Dienstag, 20. Juni 2017


Eintreffen, Registrierung



Bruno Kaiser, Präsident SI-SE
Ivo Ruckstuhl, Organisator Fachtagung


Developing Artificial Intelligence Systems: Challenges and Best Practices

Prof. Jana Koehler, Hochschule Luzern

Implementing AI systems in real-world applications poses a number of challenges in software engineering such as dealing with complex dependencies, distributed architectures, and computation-intensive algorithms. In this talk, I summarize my lessons learned and insights from several AI projects in industrial applications. I discuss application scenarios and potential risks of key AI technologies and how extensive quality assurance and testing can or cannot address these risks.


Uncovering the Unreasonable Success of Deep Learning

Gabriel Krummenacher, Data Scientist, Zühlke

The renaissance of artificial neural networks has led to breakthrough results in different fields of machine learning: from detecting objects in images and videos to winning poker against humans, the success of deep learning spans a broad range of domains. But how do modern neural networks work? Why are they so successful? What can we expect in the future? In this talk, we will look behind the scenes of deep learning to answer these questions.




Variational AutoEncoder Generative Adversarial Networks

Gilbert Francois Duivesteijn, Deep Impact

In recent years, supervised learning with Convolutional Neural Networks (CNN) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In 2014 Ian Goodfellow et al introduced a new framework for estimating generative models via adversarial networks, in which two models are trained simultaneously. The Generator is trying to create things, which look like the training data. The discriminator has to guess whether what is given is a real training example or the output of the Generator. This type of framework is notorious hard to train. During training, the Generator and Discriminator have to strike a balance so that they both get stronger. If the balance is disturbed, the entire network can end up learning something not useful at all. By combining a Variational Auto-Encoder with a Generative Adversarial Network we can use learned feature representations in the GAN discriminator as basis for the VAE reconstruction objective. It speeds up the training process dramatically and results in a better output. Moreover, the method learns an embedding in which high-level abstract visual features (e.g. wearing glasses, smiling, etc.) can be modified using simple arithmetic, which will be demonstrated during the presentation.


Generalversammlung 2017 SI-SE

Bruno Kaiser, Präsident SI-SE




Artificial Intelligence as a Driver for Digitalization

Dr. Felix von Reischach, Head of Artificial Intelligence and Machine Learning Group, Swisscom

Swisscom has the mission to support its enterprise customers in their digitalization journey. In this endeavour, Artificial Intelligence plays a central role. The new paradigms of programming, enabling computers to act like humans provides a vast potential to businesses of all sorts. Swisscom is on top of this trend and has built a competence center consisting of world-class product people, engineers, designers, and last but not least, data scientists. In co-innovation with the largest Swiss enterprises, the group identifies customer pains that require automatic understanding of human language and unstructured text and solves them using the latest technology. The talk gives an overview of why Artificial Intelligence has become so important and presents some of the lighthouse projects Swisscom is working on.


Fraud Detection across Financal Transactions using Machine Learning

Adrian Berger, Managing Partner Finance Solutions, Ergon

Die Angriffe auf den elektronischen Zahlungsverkehr werden immer gewiefter, entsprechend baucht es stets auch intelligentere Schutzmassnahmen. Dieser Gedanke stand am Anfang des Machine Learning Entwicklungsprojekts von Ergon zur Entdeckung und Bekämpfung von Betrugsfällen (Fraud Detection) im Zahlungsverkehr. Im vorliegenden Referat wird Adrian Berger die im Verlaufe des Projektes gewonnenen Erkenntnisse und die erzielten Ergebnisse erläutern und klären, inwiefern das selbstlernende System besser ist als die bekannten traditionellen Sicherheitsmassnahmen.




Conversational ChatBots, AI, and Machine Learning

Mike Wise, Incubation Architect, Microsoft

There has been a huge surge in interest in both ChatBots and AI recently, and one could be forgiven if they expect human-level intelligent chatbots to suddenly appear around us. In fact one could argue that the marketing we are seeing for Cortana, Watson, Sira, and Alexa all make these claims implicitly. But that is marketing, what is really behind it all? In this talk we will discuss the actual capabilities that ML can give to ChatBots today. Additionally we will also talk about what is missing for true human-level capabilities, and how chatbot-related ML and AI research could lead there.


AlphaGo from a Go Player and an AI Perspective

Prof. Thomas Koller, Hochschule Luzern

The computer program AlphaGo from Google DeepMind has recently defeated one of the strongest Go players, Lee Sedol, to the astonishment of both the go and the AI community. In the talk, we will have a look at the concepts and goals in the game of Go and why it was believed to be such a hard problem to solve. We will discuss AlphaGo’s approach and how it differentiates from previous attempts and look at some exciting moments in the games. Does AlphaGo play like a human? Does it make mistakes? Can we learn from the computer? Does it understand Go concepts? A look at the reactions from the Go community and an outlook on the future of AlphaGo rounds up the talk.



Bruno Kaiser, Vorstand SI-SE
Ivo Ruckstuhl, Organisator Fachtagung


Ende der Tagung