Research
We focus on the design and analysis of machine learning and optimization algorithms. The two topics are very related: Machine learning is just a form of stochastic optimization. In particular, we focus on online learning, empirical processes, and stochastic algorithms for convex and non-convex problems.
Job Openings
I am looking for 1-2 post-docs and 1-2 PhD students to work on practical and theoretical aspects of online learning, stochastic optimization, and training of LLMs. The ideal candidate has an exceptional mathematical background and is proficient in coding. If you are interested, send me an email with your CV and your transcript too if your are applying for a PhD you. Due to the volume of emails and my limited time, I might not answer to everyone: do not take it personally, you were probably not a good match for my lab.
News
5/8/24 One paper accepted at COLT!
1/17/24 One paper accepted at ICLR!
10/22/23 One paper accepted at IEEE Transactions on Information Theory!
06/22/23 One paper accepted at the ICML workshop on Duality Principles for Modern Machine Learning
06/15/23 One paper accepted at the ICML workshop on PAC-Bayes Meets Interactive Learning
06/09/23 We moved to KAUST!
05/14/23 One COLT paper accepted!
04/25/23 Two ICML papers accepted!
09/14/22 One NeurIPS paper accepted!
08/08/22 One TMLR paper accepted!
01/09/22 Three ALT'22 papers accepted!
12/01/21 One AAAI'22 paper accepted!
09/28/21 One NeurIPS'21 paper accepted!
05/08/21 Two ICML'21 papers accepted!
03/01/21 One paper published at the Annual Review of Statistics and Its Application journal
02/02/21 Francesco Orabona has been awarded the NSF CAREER
09/25/20 One NeurIPS'20 paper accepted!
06/27/20 One ICML'20 Workshop on Beyond First Order Methods in ML Systems paper accepted!
05/19/20 Francesco Orabona and Ashok Cutkosky will give a tutorial at ICML'20 on Parameter-free Online Optimization
01/24/20 One ICASSP'20 paper accepted!
10/02/19 One NeurIPS'19 Workshop on Meta-Learning paper accepted!
10/01/19 One NeurIPS'19 Workshop on Privacy in Machine Learning paper accepted!
09/03/19 Two NeurIPS'19 papers accepted!
07/27/19 NSF founded our project in collaboration with Omkant Pandey
04/21/19 One ICML'19 paper accepted!
04/18/19 One COLT'19 paper accepted!
12/22/18 One AISTATS'19 paper accepted!
09/01/18 We moved to Boston University
12/19/17 The website is up!
People
Director of the lab, Associate Professor
Margarita Akhmejanova
Post-doc
ML Engineer
(co-supervisioned with Jürgen Schmidhuber)
4th year CS PhD Student (at Boston University)
Jeren Koh
Visiting Student
Alumni
Visiting PhD Student
CS PhD Student, graduated in 2022, now Research Scientist at Meta
SE PhD Student, graduated in 2022, now Data Scientist at Microsoft
Post-doc 2018-2019, now Assistant Professor at the University of Arizona
Post-doc 2020-2021, now Assistant Professor at George Mason University
CS MS Student, now ?
Visiting PhD Student 2020, now ?
Preprints
F. Orabona. A Modern Introduction to Online Learning. arXiv [PDF]
Papers
I. Kuzborskij, K.-S. Jun, Y. Wu, K. Jang, F. Orabona. Better-than-KL PAC-Bayes Bounds. COLT 2024
A. Meterez, A. Joudaki, F. Orabona, A. Immer, G. Rätsch, and H. Daneshmand. Towards Training Without Depth Limits: Batch Normalization Without Gradient Explosion. ICLR 2024
F. Orabona, K.-S. Jun. Tight Concentrations and Confidence Sequences from the Regret of Universal Portfolio. IEEE Transactions on Information Theory, 2024
A. Cutkosky, H. Mehta, F. Orabona. Optimal Stochastic Non-smooth Non-convex Optimization through Online-to-Non-convex Conversion. ICML'23
K. Chen, F. Orabona. Generalized Implicit Follow-The-Regularized-Leader. ICML'23
K. Chen, F. Orabona. Implicit Interpretation of Importance Weight Aware Updates. Duality Principles for Modern Machine Learning Workshop at ICML'23
K. Jang, K.-S. Jun, I. Kuzborskij, F. Orabona. Improved Time-Uniform PAC-Bayes Bounds using Coin Betting. PAC-Bayes Meets Interactive Learning Workshop at ICML'23
K. Jang, K.-S. Jun, I. Kuzborskij, F. Orabona. Tighter PAC-Bayes Bounds Through Coin-Betting. COLT'23
M. Crawshaw, M. Liu, F. Orabona, W. Zhang, and Z. Zhuang. Robustness to Unbounded Smoothness of Generalized SignSGD. NeurIPS'22 [PDF] [CODE]
Z. Zhuang, M. Liu, A. Cutkosky, and F. Orabona. Understanding AdamW through Proximal Methods and Scale-Freeness. Transactions on Machine Learning Research, Aug. 2022. [PDF] [CODE]
K. Chen, A. Cutkosky, and F. Orabona. Implicit Parameter-free Online Learning with Truncated Linear Models. ALT 2022 [PDF]
X. Li, M. Liu, and F. Orabona. On the Last Iterate Convergence of Momentum Methods. ALT 2022 [PDF]
M. Liu and F. Orabona. On the Initialization for Convex-Concave Min-max Problems. ALT 2022 [PDF]
K. Chen, J. Langford, and F. Orabona. Better Parameter-free Stochastic Optimization with ODE Updates for Coin-Betting. AAAI 2022 [PDF]
J. Negrea, B. Bilodeau, N. Campolongo, F. Orabona, and Daniel M. Roy. Minimax Optimal Quantile and Semi-Adversarial Regret via Root-Logarithmic Regularizers. NeurIPS 2021 [PDF]
X. Li, Z. Zhuang, and F. Orabona. A Second look at Exponential and Cosine Step Sizes: Simplicity, Adaptivity, and Performance. ICML 2021 [PDF] [CODE]
G. Flaspohler, F. Oraboba, J. Cohen, S. Mouatadid, M. Oprescu, P. Orenstein, and L. Mackey. Online Learning with Optimism and Delay. ICML 2021
N. Cesa-Bianchi and F. Orabona. Online Learning Algorithms. Annual Review of Statistics and Its Application. 2021 [PDF]
N. Campolongo and F. Orabona. Temporal Variability in Implicit Online Learning. NeurIPS 2020 [PDF]
X. Li and F. Orabona. A High Probability Analysis of Adaptive SGD with Momentum. ICML 2020 Workshop on Beyond First Order Methods in ML Systems. [PDF]
Z. Zhuang, Y. Wang, K. Yu, and S. Lu. No-regret Non-convex Online Meta-Learning. ICASSP 2020 [PDF]
K.-S. Jun and F. Orabona. Parameter-Free Locally Differentially Private Stochastic Subgradient Descent. NeurIPS 2019 Workshop on Privacy in Machine Learning [PDF]
Z. Zhuang, K. Yu, S. Lu, L. Glass, and Y. Wang. Online Meta-Learning on Non-convex Setting. NeurIPS 2019 Workshop on Meta-Learning [PDF]
A. Cutkosky and F. Orabona. Momentum-Based Variance Reduction in Non-Convex SGD. NeurIPS 2019 [PDF]
K.-S. Jun, A. Cutkosky, F. Orabona. Kernel Truncated Randomized Ridge Regression: Optimal Rates and Low Noise Acceleration. NeurIPS 2019 [PDF]
K.-S. Jun and F. Orabona. Parameter-free Online Convex Optimization with Sub-Exponential Noise. COLT 2019 [PDF]
Z. Zhuang, A. Cutkosky, F. Orabona. Surrogate Losses for Online Learning of Stepsizes in Stochastic Non-Convex Optimization. ICML 2019 [PDF] [CODE]
X. Li and F. Orabona. On the Convergence of Stochastic Gradient Descent with Adaptive Stepsizes. AISTATS 2019 [PDF]
A. Cutkosky and F. Orabona. Black-Box Reductions for Parameter-free Online Learning in Banach Spaces. COLT 2018 [PDF]
F. Orabona and D. Pal. Scale-free Online Learning, Theoretical Computer Science, 716. 2018 [PDF]
F. Orabona and T. Tommasi. Training Deep Networks without Learning Rates Through Coin Betting, NeurIPS 2017 [PDF] [CODE]
Sponsors
Our research has been founded by