Machine Learning Lab
Computer Science Department and Centre for Cognitive Science, TU Darmstadt, Altes Hauptgebäude, Room 074, Hochschulstrasse 1, 64289 Darmstadt, Germany
Office and important requests: +49-6151-16-20820 ira.tesar (at) cs (dot) tu-darmstadt (dot) de
Meetings by appointment, general consultation: thursdays, 13:30-14:30 o'clock


Mission. The Machine Learning lab would like to make computers learn so much about the world, so rapidly and flexibly, as humans. This poses many deep and fascinating scientific problems: How can computers learn with less help from us and data? How can computers reason about and learn with complex data such as graphs and uncertain databases? How can pre-existing knowledge be exploited? How can computers decide autonomously which representation is best for the data at hand? Can learned results be physically plausible or be made understandable by us? How can computers learn together with us in the loop? To this end, we develop novel machine learning (ML) and artificial intelligence (AI) methods, i.e., novel computational methods that contain and combine for example search, logical and probabilistic techniques as well as (deep) (un)supervised and reinforcement learning methods.

News.
Oct. 2018: Chancellor Merkel describes the TU Darmstadt as a "jewel in questions of AI with all its sub-areas".
Oct. 2018: Invited talk on "Towards the Democratization of Machine Learning using Probabilistic Programming" at the Inaugral Conference on Probabilistic Programming (ProbProg) 2018.
Oct. 2018: Invited talk on "Feeding the world with big data: machines uncover spectral characteristics and dynamics of stressed plants" at "Plants and Animals: Bridging the Gap in Breeding Research" 2018.
Sept. 2018: Invited talk on "The Automatic Data Scientist" at the Logic and Learning section of the Highlights of Logic, Games and Automata 2018 conference.
Sept. 2018: Our KI 2018 tutorial on "Statistical Relational AI" in Berlin was a great success. Thanks to Tanya and Ralf!
August 2018: TUDa is the #1 German and #2 European AI institution (based on the number of publication at top venues over the last 10 years in the combined areas of Artificial Intelligence, Computer Vision, Machine Learning, Natural Language Processing, and Robotics) according to csrankings.org.
August 2018: Our European Association for AI (EurAI) Advanced Course on AI (ACAI) "Statistical Relational AI" in Ferrara (Italy) was a great success. Great lecture materials and recordings! Thanks to all lecturers and the whole team for the great event!
August 2018: The German Workshop on Knowledge Discovery and Machine Learning (KDML) 2018 we organized in Mannheim (Germany) was a lot of fun. Thanks to the whole team for the great organization and support!
August 2018: The International Conference on Uncertainty in AI (UAI) 2018 we helped organizing in Monterey (USA) was a lot of fun. Thanks to the whole team for the great organization and support!
July 2018: Invited talk on "Systems AI" at IJCAI 2018 workshop on Learning and Reasoning Workshop (LR)
July 2018: Hosted interim meeting of our "Artificial Intelligence" Research Training Group of the German National Academic Scholarship Foundation at the TU Darmstadt. Thanks for all the exciting discussions!
June 2018: Two Dagstuhl seminars accepted: "Logic and Learning" and "Software Engineering for AI-ML-based Systems." Exciting!
June 2018: The call for the establishment of a confederation of laboratories for AI research in Europe (claire-ai.org) is out. We are a key supporter!
June 2018: Participated in BMBF expert panel on "Made in Germany: Self-learning systems as Advantage for Tomorrow" at CEBIT 2018.
June 2018: Three TPM@ICML2018 papers accepted. Congrats guys!
June 2018: Founding EiC of the new Frontiers in Big Data journal on "ML and AI". Check it out!
May 2018: Keynote on "Automatic Data Scientist" at Data Science Day Jena 2018. Thanks for inviting!
April 2018: One IJCAI-ECAI survey paper on "Systems AI" and learning-based programming accepted! Congrats Parisa!
April 2018: Joint the Technical Review Board of the Robert Bosch Centre for Data Science and AI at IIT Madras, India. This is an interdisciplinary research centre for AI set up at IIT Madras. Exciting!
April 2018: Two IJCAI-ECAI papers on symbolic-numerical and lifted inference accepted! Congrats to all!
Feb. 2018: Joint the working group "Technoligical Enablers and Data Science" of the Plattform Self-Learning Systems of the Federal Ministry for Education and Research (BMBF). Excited to help shaping the future of AI in Germany.
Jan. 2018: One SysML and one ICRA paper accepted! Congrats Rudolph and Alejandro!
Dec. 2017: Our NIPS 2017 tutorial and workshop went very well. Thanks for attending!
Nov. 2017: And the P5 Award 2017 goes to the Project Group 608 "Manipulation". Congratualtions guys!
Nov. 2017: Three AAAI papers about core sets, sum-product networks, resp. autoencoders accepted. Congrats guys!
Oct. 2017: Children's university lecture on "What is actually Artificial Intelligence?" at the comprehensive school Gänsewinkel in Schwerte, Germany.
Oct. 2017: Consider to submit to our NIPS 2017 Highlights workshop!
Oct. 2017: New DFG project on Argumentative Machine Learning (CAML) within the SPP RATIO. Looking forward to our collaboration, Matthias!
Oct. 2017: Invited talk on "Tractable Data Journalism" at the Berlin Machine Learning Meetup Group.
Sept. 2017: BIBM paper on predicting the number of Angioplasty procedures using Poisson Dependency Networks accepted. Congrats Shuo!
Sept. 2017: Invited talk on declarative machine learning systems and lifted quadratic programming at the KI 2017 Sister Conference Session.
Sept. 2017: Course on "Tractable Probabilistic Graphical Models" at the 4th International Summer School on Resource-aware Machine Learning.
Sept. 2017: Consider to join our EurAI sponsored ACAI Summer School on Statistical Relational AI in Ferrara, Italy, 2018.
Sept. 2017: Consider to submit to our AAAI 2018 workshop on Declarative Learning Based Programming (DeLBP).
Sept. 2017: First week of our "AI - Facts, Chances, Risks" Research Training Group of the German National Academic Scholarship Foundation.
Sept. 2017: Invited Talk on "Tractabel Data Journalism" at SciCAR.
Sept. 2017: Consider to join our NIPS 2017 tutorial Statistical Relational AI on how to combine relational and probabilistic AI.
Aug. 2017: ICDM paper on randomized high-dimensional Weisfeiler-Lehman graph kernels accepted. Congrats Christopher!
Aug. 2017: Organizing UAI 2017 in Sydney was a lot of work but also fun. Thanks to everyone making UAI 2017 such a great experience.
Aug. 2017: Our KDD 2017 workshop on "Data Science meets Journalism" (DS+J) went very well. Thanks to everyone!
July 2017: Our BLE project on Deep Learning for Plant Phenotyping finally started at the TU Darmstadt. Welcome Patrick!
June 2017: Invited talk at CPAIOR 2017 about declarative machine learning systems and lifted quadratic programming.
May 2017: Moved to TU Darmstadt. Thanks to TU Dortmund for wonderful years.
April 2017: Consider to submit to our KDD 2017 workshop on "Data Science meets Journalism" (DS+J)
April 2017: PhD course on how to combine relational and statistical AI at the University of Trento. Thanks Andrea for a wonderful stay.
Jan. 2017: Distinguished lecture on declarative ML systems at the University of Jena.
Jan. 2017: Three papers at AAAI 2017. Congratulations guys!
Dec. 2016: Invited talk on ML Systems at the NIPS 2016 Workshop on "Neurorobotics: A Chance for New Ideas, Algorithms and Approaches"
Sept. 2016: Invited talk at DS 2016 on mathemtical models of collective attention.
Sept. 2016: Invited talk at BeneLearn 2016 on declarative machine learning systems.



People's Infos

Dr. Jing Feng (PostDoc)
currently
more info

Dr. Babak Ahmadi (PhD)
BitStar
Elena Erdmann (PhD)
ZEIT Online
Dr. Fabian Hadiji (PhD)
goedle.io
Dr. Ahmed Jawad (PhD)
Allianz
Martin Mladenov (PhD)
Google Research
Dr. Marion Neumann (PhD)
Univ. of Washington, St. Louis
Dr. Mirwaes Wahabzada (PhD)
Univ. of Bonn
Dr. Zhao Xu (PostDoc)
Fraunhofer FIT



Teaching

Course on "Statistical Relational Artificial Intelligence." winter 2018
Course on "Probabilistic Graphical Models." winter 2018
Course on "Artificial Intelligence". summer 2018
Course on "Deep Learning". summer 2018
Extended Seminar on "Interactive Machine Learning." winter 2017
Course on "Statistical Relational Artificial Intelligence." winter 2017
Course on "Probabilistic Graphical Models." winter 2017
Project group on Manipulation "Entwicklung und kritische Beleuchtung eines Frameworks zur Extraktion und Übertragung semantischer Informationen zwischen Videos." summer-winter 2017
Ph.D. course on Statistical Relational AI at the Università degli Studi di Trento, Italy. spring 2017
Ph.D. course on Lifted Inference and Collective Attention at the Università di Bari, Italy. summer 2016
Course on Wissensentdeckung in Datenbanken. summer 2014, 2016
Course on Statistical Relational Learning together with F. Riguzzi (U. Ferrara, Italy) as part of ERASMUS+. summer 2016
Course on Foundations of Data Science. summer 2015, winter 15, 2016
Course on Mathematik fuer Informatiker 1. summer 2015
Course on Probabilistic Graphical Models. winter 2013, 2014, 2015, 2016
Proseminar on Big Data Mining. summer 2014, 2015
Project group on DeepNewsDive: Maschinen lesen Zeitungen. summer-winter 2016
Project group on Infoscreens. summer-winter 2015
Seminar on Big Data Mining. winter 2013
Project lab on Across Scale Data Analysis. summer 2013
Course on Probabilistic Graphical Models. winter 2012
Course on Geoalgorithms and geo data structurs. winter 2012
Project lab on Data Mining and Pattern Recognition. winter 2012
Seminar on Geoinformation III. winter 2012
Course on Probabilistic Graphical Models. winter 2012
Course on Probabilistic Graphical Models. winter 2011
Practical lab with topics on Lifted Inference and IR. winter 2010
Course on Probabilistic Graphical Models. summer 2009, 2010
Seminar on Machine Learning for Computer Games. winter 2008
Course on Bayesian networks as part of the Advanced AI course. winter 2006





Publications








Funding


Argumentative Machine Learning (KE 1686/3-1, SPP 1999, DFG)
In this DFG project, we will investigate radically novel machine learning approaches in detail and develop the new field of “argumentative machine learning” in general: a tight integration of Computational Argumentation and Machine Learning. This has several benefits. The use of argumentation techniques allows to obtain classifiers, which are by design able to explain their decisions, and therefore addresses the recent need for Explainable AI : classifications are accompanied by a dialectical analysis showing why arguments for the conclusion are preferred to counterarguments; this automatic deliberation, validation, reconstruction and synthesis of arguments helps in assessing trust in the classifier, which is fundamental if one plans to take action based on a prediction. Argumentation techniques in machine learning also allows the easy integration of additional expert knowledge in form of arguments.
Industrial Data Science (BMBF)
In this BMBF project, the goal is to help to bridge the gap between Machine Learning and Production. We are developing novel teaching material on machine learning in engineering with a focus on production. The goal is to make it easier for engineers in production to get familar with core machine learning concepts, techniques, and algorithms. To this end, we will embed them within real applications from production.
Connecting Editors and Researchers (Zeit Online, GRK 1994, DFG)
Data Journalism, News Recommendation, and Fake News Detection: more and more AI techniques find there way into (on-line) journalism application. However, the use of Data Science, Natural Language Processing and Machine Learning in Journalism is still at the beginning. So far, there is a significant discrepancy between research and Editor's everyday life. In this cooperation between Zeit Online and the TU Darmstadt as well as the DFG Research Training Group GRK 1994 "Adaptive Preparation of Information from Heterogeneous Sources" (AIPHES), we aim at closing this gap and build a bridge between the disciplines.
Deep Inference Machines (GRK 1994, DFG)
Inference machines, viewing inference computations as trainable computation graphs, have paved the way to “deepify” classical language models. Viewing paraphrasing and harmonization as an inference task in data-driven relational probabilistic models, we therefore recast relational inference using inference machines in this project within the DFG Research Training Group GRK 1994 "Adaptive Preparation of Information from Heterogeneous Sources" (AIPHES. This allows us to lift recent advances in deep (language) modeling and learning to relational domains, consisting of (textual and visual) objects and relations among them, and to explore the resulting deep relational inference machines for data-driven textual and visual inference over heterogeneous domains.
Deep Phenotyping (BLE)
The goal of this BLE project is the optimization and objectification of phenotyping routines for crop breeding. It combines sensor technology, automation and deep learning. By using hyperspectral images and deep learning it will help to go beyond a purely visually assessed disease score for phenotyping of different genotypes.
goedle.io, EXIST
The goedle.io start-up is providing an innovative machine learning technology to maximize engagement and retention for mobile apps, e-commerce, or SaaS products. It is supported by EXIST, a support programme of the German Federal Ministry for Economic Affairs and Energy (BMWi). This programme aims at improving the entrepreneurial environment at universities and research institutes. It also aims at increasing the number and success of technology and knowledge based business start-ups.
Resource-efficient Graph Mining (SFB 876,DFG)
Linked data and networks occur often in the context of embedded systems. Sensors, RFID-chips, cameras, etc. of products of our daily life continuously produce data and communicate with each other as well as the user. A natural representation of linked data are graphs where objects correspond to the vertices of the graph and the links to its edges. In this project, we will develop new approaches and algorithms for the classification of graphs and linked data sets under resource constraints. To this aim, randomized approaches from algorithmic theory, approaches for mining and learning with graphs (in particular graph kernels) and algorithmic engineering approaches have been combined in this SFB876 research project.
Analysis and Communication for Dynamic Traffic Prognosis (SFB 876, DFG)
The goal os this SFB876 research project is the development of high precision prediction methods for the dynamic behavior of road traffic based on resource-efficient transmission of extended Floating Car Data (xFCD) and other data sources. With the help of collected data from vehicles, triggers for disturbances of the traffic flow should be detected early and countermeasures are applied in real-time. New dynamic microscopic traffic models are needed. Applying Data Mining strategies, these models are re-parameterized in real time in order to handle the heterogeneity of urban traffic.
Efficient Inference for Probabilistic Relational Models using Symmetries and Linear Programming Relaxations (GIF)
Say we know that some people in a social network are friends and some are smokers, how can we infer whether others are smokers and friends? For thousands of people this seems like a daunting computational task. However, such tasks often have strong symmetries (i.e.,repeated structures) that should intuitively translate into fewer computations. In this project, we proposed a novel approach to designing “symmetry aware” algorithms. We built on linear programming (LP) relaxations as the key underlying framework. Despite the popularity of LP relaxations in the graphical models community, they have seen very little use within the SRL literature. In this GIF project, we developed the theory and algorithms needed for applying LPs to SRL, while making effective use of symmetries.
Flexible Skill Acquisition and Intuitive Robot Tasking for Mobile Manipulation in the Real World (EU)
Flexible Skill Acquisition and Intuitive Robot Tasking for Mobile Manipulation in the Real World was a project funded by the European Commission within FP7. The goal of First-MM is to build the basis for a new generation of autonomous mobile manipulation robots that can flexibly be instructed to perform complex manipulation and transportation tasks. The project has focussed on developing a novel robot programming environment that allows even non-expert users to specify complex manipulation tasks in real-world environments. To this aim, we have built upon and extend results in robot programming, navigation, manipulation, perception, learning by instruction, and statistical relational learning to develop advanced technology for mobile manipulation robots that can flexibly be instructed even by non-expert users to perform challenging manipulation tasks in real-world environments.
Relational exploration, learning and inference (SPP 1527, DFG)
The core approach of this DFG project is to organize exploration, learning and inference on appropriate relational representations implying strong prior assumptions on the world structure. On these representations we can learn from uncertain experience compact models of action effects that generalize across objects. We transfer existing exploration theories to relational representations—leading to a novel level of explorative behavior that decidedly aims to explore objects to which the current knowledge does not generalize. This project was to our knowledge the first to combine statistical relational learning methods to tackle core problems in intelligent robotics, fueling the hope for a major advance in the field. We have demonstrated our methods on real-world robot platforms manipulating their environments.
STREAM, Fraunhofer ATTRACT
The ability to build computing systems that can observe, understand and act on human activity has long been a goal of computing research. Such systems could have profound conceptual and practical implications. Since the ability to reason and act based on activity is one of the central aspects of human intelligence, from a conceptual viewpoint such a system could cast light on computational models of intelligence. More tangibly, perhaps, machines that reason about human activity could aid humans in aspects of their lives that are today considered outside the domain of machines.
Most existing activity mining approaches indeed take uncertainties into account, but they do not consider the rich relations among people and objects that exist in the real world. STREAM’s goal was to develop formalisms, models, and algorithms for effective and robust statistical relational activity mining that enables one to develop socio-cognitive aware systems and to apply them on significant real-life applications. To this aim STREAM develop statistical relational learning methods using a Fraunhofer ATTRACT fellowship.
APRIL 1&2, EU
One of the key open questions of artificial intelligence concerns "probabilistic logic learning", i.e.the integration of probabilistic reasoning, with first order logic representations and machine learning. The overall goal of the APrIL II project is therefore to develop a sound theoretical understanding of "probabilistic logic learning" that enables one to develop effective probabilistic logic learning systems and to apply them on significant real-life applications. To realize this aim, the APrIL II consortium will (1) develop a number of significant "show-case" applications of "probabilistic logic learning" in the area of bio-informatics, more specifically, concerning protein folding, metabolic pathways, and genetics.(2) develop the needed theory, probabilistic representations, learning algorithms and systems for learning interesting probabilistic logic models in real-life applications on the basis of data. The methodology applied is that of the field of inductive logic programming, which explains the title of the project.