Tamir Hazan

Our research interests involve the theoretical and practical aspects of machine learning. Our research focuses on mathematically founded solutions to modern real life problems that demonstrate non-traditional statistical behavior. Recent examples include efficient learning of high dimensional statistics using Gumbel-max perturbation mdoels in discriminative learning, generative learning and reinforcement learning. We also learn graph based attention models across modalities and consider different aspects of Bayesian deep learning using Gaussian perturbations of their parameters. The practice of our work is motivated by many visual and language problems.



tamir.hazan@technion.ac.il

Edited Volumes

Perturbations, Optimization, and Statistics (2016)
Tamir Hazan, George Papandreou, Daniel Tarlow (Editors).
Neural Information Processing series, MIT Press.
[MIT Press], [amazon]

Students

  • Alex Schwing Asisstant professor at UIUC.
  • Alon Cohen Associate professor at TAU.
  • Idan Schwartz Postdoc at TAU.
  • Guy Lorberbom (Ph.D. student)
  • Itai Gat (Ph.D. student)
  • Hedda Cohen (Ph.D. student)
  • Adi Manos (M.Sc. student)
  • Ram Yazdi (M.Sc. student)
  • Chana Ross (M.Sc. student)
  • Bar Mayo (M.Sc. student)
  • Tom Ron (M.Sc. student)
  • Alon Berliner (M.Sc. student)
  • Vered Halperin (M.Sc. student)
  • Gilad Goldreich (M.Sc. student)
  • Meghan Lahmi (M.Sc. student)
  • Avrech Ben-David (M.Sc. student)

Research

Attention models: Deep learning revolutionized AI and machine learning techniques can be used to achieve human-like behavior. We explore attention models that allow to interprate and improve the prediction process of a learner.
Relevant materials: High Order Attention Models for Visual Question Answering, Factor Graph Attention, Audio-Visual Scene-Aware Dialog.
visual question answering factor graph attention scene aware dialog

Perturbation models: Statistically reasoning about complex systems involves a probability distribution over exponentially many configurations. For example, semantic labeling of an image requires to infer a discrete label for each image pixel, hence resulting in possible segmentations which are exponential in the numbers of pixels. Standard approaches such as Gibbs sampling are slow in practice and cannot be applied to many real-life problems. Our goal is to integrate optimization and sampling through extreme value statistics and to define new statistical framework for which sampling and parameter estimation in complex systems are efficient. This framework is based on measuring the stability of prediction to random changes in the potential interactions.
Relevant materials: discrete variational auto encoders, UAI 2014 tutorial, correlation clustering, online structured learning, measure concentration, learning with inverse optimization, entropy bounds and interactive annotations, sampling from the Gibbs distribution using max-solvers, learning with super-modular tasks and non-decomposable loss functions, the partition function and extreme value statistics.
max-solution perturb-max samples

Markov random fields, convex duality and message-passing: To efficiently predict outcomes in complex systems one can use graphical models and structured predictors. In this setting each predictor provides partial outcomes, e.g., the semantic labels of a region in an image, and global consistency for the structured prediction is maintained by passing messages between these regions. These concepts, emerging from Judea Pearl’s belief propagation algorithm, can be interpreted in terms of optimization theory. Applying Fenchel duality we develop the convex norm-product belief propagation, and its high-oder extensions, which enforce consistency between overlapping predictors using dual block coordinate descent. This provides us with the means to use cloud computing platforms to distribute and parallelize the prediction while maintaining consistency between its subproblems (code available, also for Amazon EC2). Estimating the parameters of region based predictors increase their accuracy in many real-life programs, and currently we achieve with these methods state-of-the-art results in various computer vision applications.

Publications

    2023

  1. Learning Constrained Structured Spaces with Application to Multi-Graph Matching. Hedda Indelman, Tamir Hazan. International Conference on Artificial Intelligence and Statistics
  2. [ PDF, BibTeX ]
  3. Layer Collaboration in the Forward-Forward Algorithm. Guy Lorberbom, Itai Gat, Yossi Adi, Alex Schwing, Tamir Hazan. arXiv preprint arXiv:2305.12393
  4. [ PDF, BibTeX ]

    2022

  5. Video and text matching with conditioned embeddings. Ameen Ali, Idan Schwartz, Tamir Hazan, Lior Wolf. Proceedings of the IEEE/CVF winter conference on applications of computer vision
  6. [ PDF, BibTeX ]
  7. Latent space explanation by intervention. Itai Gat, Guy Lorberbom, Idan Schwartz, Tamir Hazan. Proceedings of the AAAI Conference on Artificial Intelligence
  8. [ PDF, BibTeX ]
  9. Walking direction estimation using smartphone sensors: a deep network-based framework. Adi Manos, Tamir Hazan, Itzik Klein. IEEE Transactions on Instrumentation and Measurement
  10. [ PDF, BibTeX ]
  11. Dual Decomposition of Convex Optimization Layers for Consistent Attention in Medical Images. Tom Ron, Tamir Hazan. International Conference on Machine Learning
  12. [ PDF, BibTeX ]
  13. A functional information perspective on model interpretation. Itai Gat, Nitay Calderon, Roi Reichart, Tamir Hazan. International Conference on Machine Learning
  14. [ PDF, BibTeX ]
  15. On the Importance of Gradient Norm in PAC-Bayesian Bounds. Itai Gat, Yossi Adi, Alex Schwing, Tamir Hazan. Advances in Neural Information Processing Systems
  16. [ PDF, BibTeX ]

    2021

  17. Learning randomly perturbed structured predictors for direct loss minimization. Hedda Indelman, Tamir Hazan. International Conference on Machine Learning
  18. [ PDF, BibTeX ]
  19. Optimizing Memory Placement using Evolutionary Graph Reinforcement Learning. Shauharda Khadka, Estelle Aflalo, Mattias Marder, Avrech Ben-David, Santiago Miret, Shie Mannor, Tamir Hazan, Hanlin Tang, Somdeb Majumdar. International Conference on Learning Representations
  20. [ PDF, BibTeX ]
  21. Visual navigation with spatial attention. Bar Mayo, Tamir Hazan, Ayellet Tal. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition
  22. [ PDF, BibTeX ]
  23. Learning generalized gumbel-max causal mechanisms. Guy Lorberbom, Daniel Johnson, Chris Maddison, Daniel Tarlow, Tamir Hazan. Advances in Neural Information Processing Systems
  24. [ PDF, BibTeX ]
  25. Learning Discrete Structured Variational Auto-Encoder using Natural Evolution Strategies. Alon Berliner, Guy Rotman, Yossi Adi, Roi Reichart, Tamir Hazan. ICLR
  26. [ PDF, BibTeX ]

    2020

  27. Direct policy gradients: Direct optimization of policies in discrete action spaces. Guy Lorberbom, Chris Maddison, Nicolas Heess, Tamir Hazan, Daniel Tarlow. Advances in Neural Information Processing Systems
  28. [ PDF, BibTeX ]
  29. On the generalization of bayesian deep nets for multi-class classification. Yossi Adi, Yaniv Nemcovsky, Alex Schwing, Tamir Hazan. arXiv preprint arXiv:2002.09866
  30. [ PDF, BibTeX ]
  31. Stochastic proximal linear method for structured non-convex problems. Tamir Hazan, Shoham Sabach, Sergey Voldman. Optimization Methods and Software
  32. [ PDF, BibTeX ]
  33. Generalized planning with deep reinforcement learning. Or Rivlin, Tamir Hazan, Erez Karpas. arXiv preprint arXiv:2005.02305
  34. [ PDF, BibTeX ]
  35. Removing bias in multi-modal classifiers: Regularization by maximizing functional entropies. Itai Gat, Idan Schwartz, Alexander Schwing, Tamir Hazan. Advances in Neural Information Processing Systems
  36. [ PDF, BibTeX ]
  37. Equalizing data science curriculum for computer science pupils. Koby Mike, Tamir Hazan, Orit Hazzan. Proceedings of the 20th Koli Calling International Conference on Computing Education Research
  38. [ PDF, BibTeX ]

    2019

  39. Direct Optimization through $$\backslash$arg$\backslash$max $ for Discrete Variational Auto-Encoder. Guy Lorberbom, Andreea Gane, Tommi Jaakkola, Tamir Hazan. Advances in neural information processing systems
  40. [ BibTeX ]
  41. High dimensional inference with random maximum a-posteriori perturbations. Tamir Hazan, Francesco Orabona, Anand Sarwate, Subhransu Maji, Tommi Jaakkola. IEEE Transactions on Information Theory
  42. [ PDF, BibTeX ]
  43. Gravity-based methods for heading computation in pedestrian dead reckoning. Adi Manos, Itzik Klein, Tamir Hazan. Sensors
  44. [ PDF, BibTeX ]
  45. Factor graph attention. Idan Schwartz, Seunghak Yu, Tamir Hazan, Alexander Schwing. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition
  46. [ PDF, BibTeX ]
  47. A simple baseline for audio-visual scene-aware dialog. Idan Schwartz, Alexander Schwing, Tamir Hazan. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition
  48. [ PDF, BibTeX ]
  49. A formal approach to explainability. Lior Wolf, Tomer Galanti, Tamir Hazan. Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society
  50. [ PDF, BibTeX ]
  51. Perturbation based learning for structured NLP tasks with application to dependency parsing. Amichay Doitch, Ram Yazdi, Tamir Hazan, Roi Reichart. Transactions of the Association for Computational Linguistics
  52. [ PDF, BibTeX ]

    2018

  53. Co-segmentation for space-time co-located collections. Hadar Averbuch-Elor, Johannes Kopf, Tamir Hazan, Daniel Cohen-Or. The Visual Computer
  54. [ PDF, BibTeX ]
  55. Hinge-minimax learner for the ensemble of hyperplanes. Dolev Raviv, Tamir Hazan, Margarita Osadchy. The Journal of Machine Learning Research
  56. [ PDF, BibTeX ]
  57. Gravity direction estimation and heading determination for pedestrian navigation. Adi Manos, Itzik Klein, Tamir Hazan. 2018 International Conference on Indoor Positioning and Indoor Navigation (IPIN)
  58. [ PDF, BibTeX ]
  59. LinkedResearch-LR: A Suggested Platform to Make Research Exponential. Rick Reis, Avi Salmon, Avital Binah-Pollak, Yael Dubinsky, Israel StepAhead, Tzahi Harari, Israel Effect-Tiv, Tamir Hazan, Orit Hazzan, Ronit Lis-Hacohen.
  60. [ BibTeX ]

    2017

  61. High-order attention models for visual question answering. Idan Schwartz, Alexander Schwing, Tamir Hazan. Advances in Neural Information Processing Systems
  62. [ PDF, BibTeX ]
  63. Psychological forest: Predicting human behavior. Ori Plonsky, Ido Erev, Tamir Hazan, Moshe Tennenholtz. Proceedings of the AAAI Conference on Artificial Intelligence
  64. [ PDF, BibTeX ]
  65. Tight bounds for bandit combinatorial optimization. Alon Cohen, Tamir Hazan, Tomer Koren. Conference on Learning Theory
  66. [ PDF, BibTeX ]

    2016

  67. Multicuts and perturb \& map for probabilistic graph clustering. J{\"o}rg Kappes, Paul Swoboda, Bogdan Savchynskyy, Tamir Hazan, Christoph Schn{\"o}rr. Journal of Mathematical Imaging and Vision
  68. [ BibTeX ]
  69. A theoretical framework for deep transfer learning. Tomer Galanti, Lior Wolf, Tamir Hazan. Information and Inference: A Journal of the IMA
  70. [ PDF, BibTeX ]
  71. active congruency-Based reranking. Itai Ben, Noga Levy, Lior Wolf, Nachum Dershowitz, Adiel Ben, Roni Shweka, Yaacov Choueka, Tamir Hazan, Yaniv Bar. Frontiers in Digital Humanities
  72. [ PDF, BibTeX ]
  73. Perturbations, optimization, and statistics. Tamir Hazan, George Papandreou, Daniel Tarlow.
  74. [ PDF, BibTeX ]
  75. Online learning with feedback graphs without the graphs. Alon Cohen, Tamir Hazan, Tomer Koren. International Conference on Machine Learning
  76. [ PDF, BibTeX ]
  77. Blending learning and inference in conditional random fields. Tamir Hazan, Alexander Schwing, Raquel Urtasun. The Journal of Machine Learning Research
  78. [ PDF, BibTeX ]
  79. Distributed algorithms for large scale learning and inference in graphical models. Alexander Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun. IEEE transactions on pattern analysis and machine intelligence
  80. [ PDF, BibTeX ]
  81. Multicuts and Perturb \& MAP for Probabilistic Graph Clustering. J{\"o}rg Hendrik, Paul Swoboda, Bogdan Savchynskyy, Tamir Hazan, Christoph Schn{\"o}rr. arXiv e-prints
  82. [ BibTeX ]
  83. Constraints Based Convex Belief Propagation. Alex Schwing, Kevin Gimpel, Tamir Hazan. Advances in Neural Information Processing Systems
  84. [ PDF, BibTeX ]
  85. Constraints based convex belief propagation. Yaniv Tenzer, Alexander Schwing, Kevin Gimpel, Tamir Hazan. Proceedings of the 30th International Conference on Neural Information Processing Systems
  86. [ PDF, BibTeX ]

    2015

  87. Probabilistic correlation clustering and image partitioning using perturbed multicuts. J{\"o}rg Kappes, Paul Swoboda, Bogdan Savchynskyy, Tamir Hazan, Christoph Schn{\"o}rr. Scale Space and Variational Methods in Computer Vision: 5th International Conference, SSVM 2015, L{\`e}ge-Cap Ferret, France, May 31-June 4, 2015, Proceedings 5
  88. [ PDF, BibTeX ]
  89. K-hyperplane hinge-minimax classifier. Margarita Osadchy, Tamir Hazan, Daniel Keren. International Conference on Machine Learning
  90. [ PDF, BibTeX ]
  91. Following the perturbed leader for online structured learning. Alon Cohen, Tamir Hazan. International Conference on Machine Learning
  92. [ PDF, BibTeX ]
  93. Probabilistic Correlation Clustering and Image Partitioning Using Perturbed Multicuts. Tamir Hazan, Christoph Schn{\"o}rr. Scale Space and Variational Methods in Computer Vision: 5th International Conference, SSVM 2015, L{\`e}ge-Cap Ferret, France, May 31-June 4, 2015, Proceedings
  94. [ PDF, BibTeX ]
  95. Steps toward deep kernel methods from infinite neural networks. Tamir Hazan, Tommi Jaakkola. arXiv preprint arXiv:1508.05133
  96. [ PDF, BibTeX ]
  97. Efficient training of structured svms via soft constraints. Ofer Meshi, Nathan Srebro, Tamir Hazan. Artificial Intelligence and Statistics
  98. [ PDF, BibTeX ]

    2014

  99. On measure concentration of random maximum a-posteriori perturbations. Francesco Orabona, Tamir Hazan, Anand Sarwate, Tommi Jaakkola. International Conference on Machine Learning
  100. [ PDF, BibTeX ]
  101. Congruency-based reranking. Itai Ben-Shalom, Noga Levy, Lior Wolf, Nachum Dershowitz, Adiel Ben-Shalom, Roni Shweka, Yaacov Choueka, Tamir Hazan, Yaniv Bar. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition
  102. [ PDF, BibTeX ]
  103. Learning with maximum a-posteriori perturbation models. Andreea Gane, Tamir Hazan, Tommi Jaakkola. Artificial Intelligence and Statistics
  104. [ PDF, BibTeX ]
  105. Computational education using latent structured prediction. Tanja K{\"a}ser, Alexander Schwing, Tamir Hazan, Markus Gross. Artificial Intelligence and Statistics
  106. [ PDF, BibTeX ]
  107. Active boundary annotation using random map perturbations. Subhransu Maji, Tamir Hazan, Tommi Jaakkola. Artificial Intelligence and Statistics
  108. [ PDF, BibTeX ]
  109. Globally convergent parallel MAP LP relaxation solver using the Frank-Wolfe algorithm. Alexander Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun. International Conference on Machine Learning
  110. [ PDF, BibTeX ]

    2013

  111. On sampling from the gibbs distribution with random maximum a-posteriori perturbations. Tamir Hazan, Subhransu Maji, Tommi Jaakkola. Advances in Neural Information Processing Systems
  112. [ PDF, BibTeX ]
  113. Learning efficient random maximum a-posteriori predictors with non-decomposable loss functions. Tamir Hazan, Subhransu Maji, Joseph Keshet, Tommi Jaakkola. Advances in Neural Information Processing Systems
  114. [ PDF, BibTeX ]

    2012

  115. Continuous markov random fields for robust stereo estimation. Koichiro Yamaguchi, Tamir Hazan, David McAllester, Raquel Urtasun. Computer Vision--ECCV 2012: 12th European Conference on Computer Vision, Florence, Italy, October 7-13, 2012, Proceedings, Part V 12
  116. [ PDF, BibTeX ]
  117. Efficient structured prediction with latent variables for general graphical models. Alexander Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun. Proceedings of the 29th International Conference on Machine Learning
  118. [ PDF, BibTeX ]
  119. On the partition function and random maximum a-posteriori perturbations. Tamir Hazan, Tommi Jaakkola. Proceedings of the 29th International Conference on Machine Learning
  120. [ PDF, BibTeX ]
  121. Globally convergent dual MAP LP relaxation solvers using Fenchel-Young margins. Alex Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun. Advances in Neural Information Processing Systems
  122. [ PDF, BibTeX ]
  123. Tightening fractional covering upper bounds on the partition function for high-order region graphs. Tamir Hazan, Jian Peng, Amnon Shashua. The 28th Conference on Uncertainty in Artificial Intelligence
  124. [ PDF, BibTeX ]
  125. Approximate inference by intersecting semidefinite bound and local polytope. Jian Peng, Tamir Hazan, Nathan Srebro, Jinbo Xu. Artificial Intelligence and Statistics
  126. [ PDF, BibTeX ]
  127. Distributed structured prediction for big data. Alexander Schwing, T Hazan, M Pollefeys, R Urtasun. NIPS workshop on big learning
  128. [ BibTeX ]

    2011

  129. Distributed message passing for large scale graphical models. Alexander Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun. CVPR 2011
  130. [ PDF, BibTeX ]
  131. Large Scale Structured Prediction with Hidden Variables. Alexander Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun. The Learning Workshop
  132. [ BibTeX ]
  133. Pac-bayesian approach for minimization of phoneme error rate. Joseph Keshet, David McAllester, Tamir Hazan. 2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
  134. [ PDF, BibTeX ]
  135. Convex max-product over compact sets for protein folding. Jian Peng, Tamir Hazan, Raquel Urtasun, David Mcallester. Proceedings of the 28th International Conference on Machine Learning (ICML-11)
  136. [ PDF, BibTeX ]

    2010

  137. Norm-product belief propagation: Primal-dual message-passing for approximate inference. Tamir Hazan, Amnon Shashua. IEEE Transactions on Information Theory
  138. [ PDF, BibTeX ]
  139. A primal-dual message-passing algorithm for approximated large scale structured prediction. Tamir Hazan, Raquel Urtasun. Advances in neural information processing systems
  140. [ PDF, BibTeX ]
  141. Direct loss minimization for structured prediction. Tamir Hazan, Joseph Keshet, David McAllester. Advances in neural information processing systems
  142. [ PDF, BibTeX ]

    2009

  143. Algebraic Methods for Learning in Computer Vision. Tamir Hazan.
  144. [ BibTeX ]

    2008

  145. Convergent message-passing algorithms for inference over general graphs with convex free energies. Tamir Hazan, Amnon Shashua. The 24th Conference on Uncertainty in Artificial Intelligence
  146. [ PDF, BibTeX ]
  147. A parallel decomposition solver for svm: Distributed dual ascend using fenchel duality. Tamir Hazan, Amit Man, Amnon Shashua. 2008 IEEE Conference on Computer Vision and Pattern Recognition
  148. [ PDF, BibTeX ]

    2007

  149. Modeling appearances with low-rank svm. Lior Wolf, Hueihan Jhuang, Tamir Hazan. 2007 IEEE Conference on Computer Vision and Pattern Recognition
  150. [ PDF, BibTeX ]
  151. Plsa for sparse arrays with Tsallis pseudo-additive divergence: noise robustness and algorithm. Tamir Hazan, Roee Hardoon, Amnon Shashua. 2007 IEEE 11th International Conference on Computer Vision
  152. [ PDF, BibTeX ]
  153. Analysis of l2-loss for probabilistically valid factorizations under general additive noise. Tamir Hazan, Amnon Shashua. Technical Report 2007-13, The Hebrew University
  154. [ PDF, BibTeX ]
  155. An efficient algorithm for maximum Tsallis entropy using Fenchel-duality. Tamir Hazan, Amnon Shashua.
  156. [ PDF, BibTeX ]

    2006

  157. Multi-way clustering using super-symmetric non-negative tensor factorization. Amnon Shashua, Ron Zass, Tamir Hazan. Computer Vision--ECCV 2006: 9th European Conference on Computer Vision, Graz, Austria, May 7-13, 2006, Proceedings, Part IV 9
  158. [ PDF, BibTeX ]

    2005

  159. Non-negative tensor factorization with applications to statistics and computer vision. Amnon Shashua, Tamir Hazan. Proceedings of the 22nd international conference on Machine learning
  160. [ PDF, BibTeX ]
  161. Sparse image coding using a 3D non-negative tensor factorization. Tamir Hazan, Simon Polak, Amnon Shashua. Tenth IEEE International Conference on Computer Vision (ICCV'05) Volume 1
  162. [ PDF, BibTeX ]

    2004

  163. Algebraic set kernels with application to inference over local image representations. Amnon Shashua, Tamir Hazan. Advances in neural information processing systems
  164. [ PDF, BibTeX ]