Machine Learning in Nature

Apart from occassional news entries, comptuer science rarely makes it into Nature. A quick count in the Web of Science results in 33 articles, the last one – a year ago – by Ekert, A. & Renner, R. 2014. The ultimate physical limits of privacy. Nature, 507, (7493), 443-447, and the most prominent one surely the one with 3,200 citations by Strogatz, S. H. 2001. Exploring complex networks. Nature, 410, (6825), 268-276.

Now, machine learning has made it into Nature: The group of DeepMind Technologies founded by Demis Hassabis in 2011 as a start-up company, and purchased by Google for approx. 400 Million USD in 2014, has published a paper, which appeared today, 26.02.2015:

Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A., Antonoglou, I., King, H., Kumaran, D., Wierstra, D., Legg, S. & Hassabis, D. 2015. Human-level control through deep reinforcement learning. Nature, 518, (7540), 529-533.

Abstract: The theory of reinforcement learning provides a normative account, deeply rooted in psychological and neuroscientific perspectives on animal behaviour, of how agents may optimize their control of an environment. To use reinforcement learning successfully in situations approaching real-world complexity, however, agents are confronted with a difficult task: they must derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations. Remarkably, humans and other animals seem to solve this problem through a harmonious combination of reinforcement learning and hierarchical sensory processing systems, the former evidenced by a wealth of neural data revealing notable parallels between the phasic signals emitted by dopaminergic neurons and temporal difference reinforcement learning algorithms. While reinforcement learning agents have achieved some successes in a variety of domains, their applicability has previously been limited to domains in which useful features can be handcrafted, or to domains with fully observed, low-dimensional state spaces. Here we use recent advances in training deep neural networks to develop a novel artificial agent, termed a deep Q-network, that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning. We tested this agent on the challenging domain of classic Atari 2600 games. We demonstrate that the deep Q-network agent, receiving only the pixels and the game score as inputs, was able to surpass the performance of all previous algorithms and achieve a level comparable to that of a professional human games tester across a set of 49 games, using the same algorithm, network architecture and hyperparameters. This work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks.

Subject terms: Computer Science

The editors summary: For an artificial agent to be considered truly intelligent it needs to excel at a variety of tasks considered challenging for humans. To date, it has only been possible to create individual algorithms able to master a single discipline — for example, IBM’s Deep Blue beat the human world champion at chess but was not able to do anything else. Now a team working at Google’s DeepMind subsidiary has developed an artificial agent — dubbed a deep Q-network — that learns to play 49 classic Atari 2600 ‘arcade’ games directly from sensory experience, achieving performance on a par with that of an expert human player. By combining reinforcement learning (selecting actions that maximize reward — in this case the game score) with deep learning (multilayered feature extraction from high-dimensional data — in this case the pixels), the game-playing agent takes artificial intelligence a step nearer the goal of systems capable of learning a diversity of challenging tasks from scratch.

More information:


Feb, 17, 2015 > Seminar Talk by Hubert Wagner

Title: Topological analysis of text data.

Lecturer: Hubert WAGNER <expertise>

Abstract: In this talk an ongoing effort will be described to apply persistent homology in the area of text data mining. Persistent homology is the main tool of topological data analysis. In essence, it allows to robustly describe the shape of a data set, and compare the shapes of different data sets.
First, persistent homology will be explained, emphasizing its intuitive side.
Then, it will be demonstrated how persistent homology can be applied in the context of analyzing sets of text documents. Using the vector space model interpretation, each document becomes a point in a high-dimensional space, and it is intuitive to ask about the shape of such a point cloud. It wil be discussed, how this information can be used for knowledge discovery. Finally, an algorithmic aspect is emphasized, which is crucial if industrial applications are to be tackled.

Biography: Hubert Wagner is a computer scientist, currently working as a Postdoc at the Institute of Science and Technology Austria (IST-Austria) at the Edelsbrunner Group. Having worked as a software engineer, he moved towards science and obtained a PhD degree in 2014 from the Jagiellonian University in Krakow, Poland. Hubert is interested in the application of computational geometry and topology and related algorithmic questions. He is convinced that tools such as persistent homology may offer novel and robust solutions to many problems he encountered as an engineer, including e.g. problems in text mining. This line of his research was supported by a Google Research Grant from 2011 to 2012 (with Prof. Marian Mrozek and Dr. Paweł Dłotko) and is now continued within the Topological Complex Systems (TOPOSYS) grant. Efficient algorithms and their implementations are an important part of his work.

More Information:

Topological Analysis for Text Data

Topological Analysis for Text Data

January, 27, 2015, Seminar Talk by Barbara Di Fabio

Title: Geometric-topological tools for shape description

Lecturer: Barbara DI FABIO

Abstract: In shape comparison a widely used scheme is to measure the dissimilarity between signatures associated with each shape rather than matching shapes. In this context, computational topology plays an important role, offering a series of techniques and measures with an extremely high abstraction power. Persistent homology and Reeb graphs provide signatures able to describe shapes from topological and geometrical perspectives, being approaches grounding in the classical Morse Theory. The common idea underlying these methods, indeed, is to perform a topological exploration of the shape according to some quantitative geometric properties provided by a real-valued function defined on the shape and chosen to extract shape features. This seminar  provides an overview of these shape descriptors with related comparison methods, their main properties and drawbacks, some of the main theoretical and experimental results, recent developments, open issues and future perspectives.

Biography: Barbara DI FABIO is born in Lanciano (Italy) in 1977. In 2004, she graduated cum laude and, in 2009, received her Ph.D. degree in Mathematics at the University of Bologna with a work on the enhancement of geometrical tools for pattern recognition. Since then, she has been post-doctoral fellow at the excellence centre ARCES ”E. De Castro” (University of Bologna) and at the Department of Mathematics (Prof. Massimo FERRI, University of Bologna). Barbaras main research interests are focused on computational geometry and topology and include problems of shape analysis and understanding with related applications in computer vision, computer graphics and pattern recognition – highly relevant for machine learning and knowledge discovery. She attended several postgraduate schools and workshops, participated in and was author of several communications in national and international scientific conferences. She is author of 9 peer-reviewed papers, 5 proceedings and 1 preprint. She is a referee for several international journals. Since 2005 she has been teaching in undergraduate courses in Engineering and Economics, University of Bologna. At present, supported by an ESF exchange visit grant, she is working with Professor Neza Mramor Kosta at the Faculty of Computer and Information Science, University of Ljubljana.

More Information:

Geometric-topological tools for shape description

Geometric-topological tools for shape description


Merry Christmas and a Happy 2015 from the Holzinger Group

Merry Christmas and a Happy 2015 from the Holzinger Group

Merry Christmas and a Happy 2015 from the Holzinger Group

Open Postdoc Position in interactive Machine Learning with complex biomedical data

A postdoc position in “knowledge discovery and interactive machine learning with complex biomedical data sets” is available immediately at the Holzinger Group ( in Graz, Austria. The postdoc will be financed for four years, with an option to continue for another four years by the newly formed CBmed – Center for Biomarker Discovery and supported by the PhD school “Biomarker discovery”, which is starting with October, 1, 2015.

The challenge: Worldwide there is raising interest in biomarker discovery as an important step towards P4-medicine. The data results from various sources in different structural dimensions, and a systematic and comprehensive exploration of all these data provides a mechanism for data driven hypotheses generation. A grand challenge is to make sense of this complex data sets by applying machine learning algorithms based on the “human-in-the-loop” concept, which is of emerging interest for the international research community.

The applicant should:
1) hold a PhD in machine learning, data mining, knowledge discovery or related area of modern data science;
2) have a strong research record, documented by publications at first-tier related conferences and journals;
3) having interest in advanced methodological approaches and enjoy working in a young research group following the motto
“Science is to test crazy ideas, engineering is to bring these ideas into Business”

The successful candidate shall take an active role in the further development of our research group. Communication skills and fluency in English are required.
Conditions of employment: This post-doctoral position is provided for four years with an option for another four years. The starting date is flexible; there is no fixed deadline, so applications will be considered until the position is filled with the optimal candidate.

Application procedure: Formal applications should include:
1) A scientific curriculum vitae, including a full list of publications;
2) A statement of research interests with an outlook for the coming 4 (8) years;
3) Contact details of three reference persons.

Apply by sending your application as one single PDF document, indicating Postdoc HCI-KDD in the header directly to
Prof.Dr. Andreas HOLZINGER via e-Mail:

About the group: The Holzinger Group works consistently on a synergistic combination of methodologies and approaches of two areas that offer ideal conditions towards unraveling these problems: Human-Computer Interaction (HCI) and Knowledge Discovery/Data Mining (KDD), with the goal of supporting human intelligence with machine learning – human-in-the-loop – to discover novel, previously unknown insights into the data.
For more details please refer to:

Note: The language both of the Holzinger group and the language of the PhD school is English.

Keywords: interactive machine learning, knowledge discovery, data mining, human-in-the-loop, biomedical informatics

Nature repository for openly accessible scientific data from all disciplines

Nature Scientific Data is a recently launched open-access, online-only journal for openly accessible scientific data from all disciplines. The articles are called Data Descriptors, and combine traditional narrative content with curated descriptions of research data to support reproducibility, as this may accelerate scientific discovery, see:

Goudiaby, V., Zuidema, P. A. & Mohren, G. M. J. 2014. Data storage: Overcome hurdles to global databases. Nature, 511, (7510), 410-410.

Editorial to Nature Volume 515, Issue 7527 > Data-access practices strengthened

CfP BIH & AMT, London, 30.8.-2.9.2015 > LNAI Call due to April, 5, 2015

The 2015 International Conference on Brain Informatics and Health (BIH’15)

Informatics for Brain Science, Human Behavior and Brain Health

August 30 – September 2, 2015, London, UK



*** KEYNOTE SPEAKERS (Confirmed) ***

Allan Jones (Allen Institute for Brain Science, USA)
Henry Markram (EPFL, Switzerland)
David Van Essen (Washington University School of Medicine, USA)


Brain research is rapidly advancing with the application of big data technology to neuroscience as can be seen in major international initiatives in the US, Europe and Asia. BIH’15 reflects that brain informatics has emerged as a distinct field and crosses the disciplines of neuroscience, cognitive science, computer science, signal processing, and neuroimaging technologies as well as data science. Following the success of past conferences in this series, BIH’15 will take place at Imperial College London, in UK, gathers the researchers from major international brain research projects, and plans an industrial exhibition.

BIH’15 draws special attention to informatics for brain science, human behavior and brain health. BIH’15 will address big data approaches to both the brain and behaviour, with a strong emphasis on emerging trends of big data analysis and management technology for BI, active media technology in behavior learning, and real-world applications for brain and mental health.

BIH’15 welcomes paper submissions (full paper and abstract submissions). Both research and application papers are solicited. All submitted papers will be reviewed on the basis of technical quality, relevance, significance and clarity. Accepted full papers will be included in the proceedings by Springer LNCS/LNAI.

Tutorial, Satellite symposium and Special-Session proposals and Industry/Demo-Track papers are also welcome.


Satellite symposium proposal submission: March 10, 2015
Notification of satellite symposium acceptance: March 30, 2015
Submission of full papers: April 5, 2015
Submission of abstracts: May 20, 2015
Submission of satellite symposium papers: May 20, 2015
Notification of full paper acceptance: May 25, 2015
Notification of abstract acceptance: June 10, 2015
Notification of satellite symposium paper acceptance: June 10, 2015
Tutorial proposal submission: May 15, 2015
Satellite symposiums: August 30, 2015
Main conference: August 31-September 2, 2015


TYPE-I (Full Paper Submissions; Submission Deadline: April 5, 2015):

Papers need to have up to 10 pages in LNCS format:
All full length papers accepted (and all special sessions’ full
length papers) will be published by Springer as a volume of
the series of LNCS/LNAI.

TYPE-II (Abstract Submissions; Submission Deadline: May 20, 2015):

Abstracts have a word limit of 500 words. Experimental research is
particularly welcome. Accepted abstract submissions will be included
in the conference program, and will be published as a single,
collective proceedings volume.

Title: Include in the title of the abstract all words critical for a
subject index. Write your title in sentence case (first letter is
capitalized; remaining letters are lower case). Do not bold or
italicize your full title.

Author: List all authors who contributed to the work discussed in the
abstract. The presenting author must be listed in the first author
slot of the list. Be prepared to submit contact information as well as
conflict of interest information for each author listed.

Abstract: Enter the body of the abstract and attach any applicable
graphic files or tables here. Do not re-enter the title, author,
support, or other information that is collected in other steps of the
submission form.

Presentation Preference: Authors may select from three presentation
formats when submitting an abstract: “poster only,”, “talk
preferred” or “no preference.” The “talk preferred” selection
indicates that you would like to give a talk, but will accept a poster
format if necessary. Marking “poster only” indicates that you would
not like to be considered for an oral-presentation session. Selecting
“no preference” indicates the author’s willingness to be placed in the
best format for the program.

Each paper or abstract requires one sponsoring attendee (i.e. someone
who registered and is attending the conference). A single attendee can
not sponsor more than two abstracts or papers.

Oral presentations will be selected from both full length papers and

*** Post-Conference Journal Publication ***

The BIH conferences have the formal ties with Brain Informatics
journal (Springer, Accepted papers
from the conference, including their Best Paper Award papers, will be
expended and revised for possible inclusion in the Brain Informatics
journal each year. It is fully sponsored and no any
article-processing fee charged for BIH authors.

Selected submissions will be considered for publication in special
issues of international journals after their papers are extended to a
full-length paper and pass a review process. More information can be
found at

*** Topics and Areas ***

Please find the topics and areas of interest of the 2015 International
Conference on Brain Informatics and Health (BIH’15)

*** AMT’15 Session ***

The advance of wearable sensor technology makes the monitoring of
human behavior and life style becomes feasible. This development gives
the active media technology a new dimension which is more closely
related to the healthcare and cognitive studies. Following the success
of past conferences in this series, AMT’15 will be jointly held with
BIH’15 as a special session.


General Chairs:
Karl Friston, University College London, UK
Yike Guo, Imperial College London, UK

Program Chairs:
Aldo Faisal, Imperial College London, UK
Sean Hill, EPFL, Switzerland
Hanchuan Peng, Allen Institute for Brain Science, US

Workshop/Special-Session Chairs:
Andreas Holzinger, Medical University Graz, Austria
Zhisheng Huang, Vrije University of Amsterdam, Netherlands
David Powers, Flinders University of South Australia, Australia

Publicity Chairs:
Jessica Turner, Georgia State University, US
Juan D. Velasquez, University of Chile, Chile
Yi Zeng, Institute of Automation, CAS, China

Local Organizing Chairs:
Thomas Henis, Imperial College London, UK
Kai Sun, Imperial College London, UK
Chao Wu, Imperial College London, UK

Exhibition/Sponsorship Chair:
Caroline Li, University Kent, UK

Steering Committee Co-Chairs:
Ning Zhong, Maebashi Institute of Technology, Japan
Jiming Liu, Hong Kong Baptist University, Hong Kong

*** Contact Information ***

Chao Wu, Imperial College London, UK

Aldo Faisal, Imperial College London, UK

Banff Meeting approved

We just received the message from the Scientific Director of the BIRS – Banff International Research Station for Mathematical Discovery, that our 7th International Meeting of the expert group HCI-KDD “Advances in interactive Knowledge Discovery and Data Mining in complex and big data sets” (BIRS Approval Nr. 15w2181) has been approved by the scientific committee for the dates July 24 – July 26, 2015. So, the Holzinger Group has now the possibility of being at the holy grail of computational mathematics 🙂

NB.: The area around Banff is the world’s most dense area of paleontological findings.
Banff is 78 miles west of Calgary at a hight of approx. 5000 ft. with subarctic climate,
temperatures during our meeting end of July are the highest of the year, in average 21 degrees Celsius.
Morning temperatures can go as low as 7 degrees Celsius – so pullovers are recommended 🙂
Those who are afraid of bears should not attend this meeting, as the Research Center is directly located within deep forest and in the morning Grizzly bears are usually foraging the trash bins.

The papers will be collected in a Springer Volume Lecture Notes in Artificial Intelligence (LNAI),
which is a more mathematical oriented topical subseries of the Lecture Notes in Computer Science (LNCS): is now

Our website appears in a new light!

We welcome our visitors to our new website. In order to keep pace with time, our website is now responsive with a fresh design.


Our digital world, full of mobile computing devices and ubiquitous sensor networks produce increasingly large, complex, high-dimensional and weakly structured data sets, and increasing volumes of unstructured information. These enormous amounts of data require novel, efficient and interactive solutions for knowledge discovery/data mining.  A synergistic combination of methodologies and approaches of two areas offer ideal conditions towards unraveling such problems: Human-Computer Interaction (HCI) and Knowledge Discovery/Data Mining (KDD), with the goal of supporting human intelligence with machine learning, putting the “human-in-the-loop “. Our goal is to interactively discover new, previously unknown insights into the data and we are passionate on extending advanced methods including time (e.g. information entropy) and space (e.g. computational topology), along with user-centered software engineering methods to create interactive software for mobile applications & content analytics techniques.

Science is to test crazy ideas – Engineering is to put these ideas into Business!