Schedule

I will introduce the AI opportunities fo Deutsche Telekom. You will have a clear picture about where we are at DT on the road of Artifical Intelligence possibilities. You can see how much homework we have to do to scale and generate business impact through AI.

Based on several questionnaires and studies we can piece together humans perception and value of the potential of Artificial Intelligence. This talk will take us through how we perceive AI in general and in particular as it addresses decision making either via augmentation, where the AI support the Human, or via replacement of the human decision maker. Some interesting gender differences will be shown in the presentation.

There is a huge amount of data collected every day. Using such a big data, recent developments in AI allow for training complicated models with millions of parameters that are shown to outperform humans in many use cases. This talk will first overview some basics of the-state-of-the-art in AI and then go over some recent applications of AI developed by Machine2Learn such as deep learning for similarity or causality extraction. The talk will further refer to privacy preserving machine learning and conclude by addressing where AI lags behind.

We often hear from successful business developers that they are good network builders and good communicators who are able to control the business and the buyer. That is why many people say that sales is an art to be born for. But what will future bring? Are good communicators and network builders determining the trends in sales indeed? Is it possible to have a more effective way of sales? How much support can algorithms and real-time data provide? How to make data driven decisions as sales managers?

Secrets, everyone has them, and someone else will always be interested in those secrets. Otherwise, they wouldn’t be so. A secret is an incredibly complex idea that is developed and constantly reevaluated in the mind. We are making thousands of them willingly or unknowingly, all the time. But for what? Why do we keep secrets? To keep us safe, to keep us protected, to have privacy. Privacy, the abstraction that is so hard to define and even more so to safeguard in today’s cyberspace. How will we protect them in 2020 among millions of data leaking every week, daily ransomware attacks and billions of insecure IoT devices? A year when – according to Gartner – we will be unable to protect 75% of our sensitive information. Will Artificial Intelligence help us with this struggle?

Not all companies need artificial intelligence immediately, but each company should consider whether it can and if so how to use algorithms. This requires pragmatic and practical strategic planning. In the presentation, we present those options and methods with which business people can now begin this process. We show the areas where machine learning and artificial intelligence technologies are developed enough to bring real efficiency gains and dispel some misconceptions about what tasks can be or can not be solved with these technologies.

The seminar discusses how machine intelligence on edge will enable a new class of machine vision enabled IoT applications. Edge sites can act as high performance computing hubs virtually extending the capabilities of mobile devices or as a real time synchronization point for a large number of heterogenous devices solving a complex problem. AI supported machine vision has huge potentials in this domain enabling low latency real time performance for multiple deep learning based functionalities. 5G connectivity and the global coverage of telecom networks further pushes the robustness and scale of these “Internet of Eyes“ solutions. In my presentation I will highlight some opportunities in this domain and give some examples in the public safety, intelligent transport system and industrial automation application areas. In the workshop we can walkthrough these opportunities in more detail.

Based on the capabilities we take from AI and ML the benefits are shown by two exemplary use cases. The first one points out how ML can be used to automate anomaly detection within Incident/Fault Management. Results are used to invoke a recovery process or raise workforce tickets for physical repairs. The second use case gives an example how performance impact due to customer behavior can be analyzed and even predicted by using AI mechanisms. This information can be taken to the application level to allow compensating actions.

Deutsche Bahn ordered this project from T-Systems to improve train delay prognoses using modern machine learning techniques. Up to now, Deutsche Bahn used a purely statistical approach to predict train delay from historical data. Their approach works well for small current delays and short prognostic horizons. For larger current delays and longer prognostic horizons, the power of machine learning needs to be leveraged to increase accuracy of the predicted additional delays incurred over the prognostic horizon. Such a machine learning based approach was designed and implemented within the project to be presented.

The primary aim of the MediaBubble project is to help online news readers extend their filter bubbles in order to encourage them to make analytical decisions instead of being influenced. Filter bubble is a well known phenomenon since the popularization of recommender systems, especially in the case of the online and the social media. These software systems involve personalization techniques, basically serving the taste of the readers. The application of such -  typically artificial intelligence based - solutions restrict the reader to a specific segment of the content. Looking at the problem from a more traditional viewpoint, filter bubbles even existed before the appearance of online media, in the case of classical journals, as people tended to consume those news products which met their taste. With the popularization of the social media, the societal impact of the filter bubble phenomenon is unquestionable.
MediaBubble addresses the mentioned problem by recommending news articles to the reader in the the same topic but from an alternative aspect. The concrete software product is an artificial intelligence based newsreader application and a browser toolbar sitting on the top of semantically indexed news articles appearing in the online media. The project strongly relies on semantic technology, to be more exact, compositional semantic models. Semantic modeling is involved to determine the topic of the textual content. From the technological aspect, the MediaBubble project can also be treated as a content-based recommender system applied on the news articles appearing in Hungarian mainstream media.

Spotify, Netflix, Amazon, YouTube: what is the common element? Their business model is greatly relying on advance analytics, based on artificial intelligence. This, however is no longer the privilege of tech giants.
Our presentation is a case study of creating, implementing and operating a cloud based real time streaming data warehouse, and a machine learning based recommendation engine on top of it.

We have been talking to computers for a while now but do they really understand what we are saying? Vanda surely does! The new multichannel self service platform developed by T-Systems is here, featuring the most recent technological advancements in the fields of speech synthesis, speech recognition and AI-assisted machine learning. By understanding spoken and written messages received via phone, chat or email, Vanda is able to process and reply to customer requests with near-human intelligence, yet without human intervention, taking customer service automation to the next level.

Autonomous driving is here, autonomous vehicles are already on the road ! But what does it exactly means, what are the next steps, what is the timescale of a bigger transition and what role AI will exactly have in this story is still somewhat unclear. In the talk we present the current status of worldwide autonomous driving development, current and possible future application of AI in the context of self-driving and outline different possinilities for future trends and progress.

The dangers of Artificial Intelligence is generally misunderstood due to its representation in popular culture.
In reality, the rapid advancements in machine learning cause researchers to constantly face moral and ethical dilemmas, and the decisions they make either consciously or unconsciously effect our everyday lives.
How would you decide if you were responsible for programming the moral code for a self-driving car or an autonomous war drone?
Our Virtual Reality experiment Training2038 (developed by Kitchen Budapest) forces the human test subjects to make these decisions themselves.