Dear all,
Monday December 11th at 11h00CET (in the L120 room) the 4th and final seminar++ talk from the CWI semester programme on Scientific Machine Learning will take place. Giovanni Stabile from the University of Urbino in Italy will present new work on nonlinear model order reduction, the full abstract can be found below.
Online attendence is possible via:
Join Zoom Meeting
https://cwi-nl.zoom.us/j/82378289134?pwd=djNCTHI4SjE2K1NhWFVhQW5yTzRxZz09
Meeting ID: 823 7828 9134
Passcode: 139939
Feel free to share the link with others.
Kind regards,
Wouter Edeling
------------------
Title: From linear to nonlinear model order reduction: some results and perspectives
Abstract: Non-affine parametric dependencies, nonlinearities, and
advection-dominated regimes of the model of interest can result in a slow
Kolmogorov n-width decay, which precludes the realization of efficient
reduced-order models based on Proper Orthogonal Decomposition. Among the
possible solutions, there are purely data-driven methods that leverage
nonlinear approximation techniques such as autoencoders and their variants
to learn a latent representation of the dynamical system, and then evolve
it in time with another architecture. Despite their success in many
applications where standard linear techniques fail, more has to be done to
increase the interpretability of the results, especially outside the
training range and not in regimes characterized by an abundance of data.
Not to mention that none of the knowledge on the physics of the model is
exploited during the predictive phase. In this talk, in order to overcome
these weaknesses, I introduce a variant of the nonlinear manifold method
introduced in previous works with hyper-reduction achieved through reduced
over-collocation and teacher-student training of a reduced decoder. We test
the methodology on different problems with increasing complexity.
Hi all,
I just wanted to remind you about our second Amsterdam Causality Meeting is
happening *this Thursday, November 23rd.*
The aim of the seminar is to bring together researchers in causal inference
from the VU, UvA, Amsterdam UMC and CWI, but it is open to everyone.
Here are the details in short:
*Date:* Thursday, November 23rd 14:30-17:30
*Location:* New University building on the *VU campus* (De Boelelaan 1111,
Amsterdam), room NU-5A47.
*Program (abstracts below):*
14.30-15.30: Johannes Textor (RU, RUMC): Are DAGs really useful for causal
reasoning in complex systems?
15.30-16.30: Julia Kowalska (AUMC): Regression discontinuity designs from
Bayesian perspective: opportunities and challenges.
16.30-17.30: Drinks
If you're interested in this event or in the seminar series, please check
our website <https://amscausality.github.io/index>.
For announcements regarding upcoming meetings, you can also register
to our Google
group <amscausality(a)googlegroups.com>.
This meeting is financially supported by the ELLIS unit Amsterdam
<https://ellis.eu/units/amsterdam> and the Big Statistics
<https://www.bigstatistics.nl/> group.
Cheers,
Sara Magliacane
Joris Mooij
Stéphanie van der Pas
============================================================
Abstracts:
Johannes Textor (RU, RUMC): *Are DAGs really useful for causal reasoning in
complex systems?*
Much causal inference methodology is based on the assumption that the world
can be neatly and meaningfully described by DAGs or similar formalisms. I
will explore the use of simulation to investigate when and how this
assumption breaks down when facing complex systems with emergent
properties. Specifically, I will discuss a simple biophysical simulation
model of a cell for which we have both complete knowledge of the
computation graph and a good understanding of its emergent properties and
phenomenology. Using this system as an example, I will illustrate the
conceptual and ontological problems that arise when attempting to formulate
causal processes in terms of graphical models whose nodes represent
high-level descriptions of emergent phenomena. I argue that these problems
are similar to those resulting from the use of social categories as nodes
in DAGs, and that they cannot be easily addressed by existing extensions of
the DAG framework.
Julia Kowalska (AUMC): *Regression discontinuity designs from Bayesian
perspective: opportunities and challenges.*
Regression discontinuity design (RDD) is a quasi-experimental design that
aims at the causal effect estimation of an intervention that is assigned
based on a cutoff criterion. Regression discontinuity design exploits the
idea that close to the cutoff units below and above the cutoff are similar
hence can be meaningfully compared. However, the causal effect can be
estimated only locally at the cutoff point. In many experiments, the cutoff
criterion serves only as a guideline rather than a strict rule.
Nonetheless, the guidelines may not be publicly known, or the cutoff used
in practice is shifted with respect to the official one. If the analysis is
performed at a false cutoff point, it leads to meaningless results, but as
the intervention assignment is binary, the location of the cutoff may be
unclear. Through the Bayesian approach, we can incorporate prior knowledge
and uncertainty about the cutoff location in the causal effect estimation.
At the same time, RDD is a boundary point estimation problem, whereas the
Bayesian model is fitted to the whole data. Therefore, a natural challenge
arises: how to make Bayesian inference more local?
===========================================================
PhD position on Interactive Causal Explanations for Patient-centric Personalised Health (1.0 FTE, 4yrs.)
Institution : Radboud University Nijmegen / TNO, Netherlands
Keywords : causal inference, healthcare, explainable AI, machine learning
Application deadline : 6 December 2023
Website : https://www.ru.nl/en/working-at/job-opportunities/phd-candidate-interactive…
===========================================================
Summary
Are you interested in applying state-of-the-art causal methods to support patient-centric healthcare? Do you want to work in an interdisciplinary environment collaborating with machine learning experts, cognitive scientists and medical specialists? If so, then this vacancy may be for you!
Description
Within our Personalised Care in Oncology <https://www.personalisedcareinoncology.nl/> (PersOn) consortium, we are looking for a talented PhD candidate to work on applying principled causal model inference to help patients make informed decisions about their preferred treatment alternatives. We will use modern explainability techniques, such as Causal Shapley values, to translate complex counterfactual predictions into patient-friendly, understandable information, allowing patients to explore the impact of different available treatments. Together with other PhD candidates and postdoctoral researchers in the consortium, you will help build an integrated causal model of the target oncological domain, and use that as a basis for answering interactive `what if' questions tailored to characteristics and preferences of individual patients. You will work with causal model experts, as well as with researchers, healthcare practitioners, and stakeholders in explainable decision support. Your contribution to the project will be both foundational, advancing the state of the art in personalised causal inference, and applied, focusing on user requirements in terms of explanation and justification to help advance shared decision making about personalised healthcare.
We are:
The position is part of a collaboration between the Data Science group <https://www.ru.nl/datascience/> in the Faculty of Science at the Radboud University, and the Microbiology and Systems Biology research group at the Dutch research institute TNO <https://www.tno.nl/en/>.
Your home base will be the Data Science group, part of the vibrant and growing Institute for Computing and Information Sciences (iCIS) at Radboud University. The group's main research foci are 1) the design and understanding of deep/causal machine learning methods, 2) modern information retrieval and big data, and 3) computational immunology, with a keen eye on applications in other scientific domains, in particular healthcare, as well as industry. Our group currently consists of ca. 50 researchers, including 25 PhDs.
In addition, you will work in the Microbiology and Systems Biology research group of the Health, Living and Work unit of the Netherlands Organisation for Applied Scientific Research (TNO), in close collaboration with the data science group of the ICT, Strategy & Policy unit that is working on solutions that enrich information systems and artificial intelligence (AI) with human knowledge and experience. The focus of this group is to optimise health and cure lifestyle-related disease from a systems biology view. The group is a very multidisciplinary team, including biologists, data scientists, bioinformaticians, etc.
Both groups strongly promote an open, inclusive and supportive work environment.
What we expect from you:
You have an MSc degree in natural science, computer science, mathematics, or a related discipline. You have a strong interest in multidisciplinary research, especially on the interface between artificial intelligence and health. You are highly motivated, open-minded, and determined to obtain a PhD degree. As you will be working in two different research groups, you need to be flexible, communicative and able to work in a multidisciplinary team.
For more information about this vacancy and details on how to apply, see the website (above), or contact:
* Dr Tom Claassen, tel: +31 24 3652019, email: tom.claassen(a)ru.nl <mailto:tom.claassen@ru.nl> (iCIS)
* Dr Jildau Bouwman, tel: +31 88 8661678, email: jildau.bouwman(a)tno.nl <mailto:jildau.bouwman@tno.nl> (TNO)
===========================================================
Dear colleague,
The semester programme on Scientific Machine Learning is currently up and running at CWI in Amsterdam! One of the main events is the Scientific Machine Learning Workshop on December 6-8, co-organized by CWI and 4TU.AMI. This workshop aims to connect the international and national research communities, bringing together academia and industry to discuss the latest advances in scientific machine learning, a rapidly growing field that is transforming the way we solve scientific and engineering problems.
The workshop will be probably be highly interesting to you and/or your colleagues. Please forward the invitation to your research groups. We have a stellar line-up of international speakers, like Jan Hesthaven (EPFL), Sid Mishra (ETH) and Dirk Hartmann (Siemens). In addition to the keynote talks, the workshop will feature
- presentations by national experts,
- a poster session,
- a panel discussion led by the Dutch AIM (AI & Mathematics) network.
Please register before November 24, via the website of the workshop:
https://www.cwi.nl/en/events/cwi-research-semester-programs/workshop-scient…
Kind regards,
Benjamin Sanderse (Centrum Wiskunde & Informatica, The Netherlands)
Mengwu Guo (University of Twente, The Netherlands)
Alexander Heinlein (Delft University of Technology, The Netherlands)
-------------------
Dr. ir. B. (Benjamin) Sanderse
Head Scientific Computing group
Centrum Wiskunde & Informatica, Amsterdam
e: b.sanderse(a)cwi.nl
www: http://www.cwi.nl/~sanderse
t: +31 (0)20 592 4085
The following is a new meeting request:
Subject: [ml_ned] BeNeRL Reinforcement Learning Seminar: Cansu Sancaktar (Nov 16)
Organizer: "Gao Peng" <Gao.Peng(a)cwi.nl>
Time: Thursday, 16 November 2023 4:00 PM to 5:00 PM
Invitees: machine-learning-nederland(a)list.uva.nl; z.yang(a)liacs.leidenuniv.nl; machine-learning-nederland(a)list.uva.nl
*~*~*~*~*~*~*~*~*~*
From: Z. <machine-learning-nederland(a)list.uva.nl>
To: machine-learning-nederland <machine-learning-nederland(a)list.uva.nl>
Date: Friday, 10 November 2023 11:45 AM CET
Subject: [ml_ned] BeNeRL Reinforcement Learning Seminar: Cansu Sancaktar (Nov 16)
Dear colleagues,
Our next BeNeRL Reinforcement Learning Seminar (Nov 16) is coming: Speaker: Cansu Sancaktar (https://csancaktar.github.io), PhD student at Max Planck Institute. Title: Playful Exploration in Reinforcement Learning Date: November 16, 16.00-17.00 (CET) Please find full details about the talk below this email and on the website of the seminar series: https://www.benerl.org/seminar-series
The goal of the online BeNeRL seminar series is to invite RL researchers (mostly advanced PhD or early postgraduate) to share their work. In addition, we invite the speakers to briefly share their experience with large-scale deep RL experiments, and their style/approach to get these to work. We also maintain a page with summary of advice of previous speakers: https://www.benerl.org/seminar-series/summary-advice-on-reinforcement-learn…
We would be very glad if you forward this invitation within your group and to other colleagues that would be interested (also outside the BeNeRL region). Hope to see you on November 16!
Kind regards, Zhao Yang & Thomas Moerland Leiden University
——————————————————————
Upcoming talk:
Date: November 16, 16.00-17.00 (CET) Speaker: Cansu Sancaktar (https://csancaktar.github.io)
Title: Playful Exploration in Reinforcement Learning Zoom: https://universiteitleiden.zoom.us/j/65411016557?pwd=MzlqcVhzVzUyZlJKTEE0Nk…
Abstract: Designing artificial agents that explore their environment efficiently via intrinsic motivation, akin to children's curious play, remains a challenge in complex environments. In this talk, she will present CEE-US which combines multi-horizon planning, epistemic uncertainty reduction, and structured world models for sample-efficient and interaction-rich exploration, particularly in multi-object manipulation environments. She will showcase how the self-reinforcing cycle between good models and good exploration opens up another avenue: zero-shot generalization to downstream tasks via model-based planning. In the second part of the talk, she will introduce regularity as an intrinsic reward signal, which we proposed in recent work. She will present how taking inspiration from developmental psychology, they inject the notion of creating regularities/symmetries during free play and achieve superior success rates in assembly tasks in a multi-object manipulation environment.
Bio: Cansu Sancaktar is a PhD student at the Max Planck Institute for Intelligent Systems, supervised by Prof. Georg Martius. Her research focuses on curiosity and intrinsically-motivated Reinforcement Learning (RL). Taking inspiration from developmental psychology, she is developing methods to help robots explore their environment efficiently without extrinsic rewards, similar to how children perform free play. Before starting her PhD, she completed her Bachelor’s and Master’s degrees in Electrical Engineering and Information Technology at TU Munich. Her specialization was in Robotics and Automation. During her studies, she worked on various machine learning projects spanning diverse fields such as robotics, signal processing, communications and neuroscience.
Dear colleagues,
Our next BeNeRL Reinforcement Learning Seminar (Nov 16) is coming:
Speaker: Cansu Sancaktar (https://csancaktar.github.io<https://csancaktar.github.io/>), PhD student at Max Planck Institute.
Title: Playful Exploration in Reinforcement Learning
Date: November 16, 16.00-17.00 (CET)
Please find full details about the talk below this email and on the website of the seminar series: https://www.benerl.org/seminar-series
The goal of the online BeNeRL seminar series is to invite RL researchers (mostly advanced PhD or early postgraduate) to share their work. In addition, we invite the speakers to briefly share their experience with large-scale deep RL experiments, and their style/approach to get these to work. We also maintain a page with summary of advice of previous speakers: https://www.benerl.org/seminar-series/summary-advice-on-reinforcement-learn…
We would be very glad if you forward this invitation within your group and to other colleagues that would be interested (also outside the BeNeRL region). Hope to see you on November 16!
Kind regards,
Zhao Yang & Thomas Moerland
Leiden University
——————————————————————
Upcoming talk:
Date: November 16, 16.00-17.00 (CET)
Speaker: Cansu Sancaktar (https://csancaktar.github.io<https://csancaktar.github.io/>)
Title: Playful Exploration in Reinforcement Learning
Zoom: https://universiteitleiden.zoom.us/j/65411016557?pwd=MzlqcVhzVzUyZlJKTEE0Nk…
Abstract: Designing artificial agents that explore their environment efficiently via intrinsic motivation, akin to children's curious play, remains a challenge in complex environments. In this talk, she will present CEE-US which combines multi-horizon planning, epistemic uncertainty reduction, and structured world models for sample-efficient and interaction-rich exploration, particularly in multi-object manipulation environments. She will showcase how the self-reinforcing cycle between good models and good exploration opens up another avenue: zero-shot generalization to downstream tasks via model-based planning. In the second part of the talk, she will introduce regularity as an intrinsic reward signal, which we proposed in recent work. She will present how taking inspiration from developmental psychology, they inject the notion of creating regularities/symmetries during free play and achieve superior success rates in assembly tasks in a multi-object manipulation environment.
Bio: Cansu Sancaktar is a PhD student at the Max Planck Institute for Intelligent Systems, supervised by Prof. Georg Martius. Her research focuses on curiosity and intrinsically-motivated Reinforcement Learning (RL). Taking inspiration from developmental psychology, she is developing methods to help robots explore their environment efficiently without extrinsic rewards, similar to how children perform free play. Before starting her PhD, she completed her Bachelor’s and Master’s degrees in Electrical Engineering and Information Technology at TU Munich. Her specialization was in Robotics and Automation. During her studies, she worked on various machine learning projects spanning diverse fields such as robotics, signal processing, communications and neuroscience.
On behalf of the organizers, see the announcement below:
-------------------------------------------------------------------------
Dear colleagues,
On November 27 afternoon, we have organised a symposium commemorating the work of John von Neumann. His contributions have left an indelible mark on various scientific disciplines, fostering specific cross-connections between theory and practice in fields such as digital computing, game theory, algebra, artificial intelligence, and numerous other areas built upon his foundational research. With four speakers and prizes to win at a quiz, we hope to see many of you there!
Date and time: 27 November 2023, 13:45--17:00
Location: Cafe X Body & Mind room
(Sportcentrum TU Delft, Mekelweg 8)
Join us for this wonderful event and the reception afterwards! Registration is free via this link:
https://www.eventbrite.com/e/neumann-symposium-the-man-from-the-future-tick…
Dear all,
Announcing a talk by Debabrota Basu (INRIA Lille France, https://debabrota-basu.github.io/) as below:
Title: When Privacy meets Partial Information: Refining the Differential Privacy Definitions, Lower Bounds, and Algorithm Designs for Sequential Learning
Room: L120, CWI, 123 Science Park, Amsterdam
Time: 10:30 AM - 11:30 AM
Abstract: Bandits act as an archetypal model of sequential learning, where one has limited information regarding the utilities of a set of decisions and can know more about the utility of a decision only by choosing it. The goal of a bandit algorithm is either (a) to maximise the total accumulated utility over a given number of interactions, or (b) to find the decision with maximal utility through minimal number of interactions. As bandits are are progressively used for data-sensitive applications, such as designing adaptive clinical trials, tuning hyper-parameters, recommender systems etc., it is imperative to ensure data privacy of these algorithms. Motivated by this, we study the impact of preserving Differential Privacy in bandits with different goals (both (a) and (b)). We answer three questions:
i. How to define Differential Privacy in bandits as both the input and output are generated progressively through past data-driven interactions?
ii. What are the changes in the fundamental hardness of bandits problems (both (a) and (b)) if we ensure ε-Differential Privacy?
iii. How to modify existing bandit algorithms (both (a) and (b)) to simultaneously ensure ε-Differential Privacy and achieve optimal performance?
Our study yields new information-theoretic quantities and a generic algorithm demonstrating that in most of the cases, ε-Differential Privacy can be achieved almost for free in bandits.
The talk is based on the works: https://arxiv.org/abs/2209.02570 and https://arxiv.org/abs/2309.02202.
Please email me if you would like to meet with the speaker. He is visiting CWI from 6-10 Nov.
Best,
Aditya.
This might be of interest.
-Frans
--------------------------------------------------------------------------------------
*Collaborative AI and Model**l**ing of Humans (CAIHu 2024)*
*AAAI-24 Bridge Program *
21 Feb 2024
Vancouver - Canada
--------------------------------------------------------------------------------------
As a part of the *AAAI 2024 Bridge program*, we are happy to announce
the bridge titled *“Collaborative AI and Modelling of Humans.”*
In this one-day event, we aim to bridge the fields that deal with
Human-AI collaboration and user modelling – including but not limited to
interactive machine learning, cognitive science,
human-computer-interaction, human-robot interaction, and behavioural
game theory. We provide a common forum for these communities to discuss
topics relevant to Human-AI collaboration and user modelling, and hope
to catalyse sustained interdisciplinary collaboration around the theme
of our bridge.
*Call for Submissions***
Our Bridge welcomes original contributions, distillation of recent
advances, and novel perspectives on how to improve collaboration across
fields and anticipated challenges, ranging between 2-8 pages in length.
As there are no associated proceedings of this program, papers that have
been or will be submitted or published in other conferences or journals
are also welcome.
The space of disciplines covered by the relevant fields is very large
and submissions are expected to cover topics such as:
* Machine learning with human in the loop
* Computational Cognitive models/ User models
* Theory of mind
* Computational rationality
* Ad Hoc Teamwork, Multi-agent Learning for Human-AI collaboration /
Human-Machine Teaming
All accepted submissions will be invited to present a poster. We will
make all the accepted papers available on our website prior to the program.
The Bridge will be held in Vancouver on Feb 21st, 2024, as a part of the
AAAI 2024 conference.
*Submissions are due on Nov 22, 2023, 11:59 PM.*
Notifications of acceptance will be sent out by Dec 12, 2023.
*And the early registration deadline is on Dec 20, 2023, 11:59 PM.*
These deadlines are “anywhere on earth” / UTC -12.
For more detailed information please visit:
Bridge website: https://sites.google.com/view/collab-ai-and-human-modeling/
Call for proposals:
https://sites.google.com/view/collab-ai-and-human-modeling/call-for-papers
AAAI 2024 Bridge program:
https://aaai.org/aaai-conference/aaai-24-bridge-program/
All attendees should register for the Bridge program. Please see
https://aaai.org/aaai-conference/registration/
Please email us at caihu.aaai2024(a)gmail.com if you have any questions
*Organizing Committee*
* Frans A. Oliehoek; Delft University of Technology, Netherlands;
* Matthew E. Taylor; University of Alberta & Alberta Machine
Intelligence Institute, Canada
* Andrew Howes; University of Exeter, England
* Nuria Oliver; ELLIS Alicante, Spain
* Samuel Kaski; Aalto University, Finland and University of Manchester, UK