Reliable Online Controlled Experiments as a new Paradigm for Automated Software Quality Assurance

Lecturer:
Florian Auer
Researcher at QE, University of Innsbruck

Date: 30.01.2020

Abstract: The systematic and continuously execution of experiments to assess the return-on-investment of potential software features became known under the term “continuous experimentation”. It removed the guesswork and expensive evaluation of future features. In addition to its current application on feature evaluation it has the potential to be usable for areas of software engineering in which traditional testing techniques are no longer sufficient.  This includes areas like machine learning algorithms or internet of things. Thus, it represents a novel software quality assurance technique. Although, experience reports and guidelines to avoid common pitfalls exist in literature, the fundamental characteristics of an experiment for quality assurance are not researched sufficiently. Therefore, we researched about the characteristics of experiments with the goal to create a model that gives an overview on all characteristics that need to be considered for a reliable online controlled experiment. It will form the basis for a more structured experiment definition, static experiment testing, software quality assurance and many more.

 

Cross-DomainText Classification

Lecturer:
Benjamin Murauer
Researcher at DBIS, University of Innsbruck

Date: 23.01.2020

Abstract:    Text classification problems are usually solved by providing training data to machine learning algorithms which than predict the outcome on test data. Ideally, the training and testing data are as similar as possible, without overlapping. Whenever this is no longer the case and training and testing data are different in some dimension, a problem is called a cross-domain text classification problem. These dimensions range from different topics to languages, and they impact on classification problems in various ways. In this talk I will present different difficulties and solutions, while focussing on the specific classification task of authorship attribution. 

 

Politeness Counts: Perceptions of Peacekeeping Robots

Lecturer:
Ohad Inbar
Sandoz

Date: 16.01.2020

Abstract:  The 'intuitive' trust people feel when encountering robots in public spaces is a key determinant of their willingness to cooperate with these robots. We conducted four experiments to study this topic in the context of peacekeeping robots. Participants viewed scenarios, presented as static images and animations, involving a robot or a human guard performing an access-control task. The guards interacted with younger and older male and female civilians, applying polite or impolite behavior. Our results show strong effects of the robot's behaviour, specifically its politeness. Age and gender of the people interacting with the robot had no significant effect on participants' impressions of the robot's attributes. There were also no differences between responses to robot and human guards. This study advances the notion that politeness is a crucial determinant of people's perception of peacekeeping robots. 

 

The Effectiveness of Fake News Flags on Social Media Platforms

Lecturer:
Katrin Figl
WI, University of Innsbruck

Date: 09.01.2020

Abstract: Social media platforms have no direct control over the production process of their content; therefore, it is crucial for platforms to respond effectively to accusations of distributing fake news. Since fake news are mainly distributed by humans and not by bots, the current quasi-experiment investigates to what extent fake news flags influence the perception of the credibility of social media posts and how fake news flags can be made more effective. Based on a data set of 240 participants using fabricated Facebook posts on political news, the results show that fake news flags have an effective warning effect, but the "implicit truth effect" was not replicated (tagging some fake news headlines did not have the unintended side effect that untagged headlines were considered more accurate). The experiment also tested the semantic priming effect of various fake news flags. For example, a semantic association of a fake news flag with stop behavior (e.g. a stop sign) increased the time spent on a social media post with fake news. Another result was that users engaged more often with social media posts (e.g. shared or liked them) on smartphones than when using a PC, although they spent less time with the posts on a smartphone than when using a PC. The results may help inform policy decisions of social media platforms and their regulators and show that social media platforms can reduce the extent to which their users spread fake news articles by using well-designed fake news flags.

 

Biologically inspired visual-auditory processing – from brain-like computation to neuromorphic algorithms

Lecturer:
Heiko Neumann
University of Ulm

Date: 05.12.2019

Abstract: A fundamental task of sensory processing is to detect and integrate feature items to group them into perceptual units segregating them from other objects and the background. A framework is discussed which explains how perceptual grouping at early as well as higher-level cognitive stages may be implemented in cortex. Different grouping mechanisms are implemented which are attuned to basic features and feature combinations and mainly evaluated along the forward sweep of stimulus processing. However, due to limitations of local feature detection mechanisms and inherent ambiguities, top-down feedback is required to deliver contextual information helping to disambiguate initial measurements. Feedback of contextual information is demonstrated to improve object recognition performance, stabilize learning of object categories, and integrate multi-sensory representations.

The canonical principles of neural computation define a set of core operations to implement above-mentioned mechanisms of perceptual and cognitive inference. These operations can be mapped, in a simplified form, onto neuromorphic platforms to emulate brain-like computation. It is demonstrated that an architecture composed of canonical circuit mechanisms can be mapped onto neuromorphic chip technology facilitating low-energy non-von Neumann computation.

Work supported by DFG & Baden-Württemberg Foundation.

 

Symbiosis of Knowledge Graphs and Conversational AI

Lecturer:
Alexander Wahler
Researcher at STI, University of Innsbruck

Date: 28.11.2019

Abstract: Already 2023 there will be more than 8bn active voice assistants world wide. Automation is the only way how to manage all these conversations. But complex conversations are still not understood due to the lack of understanding of the underlying semantics of the conversations and connecting knowledge and conversations intelligently. In this talk I will show how knowledge graphs can dynamically drive conversations.

 

Paying with personal data – a legal perspective

Lecturer:
Paulina Jo Pesch
Karlsruhe Institute of Technology

Date: 21.11.2019

Abstract: For more than a decade, consumers have enjoyed the provision of free online services, such as search engines or social media platforms. By offering these services as solely ad-funded, providers generate revenue by monetize their customers’ personal data (e.g. to increase advertising revenues through targeting, or by selling the data to third parties). This calls into question whether such services should be classified as free. Instead, users rather seem to pay with their personal data. Consequently, the legal classification and treatment of such free services undergoes a paradigm change. However, “payments with personal data” pose considerable legal challenges under both, data protection law and contract law. The talk provides an introduction to payments with personal data from a legal perspective. It gives a gentle introduction into the relevant principles of EU law targeted at computer scientists, outlines legal requirements, and points to the remaining uncertainties in the areas of data protection as well as contract law.

 

On the security and "decentrality" of federated Byzantine agreement (work in progress)

Lecturer:
Martin Florian
Weizenbaum Institute, HU Berlin

Date: 14.11.2019

Abstract: Federated Byzantine Agreement Systems (FBAS) are a fascinating new paradigm in the context of consensus protocols. Originally proposed for powering the Stellar payment network, FBAS can be thought of as something in between typical permissionless systems (like Bitcoin) and permissioned approaches for solving consensus (like classical BFT protocols). Unlike Bitcoin and the like, "miners" can't remain anonymous but must be included into quorums by peers, ideally based on their actually perceived trustworthiness. Unlike permissioned protocols, there is no need for a system-wide agreement (or decree) about which set of nodes gets to participate in consensus. Instead, every peer is free to compose its own quorums, based on its own views about who to trust.

What kinds of global structures emerge from individual configurations? Are they secure? Are they "decentralized"?

In this ongoing project, we investigate the interplay between individual quorum configuration strategies (based on, e.g., generative models for inter-peer trust) and the properties of the resulting FBAS. In the talk, we'll discuss helpful abstractions and metrics, introduce our analysis and simulation tool written in Rust, and offer a few early insights about which strategies seem to work and which don't.

 

Artificial intelligence for theorem proving in Isabelle/HOL

Lecturer:
Yutaka Nagashima
Researcher at CL, University of Innsbruck

Date: 31.10.2019

Abstract: Despite the recent progress in automatic theorem provers, proof engineers are still suffering from the lack of powerful proof automation. In this talk I report our AI-based approach for further automating theorem proving in Isabelle/HOL.

 

Robot motion planning with stable dynamical systems: Incremental learning and human-aware execution

Lecturer:
Matteo Saveriano
Researcher at IIS/DiSC, University of Innsbruck

Date: 17.10.2019

Abstract: Motion generation with stable dynamical systems (DS) is becoming a popular approach to execute point-to-point motions on real robots. Stable dynamical systems can be effectively learned from human demonstrations and are flexible enough to represent complex trajectories. Moreover, robots driven by stable DS are guaranteed to reach the desired position, and can react in real-time to external perturbations, like changes in the target position or unexpected obstacles. This talk presents some key features of motion learning and execution with stable DS with a spacial focus on human-aware motion generation and incremental learning of stable dynamical systems.

References:

[1] S. M. Khansari-Zadeh and A. Billard, “Learning control Lyapunov function to ensure stability of dynamical system-based robot reaching motions,” Robotics and Autonomous Systems, 2014.

[2] M. Saveriano, F. Hirt, and D. Lee, “Human-aware motion reshaping using dynamical systems,” Pattern Recognition Letters, 2017.

[3] M. Saveriano and D. Lee, “Incremental Skill Learning of Stable Dynamical Systems ,” IROS, 2018.

 

Self-organization of behavior in autonomous robot development

Lecturer:
Georg Martius
MPI Tübingen

Date: 10.10.2019

Abstract: I am studying the question how robots can autonomously develop skills. Considering children, it seems natural that they have their own agenda. They explore their environment in a playful way, without the necessity for somebody to tell them what to do next. With robots the situation is different. There are many methods to let robots learn to do something, but it is always about learning to do a specific task from a supervision signal. Unfortunately, these methods do not scale well to systems with many degrees of freedom, except a good prestructuring is available. The hypothesis is that if the robots first learn to use their bodies and interact with the environment in a playful way they can acquire many small skills with which they can later solve complicated tasks much quicker. In the talk I will present my steps into this direction. Starting from some general information theoretic consideration we provide robots with an own drive to do something and explore their behavioral capabilities. Technically, this is achieved by considering the sensorimotor loop as a dynamical system, whose parameters are adapted online according to a gradient ascent on an approximated information quantity. I'll show examples of simulated and real robots behaving in a self-determined way and present future directions.

 

Celerity: High-level C++ for accelerator clusters

Lecturer:
Philip Salzmann
Researcher at DPS, University of Innsbruck

Date: 03.10.2019

Abstract: In the face of ever-slowing single-thread performance growth for CPUs, the scientific and engineering communities increasingly turn to accelerator parallelization to tackle growing application workloads. Existing means of targeting distributed memory accelerator clusters impose severe programmability barriers and maintenance burdens. The Celerity programming environment seeks to enable developers to scale C++ applications to accelerator clusters with relative ease, while leveraging and extending the SYCL domain-specific embedded language. By having users provide minimal information about how data is accessed within compute kernels, Celerity automatically distributes work and data.

We introduce the Celerity C++ API as well as a prototype implementation, demonstrating that existing SYCL code can be brought to distributed memory clusters with only a small set of changes that follow established idioms. The Celerity prototype runtime implementation is shown to have comparable performance to more traditional approaches to distributed memory accelerator programming, such as MPI+OpenCL, with significantly lower implementation complexity.