Machine-learning-nederland November 2020

machine-learning-nederland@list.uva.nl

3 participants
3 discussions

CWI Machine Learning seminar: Christian Hennig Fri 27 Nov, 15:00, *Updated Zoom Link*
by Wouter Koolen-Wijkstra 26 Nov '20

26 Nov '20

Dear all, This is a reminder of tomorrow's CWI Machine Learning seminar, with a *new and improved zoom URL*. Speaker: Christian Hennig (University of Bologna) Title: A spotlight on statistical model assumptions Date: Friday 27 November, 15:00 Location: https://cwi-nl.zoom.us/j/87252095652?pwd=aXI0K1VUdlNlbGlReEE3WGMyWXd6QT09 Please find the abstract below. Hope to see you then. Best wishes, Wouter Details: https://portals.project.cwi.nl/ml-reading-group/events/a-spotlight-on-stati… ============ A spotlight on statistical model assumptions Christian Hennig (University of Bologna) Many statistics teachers tell their students something like "In order to apply the t-test, we have to assume that the data are i.i.d. normally distributed, and therefore these model assumptions need to be checked before applying the t-test." This statement is highly problematic in several respects. There is no good reason to believe that any real data truly are drawn i.i.d. normally. Furthermore, quite relevant aspects of these model assumptions cannot be checked. For example, I will show that data generated from a normal distribution with a correlation of $\rho\neq 0$ between any two observations cannot be distinguished from i.i.d. normal data. On top of this, passing a model by a model checking test will automatically invalidate it; much literature investigating the performance of specific procedures that run model-based tests conditionally on passing a model misspecification test comment very critically on this practice. Despite all these issues, I will defend interpreting and using statistical models in a frequentist manner, by advocating an understanding of models that never forgets that models are essentially different from reality (and in this sense can never be "true"). Model assumptions specify idealised conditions under which methods work well; in reality they do not need to be fulfilled. However, situations in which the data will mislead a method need to be distinguished from situations in which a method does what it is expected to do. This defines a more appropriate task for model checking. Conditions are required for doing this job properly that some model checking currently in use does not fulfill. For better "model checking" it will be helpful to understand that this is not about "finding out whether the model assumptions hold", but about something quite different.

1 0

3 year postdoc at Delft University of Technology
by Frans Oliehoek 17 Nov '20

17 Nov '20

(apologies for cross-posting) We are pleased to announce that the Department of Intelligent Systems at TU Delft, The Netherlands, can offer a 3-year postdoc position, as part of the "Hybrid Intelligence" project, www.hybrid-intelligence-centre.nl. Closing date: January 8th -------------------------------------------------- Postdoc in (Meta-)Learning to Give Feedback in Interactive Learning (3 years) How can an intelligent learn to interact? How can it learn via interaction? For this project we are looking for a postdoc who wants to push machine learning beyond traditional settings that assume a fixed dataset. Specifically, in this project we will investigate interactive learning settings in which two or more learners interact by giving each other feedback to reach an outcome that is desirable from a system designers perspective. The goal is to better understand how to structure interactions to effectively progress to the desirable outcome state, and to develop practical learning techniques and algorithms that exploit these generated insights. The postdoc will be based at TU Delft and co-supervised by Herke van Hoof (University of Amsterdam) and myself. Given that the successful candidate will have to work with 2 supervisors at different institutions, we are looking for someone who can operate quite independently. Full requirements and application instructions: https://www.academictransfer.com/en/295565/postdoc-meta-learning-to-give-fe… More information: For more information, please see: https://www.fransoliehoek.net/wp/vacancies/ Informal inquiries are welcome and can be directed to myself: Dr. Frans Oliehoek <f.a.oliehoek(a)tudelft.nl>. -------------------------------------------------- I would be grateful if you could forward this message to suitable candidates. Best regards, -Frans Oliehoek

1 0

Fwd: Next SPIP talk: Dirk van der Hoeven
by Tim van Erven 02 Nov '20

02 Nov '20

Dear all, I'm forwarding an announcement of an online talk at the UvA by my PhD student Dirk van der Hoeven that might be of more general interest. Best regards, Tim -------- Forwarded Message -------- Subject: Next SPIP talk: Dirk van der Hoeven Date: Fri, 30 Oct 2020 19:00:47 +0100 From: SPIP Meetings <spip.meetings(a)gmail.com> To: [...] Dear All, We are happy to invite you to our next SPIP talk on *Friday, 6th November* from *16:00-17:00*. Our speaker is Dirk van der Hoeven and he will talk about ‘_Exploiting the Surrogate Gap in Online Multiclass Classification_/’//./ Dirk is a PhD student at Leiden University with Tim van Erven and he will join Nicolò Cesa-Bianchis group as a postdoc in December. _Zoom Details:_ _ _ Topic: SPIP - Dirk van der Hoeven Time: Nov 6, 2020 04:00 PM Amsterdam, Berlin, Rome, Stockholm, Vienna Join Zoom Meeting https://uva-live.zoom.us/j/82915951912 <https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fuva-live.…> Meeting ID: 829 1595 1912 / / / / /_Abstract:_/ We present Gaptron, a randomized first-order algorithm for online multiclass classification. In the full information setting we show expected mistake bounds with respect to the logistic loss, hinge loss, and the smooth hinge loss with constant regret, where the expectation is with respect to the learner's randomness. In the bandit classification setting we show that Gaptron is the first linear time algorithm with O(K sqrt(T)) expected regret, where K is the number of classes. Additionally, the expected mistake bound of Gaptron does not depend on the dimension of the feature vector, contrary to previous algorithms with O(K sqrt(T) ) regret in the bandit classification setting. We present a new proof technique that exploits the gap between the zero-one loss and surrogate losses rather than exploiting properties such as exp-concavity or mixability, which are traditionally used to prove logarithmic or constant regret bounds. The Organizer-team -- Tim van Erven <tim(a)timvanerven.nl> www.timvanerven.nl

1 0

2024

2023

2022

2021

2020

Machine-learning-nederland November 2020