Machine-learning-nederland April 2023

machine-learning-nederland@list.uva.nl

2 participants
5 discussions

PhD positions on AI-Based Pedestrian Crowd Modeling and Management
by Pires da Silva Castro, Rui 27 Apr '23

27 Apr '23

Dear Colleagues, There are two openings for PhD positions at the Eindhoven University of Technology, part of the recently funded AICrowd project: AI-Based Pedestrian Crowd Modeling and Management. I would appreciate it if you could forward this announcement to any potential candidates. Are you eager to work on a pioneering PhD project at the interface between physics of flowing matter, artificial intelligence, system identification, and statistics? Do you enjoy collaborating with researchers from different fields, and combining dynamical systems modeling, optimization and statistical learning theory? Are you eager to see your work making immediate societal impact? Then one of these positions might be for you. AICrowd project - AI-Based Pedestrian Crowd Modelling and Management vacancies: - AI & System identification for crowd flow modeling: https://jobs.tue.nl/en/vacancy/phd-in-artificial-intelligence-for-quantitat… - Statistics and AI for quantitative crowd dynamics modeling: https://jobs.tue.nl/en/vacancy/phd-in-statistics-and-ai-for-quantitative-cr… With best regards, Rui Castro — Dept. of Mathematics and Computer Science, TU/e http://www.win.tue.nl/~rmcastro | +31 40 247 2499

1 0

May 10, 15:00 Machine Learning Theory Seminar++ Meeting with Rianne de Heide
by Wouter Koolen-Wijkstra 21 Apr '23

21 Apr '23

Dear researchers, Apologies for re-sending, this is to test a config update to prevent future email bounces. Centrum Wiskunde & Informatica (CWI) kindly invites you to the fifth Seminar++ meeting on Machine Learning Theory, taking place on Wednesday May 10 from 15:00 - 17:00. These Seminar++ meetings consist of a one-hour lecture building up to an open problem, followed by an hour of brainstorming time. The meeting is intended for interested researchers including PhD students. These meetings are freely accessible without registration. Cookies, coffee and tea will be provided in the half-time break. The meeting of 10 May will be hosted by: Rianne de Heide <https://riannedeheide.github.io/> (Assistant Professor at the Vrije Universiteit Amsterdam <https://vu.nl/>) *Multiple testing with e-values: overview and open problems* Abstract: Researchers in genomics and neuroimaging often perform hundreds of thousands of hypothesis tests simultaneously. The scale of these multiple hypothesis testing problems is enormous, and with extreme dimensionality comes extreme risk for false positives. The field of multiple testing addresses this problem in various ways. Recently, the new theory of hypothesis testing with e-values has been brought to the field of multiple testing. In this talk I will give an overview of the most important frameworks in multiple testing and recent developments in multiple testing with e-values. Finally, I will open the discussion for open problems in this area, focusing on FDP estimation and confidence with e-values. This will create a framework for fully interactive multiple testing. The event takes place in room L016 in the CWI building, Science Park 123, Amsterdam. The Seminar++ Meetings are part of the Machine Learning Theory Semester Programme <https://www.cwi.nl/en/events/cwi-research-semester-programs/research-progra…>, which runs in Spring 2023. Best regards on behalf of CWI from the program committee, Wouter Koolen

1 0

May 10, 15:00 Machine Learning Theory Seminar++ Meeting with Rianne de Heide
by Wouter Koolen-Wijkstra 20 Apr '23

20 Apr '23

Dear researchers, Centrum Wiskunde & Informatica (CWI) kindly invites you to the fifth Seminar++ meeting on Machine Learning Theory, taking place on Wednesday May 10 from 15:00 - 17:00. These Seminar++ meetings consist of a one-hour lecture building up to an open problem, followed by an hour of brainstorming time. The meeting is intended for interested researchers including PhD students. These meetings are freely accessible without registration. Cookies, coffee and tea will be provided in the half-time break. The meeting of 10 May will be hosted by: Rianne de Heide <https://riannedeheide.github.io/> (Assistant Professor at the Vrije Universiteit Amsterdam <https://vu.nl/>) *Multiple testing with e-values: overview and open problems* Abstract: Researchers in genomics and neuroimaging often perform hundreds of thousands of hypothesis tests simultaneously. The scale of these multiple hypothesis testing problems is enormous, and with extreme dimensionality comes extreme risk for false positives. The field of multiple testing addresses this problem in various ways. Recently, the new theory of hypothesis testing with e-values has been brought to the field of multiple testing. In this talk I will give an overview of the most important frameworks in multiple testing and recent developments in multiple testing with e-values. Finally, I will open the discussion for open problems in this area, focusing on FDP estimation and confidence with e-values. This will create a framework for fully interactive multiple testing. The event takes place in room L016 in the CWI building, Science Park 123, Amsterdam. The Seminar++ Meetings are part of the Machine Learning Theory Semester Programme <https://www.cwi.nl/en/events/cwi-research-semester-programs/research-progra…>, which runs in Spring 2023. Best regards on behalf of CWI from the program committee, Wouter Koolen

1 0

April 19, 15:00: Machine Learning Theory Seminar++ Meeting with Dirk van der Hoeven
by Wouter Koolen-Wijkstra 14 Apr '23

14 Apr '23

Dear researchers, Centrum Wiskunde & Informatica (CWI) kindly invites you to the fourth Seminar++ meeting on Machine Learning Theory, taking place on Wednesday April 19 from 15:00 - 17:00. These Seminar++ meetings consist of a one-hour lecture building up to an open problem, followed by an hour of brainstorming time. The meeting is intended for interested researchers including PhD students. These meetings are freely accessible without registration. Cookies, coffee and tea will be provided in the half-time break. The meeting of 19 April will be hosted by: Dirk van der Hoeven <http://dirkvanderhoeven.com/about> (Postdoc at the University of Amsterdam <https://www.uva.nl/>) High-Probability Risk Bounds via Sequential Predictors *Abstract:* Online learning methods yield sequential regret bounds under minimal assumptions and provide in-expectation results in statistical learning. However, despite the seeming advantage of online guarantees over their statistical counterparts, recent findings indicate that in many important cases, regret bounds may not guarantee high probability risk bounds. In this work we show that online to batch conversions applied to general online learning algorithms are more powerful than previously thought. Via a new second-order correction to the loss function, we obtain sharp high-probability risk bounds for many classical statistical problems, such as model selection aggregation, linear regression, logistic regression, and—more generally—conditional density estimation. Our analysis relies on the fact that many online learning algorithms are improper, as they are not restricted to use predictors from a given reference class. The improper nature enables significant improvements in the dependencies on various problem parameters. In the context of statistical aggregation of finite families, we provide a simple estimator achieving the optimal rate of aggregation in the model selection aggregation setup with general exp-concave losses. *First open problem:* This open problem will be the main focus of the seminar. The result for logistic regression is nearly optimal and the algorithm is computationally efficient in the sense that the runtime is polynomial in the relevant problem parameters. However, it is a polynomial with a very high degree, making the algorithm practically not very useful for reasonable problem parameters. For in expectation guarantees it is known how to reduce the runtime to /d^2 T/ at the cost of a slightly worse excess risk bound, where /d/ is the dimension of the problem and /T/ is the number of datapoints. Unfortunately it is not immediately clear how to use the ideas from the faster algorithm to obtain a high-probability excess risk bound with a /d^2 T/ runtime algorithm. This open problem asks for the following: can we obtain a reasonable excess risk bound with high-probability in /d^2 T/ runtime for logistic regression. *Second open problem:* Our results heavily rely on a particular inequality for exp-concave losses. I would like to extend our ideas to a different class of loss function, namely self-concordant loss functions. Given previous results in statistical learning literature (see https://arxiv.org/pdf/2105.08866.pdf), I expect this to be possible. The event takes place in room L016 in the CWI building, Science Park 123, Amsterdam. The Seminar++ Meetings are part of the Machine Learning Theory Semester Programme <https://www.cwi.nl/~wmkoolen/MLT_Sem23/index.html>, which runs in Spring 2023. Best regards on behalf of CWI from the program committee, Wouter Koolen

1 0

April 12 Afternoon: Lectures on Machine Learning Theory by Peters and Lattimore
by Wouter Koolen-Wijkstra 07 Apr '23

07 Apr '23

Dear colleagues, Centrum Wiskunde & Informatica (CWI) kindly invites you to a lecture afternoon focused on theory of machine learning, with two distinguished speakers that will focus on philosophical and technical challenges in learning to model and explore the unknown environment. The program on 12 April: 14.00 Jonas Peters <http://web.math.ku.dk/~peters/> (University of Copenhagen): /Recent Advances in Causal Inference/ 15.00 Break 15.30 Tor Lattimore <https://tor-lattimore.com/> (DeepMind): /The Curse and Blessing of Curvature for Zeroth-order Convex Optimisation/ 16.30 Reception The first talk will be delivered remotely, the second live. For further details, please check the abstracts below and the website <https://www.cwi.nl/en/events/cwi-research-semester-programs/mid-semester-le…>. These research-level lectures are intended for a non-specialist audience, and are freely accessible. The event takes place in the Amsterdam Science Park Congress Centre. This "Mid Semester Lecture", marks the half-way point of the Machine Learning Theory Semester Programme <https://www.cwi.nl/en/events/cwi-research-semester-programs/research-progra…>, which runs in Spring 2023. Researchers in Machine Learning Theory may also be interested in the ongoing Seminar++ meetings <https://www.cwi.nl/en/events/cwi-research-semester-programs/seminar-part-4-…> every other Wednesday. Best regards on behalf of CWI from the program committee, Wouter Koolen Tor Lattimore /The Curse and Blessing of Curvature for Zeroth-order Convex Optimisation/ *Abstract:* Zeroth-order convex optimisation is still quite poorly understood. I will tell a story about how to use gradient-based methods without access to gradients and explain how curvature of the loss function plays the roles of both devil and angel. The main result is a near-optimal sample-complexity analysis of a simple and computationally efficient second-order method applied to a quadratic surrogate loss. The talk is based on a recent paper <https://arxiv.org/abs/2302.05371> with András György. Jonas Peters/ Recent Advances in Causal Inference/ *Abstract:* see website <https://www.cwi.nl/en/events/cwi-research-semester-programs/mid-semester-le…>

1 0

2024

2023

2022

2021

2020

Machine-learning-nederland April 2023