Highlights
News
-
Mini-symposium "Is knowledge useful?" (19/04)
On April 19th we will hold a mini-symposium titled "Is knowledge useful?". We will discuss how to represent knowledge, and when and how to use it for different applications. See link for details of the event!
External
-
Colloquium on uncertainties in AI (Feb 5th)
On Feb 5 the UAI group is organizing a colloquium on uncertainties in AI, together with partners at TU/e, RWTH Aachen Uni & KU Leuven. The event is held at MetaForum 4.208, 15:45-17:15, and open to everyone! See link for more info
External
-
Two papers accepted at AISTATS 2024
In the paper "Probabilistic Integral Circuits", Gennaro Gala, Cassio de Campos, Robert Peharz, Antonio Vergari, and Erik Quaeghebeur show how to represent and perform inference on continuous latent variable models in the language of circuits.In the paper "Attention-based Multi-instance Mixed Models", Jan P. Engelmann, Alessandro Palma, Jakub M. Tomczak, Fabian J Theis, and Francesco Paolo Casale present GMIL, a framework integrating Generalized Linear Mixed Models (GLMM) and Multiple Instance Learning (MIL), that upholds the advantages of linear models while modeling data heterogeneity uisng pre-defined embeddings.
External
-
Four papers accepted at ICLR 2024
In the paper "Learning Large DAGs is Harder than you Think: Many Losses are Minimal for the Wrong DAG", Jonas Seng, Matej Zečević, Devendra Singh Dhami, and Kristian Kersting consider the structure learning problem, which is hard for even simple DAG’s. They formally proved that measurement scale can decide which DAGs will minimize the MMSE or log-likelihood based losses such as ELBO and BIC and thus should be treated with care.
In the paper "Conserve-Update-Revise to Cure Generalization and Robustness Trade-off in Adversarial Training", Shruthi Gowda, Bahram Zonooz, and Elahe Arani propose a selective adversarial training method that enhances the trade-off between standard and robust generalization while also mitigating robust overfitting.
In the paper "The Effectiveness of Random Forgetting for Robust Generalization", Vijaya Raghavan T Ramkumar, Bahram Zonooz, and Elahe Arani propose a method that alternates between the forgetting phase, which randomly forgets a subset of weights and regulates the model's information through weight reinitialization, and the relearning phase, which mitigate robust overfitting.In the paper "Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning", Yucheng Yang, Tianyi Zhou, Qiang He, Lei Han, Mykola Pechenizkiy, and Meng Fang propose a novel disentanglement metric LSEPIN and build an information-geometric connection between LSEPIN and downstream task adaptation cost in an unsupervised reinforcement learning setting.
External
-
Joaquin Vanschoren chair of the MLCommons AI Safety working group
Joaquin Vanschoren became a chair of the MLCommons AI Safety working group, together with Percy Liang (Stanford) and Peter Mattseon (Google). You can join too! Follow the link for more information.
External
-
Two papers accepted at AAMAS 2024
In the paper "MaDi: Learning to Mask Distractions for Generalization in Visual Deep Reinforcement Learning", Bram Grooten, Tristan Tomilin, Gautham Vasan, Matthew E. Taylor, A. Rupam Mahmood, Meng Fang, Mykola Pechenizkiy, and Decebal Mocanu propose an algorithm that learns to Mask Distractions in pixel-based reinforcement learning. Through simulation and real robotic experiments this is shown to improve generalization performance on environments with random videos playing in the background. The paper "Automatic Curriculum of Unsupervised Reinforcement Learning (ACURL)", by Yucheng Yang, Tianyi Zhou, Lei Han, Meng Fang, and Mykola Pechenizkiy, introduces a combination of metrics that can evaluate diverse properties of unsupervised reinforcement learning (URL). It proposes an automatic curriculum that uses a non-stationary multi-armed bandit algorithm to select URL objectives for different learning episodes, resulting in a balanced improvement on all the metrics.
External
-
Daphne Miedema successfully defended her PhD cum laude
On Friday (January 12th, 2024), Daphne Miedema successfully defended her PhD cum laude. Her thesis is entitled On Learning SQL: Disentangling concepts in Data Systems Education. You can read more about it by visiting the TUe website or accessing her thesis.
External
-
Igor Smit wins the KHMW Young Talent Graduation Award for Data Science
Igor Smit won the national KHMW Young Talent Graduation Prize for Data Science. Under the supervision of Yingqian Zhang, Zaharah Bukhsh and Mykola Pechenizkiy, Igor developed a deep reinforcement learning (RL) approach to efficiently manage workers and robots working together to pick items in a warehouse.
External
-
ML Postdocs vacancies available
We still have several vacancies for machine learning Postdocs. Join our vibrant international research team at TU Eindhoven! Follow the link to apply.
External
-
DataEd returns at SIGMOD2024
The third iteration of the workshop DataEd - Data Systems Education: Bridging education practice with education research, will happen again at SIGMOD2024. We are looking forward to another successful iteration!
External
-
Robert Brijder joins the Database group as an Assistant Professor
Robert Brijder joins the Database group as an Assistant Professor. Robert's research aims to provide foundations for scalable and principled data management systems for Data Science and Machine Learning.
External
-
Multivariate Time Series Analytics workshop
Our colleague, Odysseas Papapetrou, is co-chairing the MulTiSa workshop (Multivariate Time Series Analytics), which will be co-located with ICDE’24. If you are working on time series, this is a great opportunity to hear about exciting problems and the state-of-the-art in this field, and to submit your work and get feedback from the leading experts.
External
-
GenAI: More than ChatGPT. A new paradigm for the industrial sector
Jakub Tomczak was interviewed by Robert Weber from the Industrial AI podcast (GenAI: More than ChatGPT. A new paradigm for the industrial sector). Follow the link to listen to the podcast.
External
-
Honorary mention for the best thesis award at BNAIC/Benelearn 2023
The MSc student Andrei Rykov, supervised by Sibylle Hess, got an honorary mention for the best thesis award at BNAIC/Benelearn 2023 for his thesis "Robust Deep Spectral Clustering". The conference received 37 theses, of which 3 were nominated for the award.
External
Mini-symposium "Is knowledge useful?" (19/04)
On April 19th we will hold a mini-symposium titled "Is knowledge useful?". We will discuss how to represent knowledge, and when and how to use it for different applications. See link for details of the event!
External
Colloquium on uncertainties in AI (Feb 5th)
On Feb 5 the UAI group is organizing a colloquium on uncertainties in AI, together with partners at TU/e, RWTH Aachen Uni & KU Leuven. The event is held at MetaForum 4.208, 15:45-17:15, and open to everyone! See link for more info
External
Two papers accepted at AISTATS 2024
In the paper "Probabilistic Integral Circuits", Gennaro Gala, Cassio de Campos, Robert Peharz, Antonio Vergari, and Erik Quaeghebeur show how to represent and perform inference on continuous latent variable models in the language of circuits.In the paper "Attention-based Multi-instance Mixed Models", Jan P. Engelmann, Alessandro Palma, Jakub M. Tomczak, Fabian J Theis, and Francesco Paolo Casale present GMIL, a framework integrating Generalized Linear Mixed Models (GLMM) and Multiple Instance Learning (MIL), that upholds the advantages of linear models while modeling data heterogeneity uisng pre-defined embeddings.
External
Four papers accepted at ICLR 2024
In the paper "Learning Large DAGs is Harder than you Think: Many Losses are Minimal for the Wrong DAG", Jonas Seng, Matej Zečević, Devendra Singh Dhami, and Kristian Kersting consider the structure learning problem, which is hard for even simple DAG’s. They formally proved that measurement scale can decide which DAGs will minimize the MMSE or log-likelihood based losses such as ELBO and BIC and thus should be treated with care.
In the paper "Conserve-Update-Revise to Cure Generalization and Robustness Trade-off in Adversarial Training", Shruthi Gowda, Bahram Zonooz, and Elahe Arani propose a selective adversarial training method that enhances the trade-off between standard and robust generalization while also mitigating robust overfitting.
In the paper "The Effectiveness of Random Forgetting for Robust Generalization", Vijaya Raghavan T Ramkumar, Bahram Zonooz, and Elahe Arani propose a method that alternates between the forgetting phase, which randomly forgets a subset of weights and regulates the model's information through weight reinitialization, and the relearning phase, which mitigate robust overfitting.In the paper "Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning", Yucheng Yang, Tianyi Zhou, Qiang He, Lei Han, Mykola Pechenizkiy, and Meng Fang propose a novel disentanglement metric LSEPIN and build an information-geometric connection between LSEPIN and downstream task adaptation cost in an unsupervised reinforcement learning setting.
External
Joaquin Vanschoren chair of the MLCommons AI Safety working group
Joaquin Vanschoren became a chair of the MLCommons AI Safety working group, together with Percy Liang (Stanford) and Peter Mattseon (Google). You can join too! Follow the link for more information.
External
Two papers accepted at AAMAS 2024
In the paper "MaDi: Learning to Mask Distractions for Generalization in Visual Deep Reinforcement Learning", Bram Grooten, Tristan Tomilin, Gautham Vasan, Matthew E. Taylor, A. Rupam Mahmood, Meng Fang, Mykola Pechenizkiy, and Decebal Mocanu propose an algorithm that learns to Mask Distractions in pixel-based reinforcement learning. Through simulation and real robotic experiments this is shown to improve generalization performance on environments with random videos playing in the background. The paper "Automatic Curriculum of Unsupervised Reinforcement Learning (ACURL)", by Yucheng Yang, Tianyi Zhou, Lei Han, Meng Fang, and Mykola Pechenizkiy, introduces a combination of metrics that can evaluate diverse properties of unsupervised reinforcement learning (URL). It proposes an automatic curriculum that uses a non-stationary multi-armed bandit algorithm to select URL objectives for different learning episodes, resulting in a balanced improvement on all the metrics.
External
Daphne Miedema successfully defended her PhD cum laude
On Friday (January 12th, 2024), Daphne Miedema successfully defended her PhD cum laude. Her thesis is entitled On Learning SQL: Disentangling concepts in Data Systems Education. You can read more about it by visiting the TUe website or accessing her thesis.
External
Igor Smit wins the KHMW Young Talent Graduation Award for Data Science
Igor Smit won the national KHMW Young Talent Graduation Prize for Data Science. Under the supervision of Yingqian Zhang, Zaharah Bukhsh and Mykola Pechenizkiy, Igor developed a deep reinforcement learning (RL) approach to efficiently manage workers and robots working together to pick items in a warehouse.
External
ML Postdocs vacancies available
We still have several vacancies for machine learning Postdocs. Join our vibrant international research team at TU Eindhoven! Follow the link to apply.
External
DataEd returns at SIGMOD2024
The third iteration of the workshop DataEd - Data Systems Education: Bridging education practice with education research, will happen again at SIGMOD2024. We are looking forward to another successful iteration!
External
Robert Brijder joins the Database group as an Assistant Professor
Robert Brijder joins the Database group as an Assistant Professor. Robert's research aims to provide foundations for scalable and principled data management systems for Data Science and Machine Learning.
External
Multivariate Time Series Analytics workshop
Our colleague, Odysseas Papapetrou, is co-chairing the MulTiSa workshop (Multivariate Time Series Analytics), which will be co-located with ICDE’24. If you are working on time series, this is a great opportunity to hear about exciting problems and the state-of-the-art in this field, and to submit your work and get feedback from the leading experts.
External
GenAI: More than ChatGPT. A new paradigm for the industrial sector
Jakub Tomczak was interviewed by Robert Weber from the Industrial AI podcast (GenAI: More than ChatGPT. A new paradigm for the industrial sector). Follow the link to listen to the podcast.
External
Honorary mention for the best thesis award at BNAIC/Benelearn 2023
The MSc student Andrei Rykov, supervised by Sibylle Hess, got an honorary mention for the best thesis award at BNAIC/Benelearn 2023 for his thesis "Robust Deep Spectral Clustering". The conference received 37 theses, of which 3 were nominated for the award.
External