Lunchtime Seminar
Archive Summer Semester 2018
Reconstruction and Visualization of Point Cloud Data Using the Point Distribution Tensor
Lecturer:
Marcel Ritter
Researcher at IGS group, University of Innsbruck
Date: Thursday, 28th of June 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
With the uprising use of sensor technology, such as light detection and ranging (LiDAR) or depth cameras, point clouds (PC) have become a widely used description for geometry. Starting from such observational data, and including particle data stemming from numerical simulations, a second order tensor is developed describing the geometric distribution of points within a point neighborhood. PC visualizations are enhanced by using the tensor, its computational optimization is investigated, and its application for the reconstruction of lines within PC data is analyzed. A multi-scale (MS) approach on a neighborhood description based on shape factors of the tensor is presented and applied in a reduced geometric reconstruction environment for lines. The MS shape factors are used for visualization and analysis and, finally, to automatically adapt to local properties in noisy PCs in the line reconstruction context.
On the Borders of Edge Computing and IoT
Lecturer:
Marjan Gushev
Professor at the Faculty of Computer Science and Engineering, University Sts. Cyril and Methodius - Skopje
Date: Thursday, 21st of June 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Modern mobile and wireless ubiquitous solutions are actually IoT application, brought by the advances of the technology in cloud-based systems. Edge computing idea to bring the computing closer to the user can be realized by several computer architectures, including, cloudlets, when the intermediate server is owned by an IT provider, mobile edge computing and fog computing, when the intermediate server is owned by the mobile operator, and recently the idea of dew computing appears as a solution that brings the computing even closer to the user.
This presentation focuses on different architectural approaches for IoT solutions explaining details on the organizational overview of micro services, server architectures, etc. The intermediate and dew computing layers are another way of solving the streaming data problem in the overall cloud-based architecture for IoT solutions. This needs to be analyzed by explanation of specific goals and requirements are specified to extract the differences along with a discussion of benefits, advantages and disadvantages.
To describe the overall architecture and organization one needs to analyze details on processing and communication needs in order to collect, process and/or offload data from the IoT sensors and devices. A case study of an ECG wearable sensor and monitoring system will reveal details on how these architectures can be implemented.
Visual perception: can a computational model inform the design of virtual reality systems?
Lecturer:
Fabio Solari
Ass.-Professor at the Department of Informatics, Bioengineering, Robotics and Systems Engineering, University of Genoa
Date: Thursday, 14th of June 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Any agent acting in a real-world scenario should perform the analysis and interpretation of visual information in a fast and reliable way. Here, a neuromimetic architecture that models the dorsal visual pathway from the acquisition stage to scene analysis is presented. The front-end layer (retina) performs a space-variant image acquisition and projects to a second layer (V1-MT) that estimates optic flow (and disparity) through a distributed population of energy neurons. The final layer (MST) performs scene interpretation that is based on a first order analysis of visual features.
The computational model is able to face real-world robotics tasks and to reproduce psychophysical data. On the one hand, such a computational model can allow an artificial agent to analyze and act in a real-world scenario, and on the other hand, can predict how users will perceive augmented and virtual environments.
A Time-Predictable Fog-Integrated Cloud Framework: One Step Forward in the Deployment of a Smart Factory
Lecturer:
Hamid Faragardi
Researcher at DPS group, University of Innsbruck
Date: Thursday, 7th of June 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
The proliferation of cyber-physical systems, the emerging of IoT, and cloud computing introduce the fourth revolution of industrialization, known as Smart Factory. The key feature of a smart factory is to implement a flexible and reconfigurable manufacturing system. In this talk, I will concentrate on cloud computing as one of the principal building blocks of a smart factory to provide a huge data storage space and a highly scalable computational capacity.
The cloud computing system used in a smart factory should be time-predictable to be able to satisfy hard real-time requirements of various applications existing in manufacturing systems. Interleaving an intermediate computing layer – called fog – between the factory and the cloud data center is a promising solution to deal with latency requirements of hard real-time applications. I will introduce a time-predictable cloud framework which is able to satisfy end-to-end latency requirements in a smart factory. To propose such an industrial cloud framework, we not only use existing real-time technologies such as Industrial Ethernet and the Real-time XEN hypervisor, but we also discuss unaddressed challenges. Among the unaddressed challenges, the partitioning of a given workload between the fog and the cloud is targeted. Addressing the partitioning problem not only provides a resource provisioning mechanism, but it also gives us a prominent design decision specifying how much computing resource is required to develop the fog platform, and how large should the minimum communication bandwidth be between the fog and the cloud data center.
Pricing Anonymity, with a Briefing on the GDPR for Techies
Lecturer:
Rainer Böhme
Professor at SEC group, University of Innsbruck
Date: Thursday, 24th of May 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
I summarize what computer scientists can do to help navigating the jungle of “known unknowns” in the transition to the EU’s renewed data protection regime (to become effective the day after the talk). I recall anonymity middle-layers as building blocks of many technical solutions. They establish unlinkability of actions on the application layer even if lower layers depend on observable identifiers (e.g., network addresses). And I present a recent research result that gives a pricing principle for practically existing markets for anonymity. It applies cooperative game theory and uses Shapley value, an established solution concept. The process has a novel property in that one party pays for anonymity but anonymity is created for many parties.
A brief introduction to Quantum Computing
Lecturer:
Barbara Kraus
Quantum Information and Computing, University of Innsbruck
Date: Thursday, 17th of May 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
In this talk she will first explain the difference between classical and quantum information. With the help of some examples she will then show that a quantum computer can outperform its classical counterpart. Finally, she will briefly explain how certain quantum computations can be compressed into an exponentionally smaller quantum computer.
Parallel Computing in the era of the Cloud and Heterogeneous Computing
Lecturer:
Rizos Sakellariou
School of Computer Science, University of Manchester
Date: Thursday, 19th of April 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Traditionally, the objective of parallel computing has been to minimize execution time. As the complexity and the costs associated with modern execution platforms and infrastructures grow, parallel execution time cannot be viewed as a single objective to achieve at any cost. Instead, with such execution platforms consuming large amounts of energy, one needs to assess improvements in execution time against other types of cost. Cloud computing platforms, which are often used to execute parallel applications, typically follow a resource-on-demand paradigm, where users can pay for what resources they need. However, the underlying infrastructures suffer from increasing complexity which is partly masked by having users pay, sometimes for more than they need.
In this respect, the talk will motivate the need to address efficiently the issues related to the concurrent use of multiple (and often heterogeneous) resources offered by Cloud providers by capturing these issues as some form of a multi-objective optimization problem, which requires a good understanding and appreciation of different trade-offs. The talk will make this argument by presenting experience and research on planning the parallel execution of scientific workflow applications on the Cloud in a way that tries to strike a balance between different trade-offs such as execution time, energy consumption and cost. Algorithms and techniques, experimental results and ongoing research will be presented.
Evaluation of Deep Learning in abstract classification
Lecturer:
Antonio Rodriguez Sanchez
Researcher at IIS group, University of Innsbruck
Date: Thursday, 12th of April 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Convolutional Neural Networks have become state of the art methods for image classification over the last couple of years. By now they perform better than human subjects on many of the image classification datasets. Most of these datasets are based on the notion of concrete classes (i.e. images are classified by the type of object in the image), that is, pattern recognition. In this talk I will introduce different image classification tasks, using abstract classes, which are to solve for humans, but variations of it are challenging for CNNs. The classification performance of popular CNN architectures is evaluated on these tasks and variations of the datasets that might be interesting for further research are identified.
Teaching robot motions via Multi-Task Gaussian Process Movement Primitives
Lecturer:
Athanasios Polydoros
Researcher at IIS group, University of Innsbruck
Date: Thursday, 22rd of March 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Movement primitives have been a very popular approach for learning robot movements from demonstrations. Nevertheless, they usually depend on a large number of hyper-parameters which also vary depending on the complexity of the movement. In the seminar will be presenting a novel, self-tuned and compact -- in terms of hyper-parameters -- representation of movement primitives. The method models a demonstrated movement by utilizing Multi-Task Gaussian Process (GP) models which are able to learn movements of various complexities with a minimum and fixed number of hyper-parameters. The method has been evaluated on its ability to learn and generate movements with various starting states, goal states and execution time horizons. Furthermore, it is applied on two manufacturing assembly tasks, namely snap-fitting and screwing.
Automated Test Case Generation for Java Enterprise Edition Applications
Lecturer:
Andreas Fuchs
Researcher at the Practical Computer Science Group, Department of Information Systems, University of Münster
Date: Thursday, 15th of March 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Test case generation is a labor-intensive task in software testing. As a result, an automation of that task has been the subject of intensive research effort. This talk presents an approach for a structural generation of test cases for object-oriented programs that are implemented on the Java Enterprise Edition platform and interacting with external systems such as databases or web-services. We have implemented a Symbolic Java Virtual Machine (SJVM) that executes Java bytecode symbolically. Symbolic execution is a program analysis technique for systematically exploring all feasible paths through a program. For each explored path, a path condition is the conjunction of all conditions (e.g. resulting from an if-statement) chosen along that path. A constraint solver can examine the satisfiability of a path condition, and thus determine whether this particular path can be triggered by any valid concrete input values. The difficult, yet interesting problem arises when a program interacts with external systems. The response of an external system often depends on the state of that system (e.g. a requested record in the database exists / does not exist), and hence these external states must also be captured and considered in the overall symbolic execution process.
Learning-Assisted Automated Reasoning
Lecturer:
Michael Färber
Researcher at CL group, University of Innsbruck
Date: Thursday, 8th of March 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Interactive proof assistants (ITPs) are programs that aid users writing machine-verified proofs. To improve user experience, there are so-called "hammer" systems for ITPs that aim to find proofs automatically. I show ongoing work to improve hammer systems, including machine learning techniques such as random forests and Monte Carlo Tree Search, as well as the integration of new automatic proof search methods.