View from Certosa di Pontignano
As previously announced, last week I had the pleasure to present our joint work with our partners from Ruhr-University Bochum on explainable reinforcement learning at the 8th Annual Conference on machine Learning, Optimization and Data science (LOD). The presentation sparked interesting questions and lead to inspiring discussions in the enchanting ambiance of the medieval monastery in Tuscany.
As the conference was held in conjunction with the Advanced Course & Symposium on Artificial Intelligence & Neuroscience (ACAIN), we could profit from a very stimulating interdisciplinary environment with talks, tutorials, and posters covering topics reaching from the biology of neuronal development to implementation details of different deep learning frameworks.
We are looking forward to LOD 2023!
The second edition of the annual Dataninja-Retreat took place this September in Tecklenburg. The different Dataninja projects presented their proceedings, we had the pleasure of attending a lecture by Prof. Xiaoyi Jiang, and fresh PhD graduates of the KI-Starter-project kindly shared experiences and some tips from their recently concluded PhD-journey. Last but not least it was of course a very pleasant and rare occasion for seeing each other offline, exchanging ideas, experiences, struggles, and successes.
Group picture of the participants
We are pleased to announce that we will present our research on explainable reinforcement learning at the 8th Annual Conference on machine Learning, Optimization and Data science (LOD).
Carthusian monastery in Pontignano Siena, Italy. Venue of LOD 2022
Starting with its first edition in 2015, the LOD is an established international and interdisciplinary forum for research and discussion of Deep Learning, Optimization, Big Data, and Artificial Intelligence. This year's 8th edition of the will be held online and onsite in Pontignano near Siena, Italy on September 18th - 22nd 2022.
Reading the conference's manifesto “The problem of understanding intelligence is said to be the greatest problem in science today and ‘the’ problem for this century” we find this prestigious conference to be the perfect place to present our work targeted at making deep reinforcement agents explainable.
We are very grateful for the opportunity to present our paper titled “Sample-based Rule Extraction for Explainable Reinforcement Learning”, which outlines the results of our ongoing research of inducing simple, transparent, human-readable rules from well-trained deep reinforcement learning agents. A link to the article will be added as soon as it is published. For those interested, early registration for the conference is available until Sunday July 31st, 2022.
In September 2021, shortly after the Dataninja retreat, we participated at the KI 2021 – 44th German Conference on Artificial Intelligence.
Alongside fellow members of the Dataninja research training group we, Raphael Engelhardt and Wolfgang Konen from TH Köln together with Laurenz Wiskott and Moritz Lange from RUB Bochum, presented our work on rule extraction from trained reinforcement learning agents in a poster session as part of the workshop "Trustworthy AI in the wild". The conference had to be held virtually which did not affect some very interesting discussions and exchange of new ideas.
Our poster as well as the extended abstract are available online.
Between the third and fourth wave of the COVID-19 pandemic, we were in September 2021 lucky enough to hold in presence the annual retreat of the Dataninja research group, which was at the same time the group's first ever in-person meeting. The rich and balanced program included presentations of the different Dataninja projects by the respective PhD candidates, scientific talks by guest speakers, networking and outdoor activities in the nature surrounding Willingen.
While the science was top notch, there is definitely room for improvement regarding steering a canoe as the following tracking picture of one of the canoes shows 🙂 ...
(Image: Christoph J Kellner)
For the research project (RL)^3, which is lead by Laurenz Wiskott (RUB) and Wolfgang Konen (TH Köln) and that is illustrated in the figure above, there is now a press release TH Köln available: https://www.th-koeln.de/hochschule/mehr-transparenz-bei-kuenstlicher-intelligenz_84905.php (sorry, in German only!).
(RL)^3 is part of the graduate school Dataninja (Trustworthy AI for Seamless Problem Solving: Next Generation Intelligence Joins Robust Data Analysis).
We are happy to announce the dataninja.nrw inauguration event that takes place virtually on Monday, May, 3rd, 16-18.
Our research group (RL)^3 at TH Köln is part of the AI graduate school dataninja.nrw with a PhD tandem together with Ruhr University Bochum.
Please see the attached PDF dataninja_inauguration_05_03 for all the details and the programme of the inauguration event and how to register.
All interested people are free to join this event!
(RL)^3 is part of the graduate school dataninja.nrw (Trustworthy AI for Seamless Problem Solving: Next Generation Intelligence Joins Robust Data Analysis).
We are happy to announce that Samineh Bagheri won the Dissertation Price 2020 of TH Köln for her PhD thesis. Her thesis “Self-Adjusting Surrogate-Assisted Optimization Techniques for Expensive Constrained Black Box Problems” deals with state-of-the art optimization algorithms supported by RBF surrogate models.
The award ceremony took place online - as usual in these times - on Monday, Feb, 22nd, 2021, and was conducted by vice president Prof. Klaus Becker (upper row, right), attended by all other members of the presidential committee, including president Prof. Stefan Herzig (upper row, left) and of course the laureate Dr. Samineh Bagheri (lower row, left) and her supervisor Prof. Wolfgang Konen (lower row, right).
Read more about the dissertation price and Samineh Bagheri in this THK-News (sorry, in German only!)
Read more about Samineh's PhD-colloquium (the first fully-online colloq at Leiden University) in this CIOP blog post.
We are happy to announce that the CIOP group of TH Köln participated with two papers and two talks at the 9th International Conference BIOMA'2020 (Bioinspired Optimization Methods and Applications), which took place November 2020, 19th-20th, and was this year a completely online event:
- "Reinforcement Learning for N-Player Games: The Importance of Final Adaptation" by Wolfgang Konen and Samineh Bagheri. The talk presented a new approach for game learning and game playing on a variety of games with 1, 2 and 3 players.
- "Time Series Encodings with Temporal Convolutional Networks" by Markus Thill, Wolfgang Konen and Thomas Bäck. The talk presented a deep learning approach for anomaly detection of otherwise hard-to-find temporal anomalies in time series.
Although a fully-online event, the single-track conference took place in a nice atmosphere, with many discussions and informal chats. A side effect of the online format was that each speaker was asked to record a backup video, which could be played in case of technical problems. Thus, if you missed the conference, you can nevertheless take a look at the paper, the slides and the video, if you like:
- "Reinforcement Learning for N-Player Games: The Importance of Final Adaptation": paper, slides, video (15 min)
- "Time Series Encodings with Temporal Convolutional Networks": paper, slides, video (15 min)
The State of North Rhine-Westphalia provides a grant for the AI Graduate College Data-NInJA („Trustworthy AI for Seamless Problem Solving: Next Generation Intelligence Joins Robust Data Analysis“), coordinated by Prof. Barbara Hammer, University Bielefeld. The grant consists of seven PhD tandems, which were selected out of 37 applications for this grant by an expert jury.
TH Köln takes part in this Graduate College with a PhD tandem together with Ruhr-University Bochum (RUB). Under the title „(RL)³: Representation, Reinforcement und Rule Learning“ the research is conducted by Prof. Laurenz Wiskott from the Neuroinformatics Institute, RUB, and Prof. Wolfgang Konen from the Cologne Institute of Computer Science, TH Köln, together with two PhD students at each institution. The reseach aims at explainable and interpretable AI models, where methods from (deep) reinforcement learning and representation learning will play an important role. The new element of this research project is that both PhD students, one at RUB and one at TH Köln, will work closely together in the PhD tandem, together with their supervisors.