Multiple Sources
Stephanie Hyland, Theofanis Karaletsos, Gunnar Rätsch.
AAAI Conference On Artificial Intelligence (AAAI) (spotlight talk) , 2016.
arXiv

Knowledge Transfer with Medical Language Embeddings
Stephanie Hyland, Theofanis Karaletsos, Gunnar Rätsch.
SDM-DMMH, also shot version in NIPS Workshop Machine Learning in Healthcare 2015 , 2016.

An Empirical Analysis of Topic Modeling for Mining Cancer Clinical Notes
Katherine Redfield Chan, Xinghua Lou, Theofanis Karaletsos, Christopher Crosbie,Stuart Gardos, David Artz, Gunnar Rätsch.
ICDM-BioDM, 2013.

Journal Publications

Pyro: Deep Universal Probabilistic Programming
Eli Bingham, Jonathan P. Chen. Martin Jankowiak, Neeraj Pradhan, Theofanis Karaletsos, Rohit Singh, Paul Szerlip, Paul Horsfall, Noah D. Goodman.
Journal of Machine Learning Research, 2018.
arXiv

RiboDiff: Detecting Changes of Translation Efficiency from Ribosome Foot-prints
Yi Zhong, Theofanis Karaletsos, Philipp Drewe, Vipin Thankam T Sreedharan, Kamini Singh, Hans-Guido Wendel, Gunnar Rätsch.
Oxford Press Bioinformatics, 2016.
BiorXiv

ShapePheno: Unsupervised extraction of shape phenotypes from biological image collections
Theofanis Karaletsos, Oliver Stegle, Christine Dreyer, John Winn, Karsten Borgwardt.
Oxford Press Bioinformatics, 2012.
Link

Workshop Papers

Gaussian Process Meta-Representations For Neural Networks
Theofanis Karaletsos, Thang Bui.
2nd Symposium on Advances in Approximate Bayesian Inference, 2019.

Generalized Hidden Parameter MDPs,
Christian Perez, Felipe Such, Theofanis Karaletsos.
International Conference For Machine Learning: Generative Modeling and Model-Based Reasoning For Robotics and AI workshop, 2019.

Applying SVGD to bayesian neural networks for cyclical time-series prediction and inference
Xinyu Hu, Paul Szerlip, Theofanis Karaletsos, Rohit Singh.
Neural Information Processing Systems: Bayesian Deep Learning Workshop, 2018.
arXiv

Efficient transfer learning and online adaptation with latent variable models for continuous control
Christian Perez, Felipe Such, Theofanis Karaletsos.
Neural Information Processing Systems: Continual Learning Workshop 2018, 2018.
arXiv

A Generative Model Of Words and Relationships from Multiple Sources
Stephanie Hyland, Theofanis Karaletsos, Gunnar Rätsch.
IWES Workshop for Learning Semantics, 2015.

Probabilistic Disease Progression Models For Retrospective Analysis Of Cancer Health Records
Theofanis Karaletsos, Stefan Stark, Gunnar Rätsch.
NIPS Workshop Machine Learning in Healthcare, 2015.

Poisson Matrix Factorization For Joint Modeling Of Genetics and Medical Text
Melanie Fernandez, Theofanis Karaletsos, Julia Vogt, Stephanie Hyland, Gunnar Rätsch, Fernando Perez-Cruz.
NIPS Workshop Machine Learning in Healthcare, 2015.

Towards an integrated dynamic model of temporal structure of clinical textnotes and interactions with genetic profiles
Theofanis Karaletsos, X. Lou, K.R.Chan, C. Crosbie, G. Rätsch.
NIPS Machine Learning for Clinical Data Analysis in Healthcare, 2013.

JigPheno: Semantic Feature Extraction From Biological Images
Theofanis Karaletsos, Oliver Stegle, John Winn, Karsten Borgwardt.
NIPS Machine Learning in Computational Biology (oral), 2010.
Video
 

Open Source Software


Pyro
Pyro is a deep, universal probabilistic programming language written in Python on top of PyTorch. For more information, check out the release blog.

Tyche
Bayesian Neural Network Toolbox for Pytorch

Patents


Representations Of Units in Neural Networks
Theofanis Karaletsos, Peter Dayan, Zoubin Ghahramani.
US Patent App., 16/356,991 2019.

Model Based Reinforcement Learning Based On Generalized Hidden Parameter Markov Decision Processes
Christian Perez, Felipe Such, Theofanis Karaletsos.
US Patent App., 62/851,858 2019.

Event detection using sensor data
NP Volk, Theofanis Karaletsos, U Madhow, JB Yosinski,TR Sumers.
US Patent App., 16/233,779 2019.

Systems and Methods For Intelligent Regularization of Neural Network Archi-tectures
Zoubin Ghahramani, Douglas Bemis, Theofanis Karaletsos.
US Patent App., 15/789,898 2018.