Accurate and efficient time-domain classification with adaptive spiking recurrent neural networks

September 14^th, 2021

Inspired by more detailed modeling of biological neurons, Spiking neural networks (SNNs) have been investigated both as more biologically plausible and potentially more powerful models of neural computation, and also with the aim of extracting biological neurons’ energy efficiency; the performance of such networks however has remained lacking compared to classical artificial neural networks (ANNs). Here, we demonstrate how a novel surrogate gradient combined with recurrent networks of tunable and adaptive spiking neurons yields state-of-the-art for SNNs on challenging benchmarks in the time-domain, like speech and gesture recognition. This also exceeds the performance of standard classical recurrent neural networks (RNNs) and approaches that of the best modern ANNs. As these SNNs exhibit sparse spiking, we show that theoretically they are one to three orders of magnitude more computationally efficient compared to RNNs with comparable performance.
Together, we argue that this positions SNNs as an attractive solution for AI hardware implementations.

Speaker:
Prof. Dr. Sander M. Bohté

Location:
Seminarraum IGI - Inffeldgasse 16b/I
Webex

When:
Friday, 17 September 2021 11.00 – 12.45

Organizer:
IGI TU Graz

AI-DA Challenge - Award ceremony

August 09^th, 2021

Big thanks to the 8 teams of TU Graz for their engagement in the AI-DA Challenge (AI-Dependability Assessment) of Siemens Mobility in the last weeks.
“What was so interesting about this task is that it wasn’t only about reaching as high accuracy as possible but also required us to consider real-world safety aspects. For example, we made trade-offs between different kinds of errors and gave bounds to our misclassification rates, so the model could actually be used in a safety-critical environment,” Patrick Deutschmann and his team partner Lukas Timpl stated.

Due to this strong participation of our TU Graz, the jury decided at the award ceremony on 16 July to increase the prize money and to award three out of eight teams from TU Graz with the Runners-up prize of € 1000 for each team:  

• Adna Ribo | Markus Schiffermueller | Lucas Weitzendorf
• Felix Muetter
• Patrick Deutschmann | Lukas Timpl

Furthermore, the CKI office at TU Graz and Siemens Mobility Graz would invite all 8 teams to pizza and beer at the beginning of the next semester, please contact christine.schichlernoSpam@tugraz.at .

DIY tabletop magnetic resonance scanner

July 19^th, 2021

The student initiative INSIGHT, operating at our Institute of Medical Engineering, introduces itself with an exciting project: Self-construction of a magnetic resonance tomograph for the desk. They broadcast live the assembly of the core piece, namely the magnet system for the small scanner. This is an assembly of iron yoke, pole pieces and two extremely strong neodymium permanent magnets. The assembly is relatively spectacular, as forces of up to one and a half tons act on the parts in the meantime and therefore great caution, some securing and some courage are required during assembly. The team of four students has been eagerly awaiting this moment for a year and wants to share the exciting scenes with all viewers. Live on YouTube: July 22^th - 3pm

The World's First Digital Model of a Cancer Cell

July 06^th, 2021

Computer models have been standard tools in basic biomedical research for many years. However, around 70 years after the first publication of an ion current model of a nerve cell by Hodgkin & Huxley in 1952, researchers at Graz University of Technology (TU Graz), in collaboration with the Medical University of Graz and the Memorial Sloan Kettering Cancer Center in New York, have finally succeeded in developing the world's first cancer cell model, thus launching "an essential tool for modern cancer research and drug development," reports a delighted Christian Baumgartner. The head of the Institute of Health Care Engineering with European Testing Center of Medical Devices at TU Graz is senior author of the publication in which the digital model is presented in the journal PLoS Computational Biology.

One award after the other for Gerhard Holzapfel

June 09^th, 2021

Biomechanist Gerhard Holzapfel can be pleased about two new science awards. His computer simulations are intended to facilitate medical prognoses.
Holzapfel can be pleased about another award. After the William Prager Medal (we reported in January), he received the next international medal: the Koiter Medal, awarded for special achievements in research and science. In addition to American scientists from renowned U.S. universities, Holzapfel is an Austrian among the list of award winners. "This is already a great honor," says the 60-year-old professor, who was already awarded the Erwin Schrödinger Prize ten years ago for his life's work.

Holzapfel's focus: the calculation of mechanical properties for vessel walls, such as an aorta or artery. He has developed his own model for this purpose. "You can map mechanical processes that happen in our body on the computer. The finer the model, the more accurate the prediction. On the virtual patient, you can thus predict what a medical intervention will be like," he explained to Kleine Zeitung. "We can make simulations that are very close to reality. That's where technology has advanced insanely."

Holzapfel began his career as a young postdoc at Stanford. "In America, there was already a lot of research going on in the field in the early 1990s, so I was there at the right time." In the meantime, he himself is the contact person for 200 students and drives countless projects forward. The TU Graz lead project "Aortic Dissection" has just been extended. This involves working on computer simulations of aortic dissections. An aortic dissection is a splitting of the wall layers of the aorta. This condition can be life-threatening. The aim of the TU project is to use algorithms and models to support diagnosis and treatment at an early stage.

Data register and data protection very well compatible

May 25^th, 2021

New encryption methods ensure the pooling of sensitive income and health data at a very high level of data protection. It is precisely for such cases that research has developed technologies that are ripe for practical use.

In the course of the current discussion about a planned data registry for times of crisis such as pandemics, which is viewed critically by data protection experts, cybersecurity expert Christian Rechberger of Graz University of Technology takes notice: "Pandemic management using merged income and health data is very much possible with the required level of data protection. Together with international colleagues, we have developed new cryptographic methods since the beginning of the pandemic that will be used for such applications, among others."

Rechberger sees the crisis database as a classic case for the first use of the new technology. This is based on so-called homomorphic decryption: it allows confidential data to be processed and interleaved without first having to be decrypted. "This allows data registers to be implemented while maintaining privacy and protecting sensitive, personal data. No data source has to copy the data elsewhere, registers do not have to be merged centrally. The data remains local, and the calculations are encrypted. Only the result of the respective study is then disclosed."

"In the future, we will have to adjust consumption to supply, not the other way around"

May 22^th, 2021

Responsibility for the climate and the environment should not be assigned to the individual, emphasizes Gerald Schweiger. "This requires technological solutions at a systemic level." The fact that the 35-year-old, who conducts research on intelligent energy systems at the Graz University of Technology (TU), sometimes digresses into socially critical and philosophical considerations may be due to his education: "After HTL, I actually thought that technology wasn't for me after all, and initially followed up with studies in philosophy and social sciences. In the end, however, I only added another engineering degree and one in environmental systems science."
The result is not only a total of four academic degrees in the passport around his name in his passport (the fifth will follow soon), but also holistic demands such as these demands like this: "One less plastic bag is not a solution. We have to rethink entire systems and equip them with intelligence. Especially in the area of sustainability new insights have an enormous impact."

Gerald Schweiger (35) is concerned with energy systems of the future and their optimization with the help of physical models and artificial intelligence. He teaches at Graz University of Technology, where he heads the "Intelligent Energy Systems" research group, as well as at two Styrian universities of applied sciences. In addition, Schweiger leads several national and international projects on the topic.

Hellomath: TU Graz student founded learning platform for math school-leaving exam

April 06^th, 2021

Calculating probabilities, uncovering the value of X, analyzing graphs - math is a nightmare for some students.However, for students Maximilian Mallweger and Matthias Hülse rmath is passion. "Yes, it exists!" they write on their website Hellomath - the self-founded learning portal where students can prepare for the centralized math Matura.

The tasks that can be solved on Hellomath are taken from the official question pool of the Ministry and also from older maturity exams that are freely available. Maximilian has also created some tasks himself, based on basic competency schemes. Already 100 active users use the learning portal regularly.

Mobile app generates data for the energy management of the future

March 24^th, 2021

As part of the GameOpSys research project, people were asked to document their own energy consumption using a study app. Researchers at TU Graz use this data for modeling and optimizing future energy systems.

The research group "Intelligente Energiesysteme und Cyber-Physical Systems" at the Institute of Software Technology at TU Graz is working on methods to make energy systems more efficient and intelligent, while also taking into account the interaction between humans and technology. In collaboration with simulation expert Niki Popper and his company dwh, the Graz-based research group led by Gerald Schweiger focused on necessary technologies for future user-centered energy services. For this purpose, a mobile app was developed that is intended to motivate users with game-like elements (gamification) to collect data on their own energy consumption (electricity, heating and cooling).

"Software development is much more than just programming"

March 16, 2021

Denis Helic explains why Software development is much more than just programming. Software development does not only include programming skills, but also requires management and team leadership skills. User experience is also becoming increasingly important. The curricula are continuously updated and expanded via electives in order to respond to new developments. That's why the TU has offered an English-language bachelor's and master's degree in software engineering and management since 2005, where 20 percent consists management topics. Helic explains that the profession demands a range of soft skills, "from project management to teamwork to communication skills and much more." In addition, he says, today's students are tomorrow's company founders, designing and marketing their own products in the software field. This refutes the stereotype of the computer scientist tapping away in a lonely chamber.

New approach found for energy-efficient AI applications

March 11, 2021
 

Computer scientist Wolfgang Maass and his doctoral student Christoph Stöckl have found a new design method for artificial neural networks. These networks get by with extremely few signals. Similar to Morse code, they also give meaning to the pauses between the signals. Maass and Stöckl were inspired by the way the human brain works. The human brain processes several trillion calculations per second, but only needs about 20 watts to do so. This low energy consumption is made possible by inter-neuronal communication using very simple electrical impulses, so-called spikes. This research work is anchored at TU Graz in the Fields of Expertise "Human and Biotechnology" and "Information, Communication & Computing", two of five areas of strength at TU Graz.

Styrian robotics specialist is bought up by US group Verizon

February 16, 2021
 

Incubed IT, a start-up founded in 2011 by seven graduates of TU Graz, is being bought by the US company Verizon (2020: $128.3 billion sales). The founders got together back in 2004 when they first formed a TU team for the Robocup. In 2011, the seven graduates of TU Graz (Julia Arnus, Stephan Gspandl, Reinhard Günther, Andreas Hechenblaickner, Michael Reip, Mate Wolfram and Christoph Zehentner) founded Incubed IT. They recently achieved annual sales in the low millions, and have already received many awards, including the "Phoenix" founder prize, the "Robotics Award" at the Hanover Trade Fair, and the Fast Forward Award.

More privacy for using WhatsApp and Co.

February 15, 2021


 

Two of our cryptography experts Daniel Kales and Christian Rechberger developed, together with their colleagues at Darmstadt University of Technology, privacy-protecting security software for mobile messenger services. "ContactGuard" was awarded the 8th German IT Security Award 2020. The researchers intend to use the prize money of 60,000 euros to further develop the security software to market maturity. Stay ahead!

Promotion Award 2021 of the Forum Technology and Society

Febuary 10, 2021
 

You completed your disseration or master thesis between February 1, 2020 and January 31, 2021? Briefly argue (3000 characters) why your dissertation or master thesis is or can be important for society. Send your arguments and convince the jury. The prize money totals 12,500 euros this year!

Martin Ebner is honored as "Styrian of the day"

02. Febuary 2021

Martin Ebner was portrayed in the Kleine Zeitung because of his development IMooX, which is an Austria-wide platform where digital teaching content of higher education institutions is managed. Ebner has been involved with digital forms of teaching and learning for more than 20 years now.

Cardiovascular diseases: New "Digital Twin" computer model improves therapy

January 22, 2021

Although treatment options are constantly improving, cardiovascular diseases are still one of the most frequent causes of death in Europe. The success of the treatment varies from patient to patient and depends on the individual pathology. Gernot Plank (Med Uni Graz) has now developed a computer model together with the mathematicians Gundolf Haase and Kristian Bredies from the University of Graz and computer scientist Thomas Pock from the Institute of Computer Graphics and Vision at Graz University of Technology, respectively, with which doctors can pre-simulate the optimal therapy and dramatically improve the success of treatment.

Gerhard A. Holzapfel receives William Prager Medal 2021

January 18, 2021
 

Prof. Holzapfel receives the prestigious prize for his outstanding contributions to the inspirational application of solid mechanics in the development of continuum theory, computational methods, simulation and experiment in the biomechanics of soft biological materials. Congratulations!

A1 Telekom Austria acquires majority stake in analytics start-up Invenium

January 14, 2021

By acquiring a majority stake in the Graz University of Technology spin-off and the Know-Center, A1 is expanding the area of movement data analyses and offering "A1 Mobility Insights", a commercial product based on the TÜV-approved and data protection-compliant platform.