- Machine-learning-nederland

Reminder: Thematic seminar: Damien Garreau Thu Nov 10, and Umut Simsekli Mon Nov 14
by Tim van Erven 09 Nov '22

09 Nov '22

Dear all, This is to remind you about the two in person talks tomorrow and on Monday in the thematic seminar at the UvA: 1. On Thursday November 10 Damien Garreau from the Université Côte d'Azur will speak about his analysis of the popular LIME method for explainable machine learning. 2. On Monday November 14, Umut Şimşekli from INRIA/École Normale Supérieure will speak about his new generalization bounds for deep neural networks. *1. Damien Garreau *(Université Côte d'Azur, https://sites.google.com/view/damien-garreau/home) *Thursday November 10*, 16h00-17h00 In person, at the University of Amsterdam Location: Science Park 904, Room B0.204 * **What does LIME really see in images? *The performance of modern algorithms on certain computer vision tasks such as object recognition is now close to that of humans. This success was achieved at the price of complicated architectures depending on millions of parameters and it has become quite challenging to understand how particular predictions are made. Interpretability methods propose to give us this understanding. In this talk, I will present a recent result about LIME, perhaps one of the most popular methods. * * *2. Umut Şimşekli* (INRIA/École Normale Supérieure, https://www.di.ens.fr/~simsekli/) *Monday November 14*, 16h00-17h00 In person, at the University of Amsterdam Location: Science Park 904, Room A1.24 *Fractal Structure and Generalization Properties of Stochastic Optimization Algorithms* Understanding generalization in deep learning has been one of the major challenges in statistical learning theory over the last decade. While recent work has illustrated that the dataset and the training algorithm must be taken into account in order to obtain meaningful generalization bounds, it is still theoretically not clear which properties of the data and the algorithm determine the generalization performance. In this talk, I will approach this problem from a dynamical systems theory perspective and represent stochastic optimization algorithms as random iterated function systems (IFS). Well studied in the dynamical systems literature, under mild assumptions, such IFSs can be shown to be ergodic with an invariant measure that is often supported on sets with a fractal structure. We will prove that the generalization error of a stochastic optimization algorithm can be bounded based on the ‘complexity’ of the fractal structure that underlies its invariant measure. Leveraging results from dynamical systems theory, we will show that the generalization error can be explicitly linked to the choice of the algorithm (e.g., stochastic gradient descent – SGD), algorithm hyperparameters (e.g., step-size, batch-size), and the geometry of the problem (e.g., Hessian of the loss). We will further specialize our results to specific problems (e.g., linear/logistic regression, one hidden-layered neural networks) and algorithms (e.g., SGD and preconditioned variants), and obtain analytical estimates for our bound. For modern neural networks, we will develop an efficient algorithm to compute the developed bound and support our theory with various experiments on neural networks. The talk is based on the following publication: Camuto, A., Deligiannidis, G., Erdogdu, M. A., Gurbuzbalaban, M., Simsekli, U., & Zhu, L. (2021). Fractal structure and generalization properties of stochastic optimization algorithms. Advances in Neural Information Processing Systems, 34, 18774-18788. Seminar organizers: Tim van Erven Botond Szabo https://mschauer.github.io/StructuresSeminar/ -- Tim van Erven<tim(a)timvanerven.nl> www.timvanerven.nl

1 0

Tenure track researcher in reinforcement learning at AI Department of the Donders Institute
by Marcel van Gerven 04 Nov '22

04 Nov '22

The AI Department of the Donders Centre for Cognition (DCC), embedded in the Donders Institute for Brain, Cognition and Behaviour, is looking for a tenure track researcher in reinforcement learning with an interest towards applications in neurotechnology. You will be part of the Dutch DBI2 consortium, which is a multi-year/multi-institute project aimed at understanding brain function from a computational and mechanistic perspective (dbi2.nl <http://dbi2.nl/>). You are a researcher in reinforcement learning with an emphasis on safety and robustness. You have a keen interest in applications in neurotechnology and other domains such as robotics, healthcare and/or sustainability. You will perform top-quality research in reinforcement learning, actively contribute to and support the DBI2 consortium, interact and collaborate with other researchers and specialists in academia and/or industry, and be an inspiring member of our staff with excellent communication skills. Furthermore, you will engage with students through teaching and master projects not exceeding 20% of your time. This tenure track position will lead to a tenured assistant or associate professorship after three years (depending on experience and performance). The Donders Institute provides excellent facilities such as computing facilities, a robot lab, a virtual reality lab, behavioural labs, and a technical support group. The AI Department is also a founding member of Radboud AI and the ELLIS Unit Nijmegen (European Excellence Network in Machine Learning). You will join the Artificial Cognitive Systems (ACS) group headed by Prof. van Gerven and interact closely with other machine learning researchers and computational neuroscientists. If you are interested, please check out https://ru.nl/en/working-at/job-opportunities/researcher-in-reinforcement-l… <https://ru.nl/en/working-at/job-opportunities/researcher-in-reinforcement-l…> and feel encouraged to apply (deadline November 23).

1 0

Thematic seminar: Damien Garreau Thu Nov 10, and Umut Simsekli Mon Nov 14
by Tim van Erven 01 Nov '22

01 Nov '22

Dear all, In November we will have two in person talks in the thematic seminar in rapid succession at the UvA. 1. On Thursday November 10 Damien Garreau from the Université Côte d'Azur will speak about his analysis of the popular LIME method for explainable machine learning. 2. And on Monday November 14, Umut Şimşekli from INRIA/École Normale Supérieure will speak about his new generalization bounds for deep neural networks. *1. Damien Garreau *(Université Côte d'Azur, https://sites.google.com/view/damien-garreau/home) *Thursday November 10*, 16h00-17h00 In person, at the University of Amsterdam Location: Science Park 904, Room B0.204 * **What does LIME really see in images? *The performance of modern algorithms on certain computer vision tasks such as object recognition is now close to that of humans. This success was achieved at the price of complicated architectures depending on millions of parameters and it has become quite challenging to understand how particular predictions are made. Interpretability methods propose to give us this understanding. In this talk, I will present a recent result about LIME, perhaps one of the most popular methods. * * *2. Umut Şimşekli* (INRIA/École Normale Supérieure, https://www.di.ens.fr/~simsekli/) *Monday November 14*, 16h00-17h00 In person, at the University of Amsterdam Location: Science Park 904, Room A1.24 *Fractal Structure and Generalization Properties of Stochastic Optimization Algorithms* Understanding generalization in deep learning has been one of the major challenges in statistical learning theory over the last decade. While recent work has illustrated that the dataset and the training algorithm must be taken into account in order to obtain meaningful generalization bounds, it is still theoretically not clear which properties of the data and the algorithm determine the generalization performance. In this talk, I will approach this problem from a dynamical systems theory perspective and represent stochastic optimization algorithms as random iterated function systems (IFS). Well studied in the dynamical systems literature, under mild assumptions, such IFSs can be shown to be ergodic with an invariant measure that is often supported on sets with a fractal structure. We will prove that the generalization error of a stochastic optimization algorithm can be bounded based on the ‘complexity’ of the fractal structure that underlies its invariant measure. Leveraging results from dynamical systems theory, we will show that the generalization error can be explicitly linked to the choice of the algorithm (e.g., stochastic gradient descent – SGD), algorithm hyperparameters (e.g., step-size, batch-size), and the geometry of the problem (e.g., Hessian of the loss). We will further specialize our results to specific problems (e.g., linear/logistic regression, one hidden-layered neural networks) and algorithms (e.g., SGD and preconditioned variants), and obtain analytical estimates for our bound. For modern neural networks, we will develop an efficient algorithm to compute the developed bound and support our theory with various experiments on neural networks. The talk is based on the following publication: Camuto, A., Deligiannidis, G., Erdogdu, M. A., Gurbuzbalaban, M., Simsekli, U., & Zhu, L. (2021). Fractal structure and generalization properties of stochastic optimization algorithms. Advances in Neural Information Processing Systems, 34, 18774-18788. Seminar organizers: Tim van Erven Botond Szabo https://mschauer.github.io/StructuresSeminar/ -- Tim van Erven<tim(a)timvanerven.nl> www.timvanerven.nl

1 0

Fwd: [ML-news] Three 3-year positions in teaching (general AI) in Groningen, NL
by Tim van Erven 22 Oct '22

22 Oct '22

Dear ML Netherlands list, Forwarding this message, which may be of interest. In general: feel free to advertise machine learning related vacancies on the list, and also do advertise your local seminars when you think they may be of broader interest. Best, Tim -------- Forwarded Message -------- Subject: [ML-news] Three 3-year positions in teaching (general AI) in Groningen, NL Date: Fri, 21 Oct 2022 08:48:29 -0700 (PDT) From: H. Jaeger <h.jaeger(a)rug.nl> To: Machine Learning News <ml-news(a)googlegroups.com> The AI Department in the Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence (https://www.rug.nl/research/bernoulli/) at the University of Groningen (https://www.rug.nl/) offers three 3-year positions for teaching in general AI, with competitive salaries, in a most welcoming and high-ranked working environment (top-100 university, https://www.rug.nl/about-ug/profile/facts-and-figures/position-internationa…). The application deadline is November 7, 2022. Formal prerequisites are a PhD degree; outstanding holders of a MSc degree will also be considered. Details can be found at https://www.rug.nl/about-ug/work-with-us/job-opportunities/?details=00347-0… - Herbert Jaeger -- Dr. Herbert Jaeger Professor of Computing in Cognitive Materials Rijksuniversiteit Groningen Faculty of Science and Engineering - CogniGron Bernoulliborg Nijenborgh 9, 9747 AG Groningen office: Bernoulliborg 402 phone: +31 (0) 631 921588 COVID home office phone: +49 (0) 4209 930403 web: www.ai.rug.nl/minds/ -- You received this message because you are subscribed to the Google Groups "Machine Learning News" group. To unsubscribe from this group and stop receiving emails from it, send an email to ml-news+unsubscribe(a)googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/ml-news/783f83fd-ea97-4546-aaa9-7bb3e78ea… <https://groups.google.com/d/msgid/ml-news/783f83fd-ea97-4546-aaa9-7bb3e78ea…>.

1 0

CWI Scientific Meeting with the two Van Wijngaarden award winners: Marta Kwiatkowska & Susan Murphy (4 Nov 2022)
by Felix Lucka 19 Oct '22

19 Oct '22

Dear colleagues, On Friday 4 November we will have a seminar with the two Van Wijngaarden award winners: Marta Kwiatkowska (University of Oxford) and Susan Murphy (Harvard University) at the Centrum Wiskunde & Informatica (CWI) in Amsterdam. The meeting will take place between 15:00 and 17:00 in the Euler room (Amsterdam Science Park Congress Center), with drinks available afterwards. The program is as follows: * 15:00-15:45: Marta Kwiatkowska: "Safety and robustness for deep learning with provable guarantees" * 15:45-16:00: Questions and small break * 16:00-16:45: Susan Murphy: "Inference for Longitudinal Data After Adaptive Sampling" * 16:45-end: Questions followed by drinks Abstracts of the talks are copied below, Best regards, Felix Lucka, Jannis Teunissen & Peter Grunwald Marta Kwiatkowska (University of Oxford): Safety and robustness for deep learning with provable guarantees Abstract: Computing systems are becoming ever more complex, with decisions increasingly often based on deep learning components. This lecture will describe progress with developing automated verification techniques for deep neural networks to ensure safety and robustness of their decisions. The lecture will conclude with an overview of the challenges in this field. Susan Murphy (Harvard University): Inference for Longitudinal Data After Adaptive Sampling Abstract: Adaptive sampling methods, such as reinforcement learning (RL) and bandit algorithms, are increasingly used for the real-time personalization of interventions in digital applications like mobile health and education. As a result, there is a need to be able to use the resulting adaptively collected user data to address a variety of inferential questions, including questions about time-varying causal effects. However, current methods for statistical inference on such data (a) make strong assumptions regarding the environment dynamics, e.g., assume the longitudinal data follows a Markovian process, or (b) require data to be collected with one adaptive sampling algorithm per user, which excludes algorithms that learn to select actions using data collected from multiple users. These are major obstacles preventing the use of adaptive sampling algorithms more widely in practice. In this work, we provide theory for common Z-estimators based on adaptively sampled data. The inference is valid even when observations are non-stationary and highly dependent over time, and (b) allow the online adaptive sampling algorithm to learn using the data of all users. Furthermore, our inference method is robust to miss-specification of the reward models used by the adaptive sampling algorithm. This work is motivated by our work in designing the Oralytics oral health clinical trial in which an RL adaptive sampling algorithm will be used to select treatments, yet valid statistical inference is essential for conducting primary data analyses after the trial is over.

1 0

Assistant Professor of Integrative Data Science with focus on health at TU Delft
by Tina Nane 17 Oct '22

17 Oct '22

Dear all, TU Delftis proud to invite applications for an Assistant Professor (Development Track) position in integrative data science with focus on health. We welcome candidates engaged in all aspects of data science, with preference on applications in health disciplines. The Assistant Professor will be appointed at the Faculty of EEMCS, in Delft Institute of Applied Mathematics, in the Applied Probability group. With this position, you will be part to the national ‘RECENTRE <https://www.4tu.nl/recentre/>: Risk-based lifEstyle Change: daily-lifE moNiToring and Recommendations’ program, funded by the 4TU.Federation <https://www.4tu.nl/>. Within RECENTRE, the aim is to empower patients to play a leading role in their lifestyle and health within their own environment. Integrated smart sensor systems and dynamic risk profiles will be designed to support high patient engagement.This can lead to personalized adaptive recommendations with considerable potential for adoption and lifestyle change. To get authentic and up-to date risk profiles, sensor and historic data will be complemented by expert and patient subjective assessments, as well as by patient personal information.Novel methods will be consequently needed to integrate these disparate data sources. Here <https://www.tudelft.nl/over-tu-delft/werken-bij-tu-delft/vacatures/details/…>you can fine more details on the position and how to submit your application. The deadline for application is October 30th. With best regards, Tina Nane Delft Institute of Applied Mathematics Delft University of Technology

1 0

Tenure track Assistant Professor in Computer Vision and Machine Learning at the University of Amsterdam
by Stratis Gavves 17 Oct '22

17 Oct '22

Hi all, we warmly invite you to submit your application if interested in joining us as a tenure track Assistant Professor in Computer Vision and Machine Learning at the University of Amsterdam, one of the most exciting AI ecosystems at the moment. The position is together with being a lab manager of the POP-AART lab, which is set up between the University of Amsterdam, the renowned Netherlands Cancer Institute, and Elekta, the leader manufacturer of Adaptive Radiotherapy equipment. The position comes with 6 strong PhDs working on generative models, geometry, segmentation, registration, reinforcement learning, inverse models, all focusing on adaptive radiotherapy. https://vacatures.uva.nl/UvA/job/Assistant-Professor-in-Computer-Vision-and… For any questions, please do not hesitate to contact me at egavves(a)uva.nl. Best, Stratis

1 0

Thematic seminar: Damien Garreau, Thu Nov 10, 16h00 at UvA
by Tim van Erven 14 Oct '22

14 Oct '22

Dear all, In November we will have two in person talks in the thematic seminar in rapid succession at the UvA. - On Thursday November 10 Damien Garreau from the Université Côte d'Azur will speak about his analysis of the popular LIME method for explainable machine learning. - And on November 14 or 15, Umut Şimşekli from INRIA/École Normale Supérieure will speak about his new generalization bounds for deep neural networks. *Damien Garreau *(Université Côte d'Azur, https://sites.google.com/view/damien-garreau/home) *Thursday November 10*, 16h00-17h00 In person, at the University of Amsterdam Room: TBA * **What does LIME really see in images? *The performance of modern algorithms on certain computer vision tasks such as object recognition is now close to that of humans. This success was achieved at the price of complicated architectures depending on millions of parameters and it has become quite challenging to understand how particular predictions are made. Interpretability methods propose to give us this understanding. In this talk, I will present a recent result about LIME, perhaps one of the most popular methods. * * *Upcoming talks: *- Nov. 14 or 15, Umut Şimşekli <https://www.di.ens.fr/~simsekli/> from INRIA/École Normale Supérieure will speak about new generalization bounds for deep neural networks. ** Seminar organizers: Tim van Erven Botond Szabo https://mschauer.github.io/StructuresSeminar/ -- Tim van Erven<tim(a)timvanerven.nl> www.timvanerven.nl

1 0

Workshop YEQT XV "Machine Learning for Stochastic Networks"
by Tim van Erven 06 Oct '22

06 Oct '22

Forwarding on behalf of the YEQT organizers: Subject: Workshop YEQT XV "Machine Learning for Stochastic Networks" From: "Verleijsdonk, Peter" <p.verleijsdonk(a)tue.nl> Date: 06/10/2022, 15:21 To: "machine-learning-nederland(a)list.uva.nl" <machine-learning-nederland(a)list.uva.nl> CC: "Sanders, Jaron" <jaron.sanders(a)tue.nl> Dear recipient, It is our pleasure to invite you to theannual ‘Young European Queueing Theorists’ (YEQT) workshop.The15th edition,YEQT 2022, will be hosted byEindhovenUniversity of Technology and held from2-4November 2022atEurandom.The event provides an excellent opportunity for developing researchers to interact and exchange ideas in an informal, friendly, yet research-focused setting inEindhoven, the Netherlands. This year’s theme for the YEQT workshop is Machine Learning for Stochastic Networks. The aim of YEQT XV is to bring together senior researchers who have made substantial contributions to advancing machine learning techniques and the optimization of stochastic networks. The methodological focus will be on research that combines theoretical stochastic modeling and optimization together with modern machine learning techniques. Confirmed keynote speakers are: * Ton Dieker (Columbia) * Tor Lattimore (Deepmind) * LaurentMassoulié (Inria) * AlexandreProutiere (KTH) Pleaseseethe website for the full program, titles and abstracts, and registration:https://www.eurandom.tue.nl/event/workshop-yeqt-xv-machine-learning-for-stochastic-networks/ <https://www.eurandom.tue.nl/event/workshop-yeqt-xv-machine-learning-for-sto…>

1 0

CWI seminar Joseph Bakarji
by Wouter Edeling 06 Oct '22

06 Oct '22

Dear all, The next speaker in our seminar on Uncertainty Quantification and Machine Learning in scientific computing is Joseph Bakarji, working at the University of Washington with Steven Brunton and Nathan Kutz (https://www.josephbakarji.com/). The topic of his talk will be the use of sparse identification and machine learning for discovering dimensionless groups, see abstract below. The (online) talk will take place on October 13th at 16h CET. Best regards, Wouter Edeling Join Zoom Meeting https://cwi-nl.zoom.us/j/83204615411?pwd=VmNLelBwWVZLaEU4QWFYYXNZYS9qZz09 Meeting ID: 832 0461 5411 Passcode: 669720 13 October 2022 16h00 CET: Joseph Bakarji (University of Washington) : Discovering dimensionless groups from data using constrained sparse identification and deep learning methods Dimensional analysis is a robust technique for extracting insights and finding symmetries in physical systems, especially when the governing equations are not known. The Buckingham Pi theorem provides a procedure for finding a set of dimensionless groups from given parameters and variables. However, this set is often non-unique which makes dimensional analysis an art that requires experience with the problem at hand. In this talk, I'll propose a data-driven approach that takes advantage of the symmetric and self-similar structure of available measurement data to discover dimensionless groups that best collapse the data to a lower dimensional space according to an optimal fit. We develop three machine learning methods that use the Buckingham Pi theorem as a constraint: (i) a constrained optimization problem with a nonparametric function, (ii) a deep learning algorithm (BuckiNet) that projects the input parameter space to a lower dimension in the first layer, and (iii) a sparse identification of differential equations method to discover differential equations with dimensionless coefficients that parameterize the dynamics. I discuss the accuracy and robustness of these methods when applied to known nonlinear systems where dimensionless groups are known, and propose a few avenues for future research.

1 0

2024

2023

2022

2021

2020

Machine-learning-nederland