Big Techday 8

In recent years, deep artificial neural networks (including recurrent ones) have won many contests in pattern recognition and machine learning. They are now widely used in industry and academia. In this talk, the field and the latest state-of-the-art results in numerous important applications will be reviewed, and it will also be discussed how artificial intelligence will fundamentally change our civilization.

Machine learning algorithms have reached human-level performance on a variety of benchmark tasks. This raises the question of whether these algorithms have also reached human-level 'understanding' of these tasks. By designing inputs specifically to confuse machine learning algorithms, we show that statistical models ranging from logistic regression to deep convolutional networks fail in predictable ways when presented with statistically unusual inputs. Fixing these specific failures allows deep models to attain unprecedented levels of accuracy, but the philosophical question of what it means to understand a task and how to build a machine that does so remains open.

Machine learning and big data are some of the biggest technology trends nowadays, and neural networks are experiencing their second renaissance. What is this fuzz about? Is it justified? And why does it happen now? This talk will discuss some developments in this field, and show sample applications where machines reach human level accuracy.

Deep Blue was the best chess computer in 1997. Since then, modern chess engines have made significant improvements and are now far stronger than any human grandmaster. This talk gives an overview of how chess engines work, covering search/evaluation and interesting high-level and low-level optimizations. It will also discuss the Fishtest distributed testing framework, a method to measure strength improvement during chess engine development.

Since the earliest days of computer chess, almost all competitive chess programs have used some variant of the alpha-beta algorithm. The alpha-beta algorithm is simple and elegant, always gives the same result as a plain minimax search, and effectively doubles the achievable search depth for a given amount of thinking time. However, the branching factor in chess is so high that even with alpha-beta pruning, a brute-force search will not get very deep in the game tree. The alpha-beta algorithm is also notoriously difficult to parallelize. This talk explains how modern chess programs reduce the tree size much further than what is possible with only alpha-beta pruning, and how to efficiently distribute the search across multiple CPU cores.

4.0 is the new release of the NoSQL Document Database from Couchbase. This is major step forward in data management for mission critical, real time interactive applications. This talk will give an overview of Use Cases suited to NoSQL, an insight into the technical architecture of a clustered NoSQL database, and introduce the new data access capabilities of the SQL for Documents query language.

Es geht um die Präsentation eines Systems zur Lösung des heiligen Grals der Computer Vision: Das Matching von Bildern und Texten und die Bildbeschreibung durch automatisch generierte Texte. Unser System stützt sich auf die Kombination von Deep Learning-Tools für Bilder und Text - nämlich Convolutional Neural Networks, word2vec und Recurrent Neural Networks - mit einem klassischen Computer-Vision-Tool, dem Fisher Vector. Der Fisher Vector wurde modifiziert, um Hybrid-Distributionen zu unterstützen, die viel besser mit Textdaten umgehen können. Unsere Methode hat sich als äußerst erfolgreich erwiesen und übertrifft mit signifikantem Abstand alle zurzeit konkurrierenden Methoden. Wenn die Zeit es zulässt, werden noch weitere Deep-Learning-Projekte wie Gesichts- und Spracherkennung oder Videoanalyse dargestellt.

What if your navigation system could project route information directly to your field of vision, or if facial recognition technology could help you identifying faces? Due to the rapid development of virtual reality glasses in the last few years, this vision can become reality soon. The Maker-Team “Red Pill” from TNG Technology Consulting developed applications that expand reality by useful additional information with the help of the VR glasses “Oculus Rift DK2”. Contrary to the typical use case of such displays, the user doesn’t enter a virtual world but maintains his normal field of vision. This happens through the extension of the VR glasses by two additional cameras. The thus gained stereoscopic picture could for instance be enriched by data for facial recognition and identification. In this talk the TNG consultants will introduce you to the world of augmented reality. You get an insight into the whole program sequence, from the readout of the pictures as far as the representation of the data on the VR display. Thereby they will illuminate the history of this actually not so new technology and point out further areas of application.

New methods and approaches for securing cloud environments are becoming increasingly more critical now that it is being widely adopted by the businesses sector. Cloud system inherently rely on hardware virtualization to enable sharing of hardware resources, and despite the fact that virtualization itself is not inherently insecure, nearly two thirds of all virtual systems are less secure than those physical systems they replace. This curious state arises primarily because traditional host security strategies are not well integrated into virtual environments: as an example, typical antivirus scans are a critical component of layered defense-in-depth, but they rapidly exhaust available CPU and memory when protecting a large number of virtual machines. Virtualization nevertheless also offers a unique opportunity: the ability to peer into a running operating system from an outside perspective, known as introspection. It is possible to observe the memory, storage, CPUs, processes, and kernel of a running virtual machine from a safe vantage point. More interestingly, it is also possible to alter the behavior of all of these components to help protect virtual systems. Over the last few years our team has worked on implementing these features for the open-source Xen hypervisor through the LibVMI library. Our work enables third-party security tools to take advantage of the extra layer of protection presented by the hypervisor, both on x86 and ARM systems. While these features present a significant improvement in cloud security tools, active introspection is a double-edged sword, as it highlights the implicit trust placed into the hands of cloud providers.

For many years enigmail has been one of the standard tools for email encryption with Mozilla Thunderbird. In the post-Snowden era the relevance and the focus of enigmail have completely changed. What was formerly a tool for Techies, has become extremely important for the implementation of the fundamental right to privacy. Usability on the one hand and the compatibility with other email tools on the other hand are the most crucial features. This talk illustrates the current stage of development and gives some insight into the depth of open source development in the age of mass surveillance.

In summer/autumn last year, badUSB devices were short-term the cause of a big safety agitation (e.g. see badUSB: “Wenn USB-Geräte böse werden” | heise Security). Responsible are not only the missing safety mechanisms of the USB printout but also the users’ carelessness or rather their curiosity. Maybe there is something interesting to find on the located USB stick? A business report, or pictures of the attractive neighbor? By means of microcontroller developing plates and a proper USB stick, every reasonably versed developer is able to program USB devices that can almost arbitrarily change their behavior: keyboard, mouse, printer, data medium etc. Already in a free programmable keyboard, that additionally pursues its own plans, a lot of fooling can be done on a victim computer. We demonstrate for example an attack on a standard Windows 7 PC, on which administrative rights are gained by means of a primed device, selective exceptions are established in the firewall and a register console is generated and started. Through the console we show subsequently the content of the hard disk which in reality can be spied on as well. Meanwhile there are collections of finished scripts for different developing plates online, which make amusing gags but also real attacks realizable. Basically every computer and every operating system is vulnerable. The declared aim of the talk is, besides a very slim depiction of the technical basics, to conduct different demonstrations on victim systems with primed devices, in order to create a consciousness for the risks of an all too careless handling with (foreign) USB devices.

The other day, after a release (like this or similar):
Chief: "And, what are the news?" Team: "All the same, except ..." Chief: "What?!? What have you done the last few weeks?" Team: "... the website is now much faster, filling the product index takes only seconds and the search engine finally finds partial words..." Chief: "But how does that work, we were online all the time, right?!?" Team: "Of course!"
The Open Source "full-text search and analytics engine" Elasticsearch doesn’t has to fear commercial competition, because in some cases it is even superior - and not only in price matters. This can be observed in the example of a hybris based web shop which was converted from a specialized, commercial product index to Elasticsearch. This talk depicts how it ever came so far, why Elasticsearch was selected, and what steps were necessary for the uninterrupted migration).

Open Source Software is free, but in order to preserve the freedom of open source and keep this model working, open source licenses also express obligations for users or re-distributors. Today the number of open source licenses is in the hundreds and in most cases, an open source package is not covered by a single license but by multiple licenses. Therefore many different obligations exist and some of them may apply when using a piece of open source software. For commercial use, it is inevitable to assess the license conditions – a kind of clearing task. Software tools support this assessment. One solution is the open source project FOSSology. In the presentation we will explain the main use case, why people should use FOSSology and how latest features improve the clearing with respect to efficiency and effectiveness.

Progressively OpenStreetMap becomes the largest supplier of geospatial data for applications of all kinds. The data is added every day by thousands of volunteers to the database. Indeed, it provides not only static prefabricated maps, but a fortune on data which is available with almost no use restrictions and therefore creates new opportunities to better understand our world. OpenStreetMap is though not just a database, but an entire ecosystem of people, software, and companies that supply and further develop the tools for processing, improving, and visualization of the dataset. The mapping goes beyond the simple recording of roads, it registers also shops, house numbers, collection times, public transport connections, footpaths and much more, in order to make this information available to the public. Errors exist in every database, but in OpenStreetMap everyone - even oneself - may fix them. This talk shall give an insight into eleven years of project history and the current state, as well as provide some introductive points on how one can collaborate.

The problems of the Record system in Haskell are one of its most long-standing and controversial issues. The community has long reached a consensus that the current state of affairs is unsatisfactory, however any solution to it has been spawning debates to this day. The debates have long been the cause of nothing actually getting implemented, despite concrete proposals periodically getting presented since at least the year 1999. Fortunately, the recent developments in GHC made it possible to reimplement the records as a library. Such a library, ideas behind it and its practical usage are the subject of this talk.

Concurrency is here to stay. Your phone is multi-core, your workstation is many-core, and your data centre is a super-computer. As an industry, we need to be able to write code that scales when our hardware scales, we need to be able to guarantee that our code doesn't have data-races or deadlocks, and we need to be able to express all this simply and concisely. Pony is a new open source programming language that aims to do all of this. It's an actor-model language (like Erlang), it's object-oriented with extensive functional features (like Scala), it compiles to fast native machine code without a VM (like C/C++), and it has an innovative type system that guarantees your program will have no data-races, won't deadlock, and will never have a null-pointer or other runtime exception. We'll be discussing some of the fun properties of both the language and the runtime library, and we'll talk about where we're going, and how to get involved.

In this talk you'll hear how the start-up Jolla managed to build a mobile OS and a mobile phone with just 100 employees. Staffed with mostly ex-Nokia developers, we managed to build, launch and ship a non-Android mobile device with our own mobile operating system in a lot shorter time than established vendors have ever been able to. Beyond the story I'll speak details on how a modern mobile operating system is built up - not only from a technical point of view, but how you make a great product and sell it and work with your customers.

The Mirage OS is an open-source library operating system that compiles code written in the OCaml functional language into a variety of hardware backends, most notably specialized unikernels that run directly on the Xen hypervisor. Mirage is particularly useful for building safe, reliable OS components such as storage or networked daemons. Instead of having to manage complex deployments such as a LAMP stack (with the associated security headaches), Mirage offers the opportunity to "compile your own cloud" from a set of protocol libraries. For example, the Mirage website is provided using DNS and HTTP servers running as distinct cloud-hosted unikernels, coordinated via the Irminsule storage stack which uses Git as its communications protocol. Deployment relies uses the Travis continuous integration tool to commit the entire compiled images to (e.g.,) Github from which they can be automatically deployed. In this talk I will describe the architecture of Mirage, present some benchmark results comparing the performance of our unikernels to traditional applications such as Apache, BIND and OpenSSH, run through the deployment workflow benefits it brings, and present our latest results using Mirage for low-latency deployments to low-power small form-factor ARM boards.

More than 30 years ago, Richard Feynman outlined the visionary concept of a quantum simulator for carrying out complex physics calculations. Today, his dream has become a reality in laboratories around the world. All this has become possible using complex experimental setups of thousands of optical elements, allowing atoms to be cooled to Nanokelvin temperatures, where they almost come to rest. The atoms can then be trapped and manipulated in arrays of millions of microscopic light traps. Such ‘light crystals’ allow an unprecedented view into the microscopic world of quantum materials and have enabled the most precise atomic clocks to date that are fundamental to next generation timing and navigation applications. In my talk, I will give an introduction how such quantum simulators can be realized at the lowest man-made known temperatures and outline some of their applications ranging from condensed matter physics over statistical physics to high energy physics with table-top experiment.

Cortical.io’s Semantic Fingerprinting technology originates in a new, fundamentally different machine learning approach: it is based on a statistics-free processing model that uses similarity as a foundation for intelligence. The Cortical.io Retina converts any kind of text into a numeric representation, a Semantic Fingerprint that encodes meaning explicitly with all contained senses and contexts. The system "understands" the relatedness of two items by simply measuring their overlap. As a result, it is very fast, reliable and easy to implement - a breakthrough technology that leverages the intelligence of the brain to enable the Natural Language Processing of Big Text Data.

In this talk we report on the latest results of our social software engineering research at TU München. SocioCortex is a web-based platform that provides a novel mix of content and model management concepts and services to support problem-solving processes in organizations. These processes involves stakeholders with different interests and background that want to use their preferred content representations (tables, hypertexts, images, drawings, maps, 3d models, matrices, mathematical formulas) which involve diverse content sources and channels. Using examples from industry projects we illustrate how SocioCortex enables the emergence of data models, access-control modes, process models and UI models shaped by the actual problem-solving processes performed on the platform.

Active-Matrix Organic Light-Emitting Diode (AMOLED) displays are among the most exciting developments in consumer electronics in the past decade. 250 million AMOLED displays with a total area of 1.6 square kilometers and a total value of 12 billion Euros were manufactured in 2013, mainly for mobile phones and small tablets. Current trends in AMOLED displays include higher resolution (beyond HD), larger screen size (55 inch diagonal) and the development of flexible backplanes for rollable or foldable displays. The latter will benefit from a technology which allows the thin-film transistors (TFTs) that control the individual display pixels to be fabricated at temperatures compatible with polymeric substrates, i.e., at temperatures below about 150 °C. One possibility are TFTs based on conjugated organic semiconductors.

It is amazing how we can use tools so easily from a simple screwdriver to dexterous usage of chopsticks, our brain seamlessly make use of these devices for fairly complex tasks. In this talk I'll touch on ways in how we can fool our brain embrace even complex robotic systems.

It seems that microservices are increasingly being hailed as the panacea of architecture. However, many organizations are completely unprepared for the technical, operational, and organizational consequences of this approach. Deploying and running microservice-based systems will bring new challenges to the devops table. Some of these challenges can be alleviated by the recent container movement (of course, not without introducing some new ones). In this talk, we'll take a look at using Docker for deploying microservice architectures as well as the options provided by (or missing from) the current ecosystem.

Analyzing classic economic indicators such as the gross domestic product and the unemployment rate Bavaria is far ahead in the national comparison. We all like to live in Bavaria - but is Bavaria really well prepared for the future? This image changes a lot if one takes instead of the classic indicators rather the future-oriented indicators such as income distribution, start-up quote, integration index and education mobility into account. From McKinsey’s perspective, the world is changing radically: structural breaks as “the end of the work as we know it”, the emergence of a “volatile world”, the shortening of resources (budgets) and the development of “disruptive technologies” will lead us to a completely different and rather unknown world. Pertaining the impacts of these breaks one can find two opposite interpretations: on the one side a world characterized by cleavage and on the other side a world as a stage for participation. New playing fields may pick up the conflict zones: we gathered a total of 15 initial ideas (e.g. digital education, integrated health care system, energy efficiency and independence, revival of the Bavarian identity). When discussing the future of Bavaria it became clear that there isn’t “a Bavaria”, but several regions that should work in synergy rather than compete.

Although Behavior Driven Development has been existing for over 10 years, the methodology hasn’t yet been very popular in the Java world. One reason for this are the existing BDD tools for Java that are cumbersome for developers to use and require a lot of maintenance. The speaker wants to change this with JGiven and provide Java developers with a framework that they like to use and at the same time satisfy the operating department with instructive reports. JGiven scenarios are written in the usual Given-When-Then formula with an embedded Java DSL. This allows developers to use all IDE features such as auto completion and refactoring tools. The resulting scenarios are indeed separately very readable, but JGiven can additionally generate more reports in different formats that can be used for collaboration with domain experts. Through a modular concept, new scenarios can be easily assembled from parts of other scenarios. This speeds up the creation of new scenarios and avoids test code duplication. Since neither Groovy nor Scala are still needed and JGiven is compatible with JUnit and TestNG, JGiven can be immediately applied in Java projects and be easily integrated into existing test infrastructures. In this presentation, the speaker will give an introduction to JGiven and, based on a short live coding session, show how quickly and easily BDD Scenarios can be written in JGiven.

Kanban has been around for 10 years already. We know it provides direct benefits for service delivery. We know that evolutionary change is a better way to adapt organizations to market uncertainties and changing market conditions. However, Kanban adoption has mostly been inside-out, starting with small teams and in the middle of service delivery workflows. Many Kanban implementations are started locally and provide only local benefits. Kanban was always intended as an end-to-end enterprise solution. Enterprise Services Planning (ESP) is the future of Kanban. ESP is an outside-in approach that looks at strategy, fitness for purpose and alignment of capability with strategy enabling better business results from the use of Kanban as a service delivery method and enabler of evolutionary change. ESP includes guidance on portfolio management, capacity planning, scheduling, forecasting, and risk. ESP is supported by new software tools that facilitate decision making and process improvement helping managers do their jobs by informing them better and making recommendations based on an understanding of business risks. ESP offers AI for your technology business. It's the future for how modern 21st Century businesses will be managed!

Modern software relies on tests to avoid bugs instead of proving code correct because traditional formal methods are prohibitively expensive. This talk illustrates a more practical approach to formal methods that combines equational reasoning in Haskell with introductory category theory to cheaply verify high-level properties.

The Spotify desktop app is built out of a large number of small, self-contained web apps called Spotlets. In this talk, I'll tell you how they allow us to divide ownership between teams, and how they allow us to remotely live-replace parts of the app in the running clients of millions of users. I will also tell you the story behind how Spotlets originally came to be, how they've evolved, and some of the surprising effects they've had on us.

Sannsyn is a small Norwegian startup which develops a platform for recommendations. Having developed our own toolbox from scratch we see that there are several challenges our customers want solved where we have to develop entirely new sets of algorithms. Collecting and processing lots of data about persons also raises important issues with regard to privacy which we have an obligation both to law and ethics to solve.

Here at JetBrains we’ve spent over a decade making all sort of development tools. We’ve worked with a great many programming languages out there (Java, Python, Ruby, C#, JavaScript, Php, etc.), and even made our own (Kotlin). However, surprisingly, C++ was not between them for a long time. But year 2015 came to change it all with two C++ products released recently - ReSharper C++ (a plugin to Visual Studio) and CLion (a cross-platform C/C++ IDE). In this talk I’ll share the long story of JetBrains coming to C++ domain, that actually has started in AppCode, our IDE for iOS/OS X development. As well as some technical difficulties and concerns we’ve met on the way, especially related to C++ parser, build systems and compiler toolchains. You’ll also find out how the CLion product was born and about our future plans and why we actually did C++ product twice (or even three times), together with a funny story of creating “CLion” name.

NP-complete problems often occur at the heart of important application domains like logistics, scheduling, or satisfiability checking. Today, no fast and correct algorithm is known for any NP-complete problem; in the worst-case, they all require exponential time. If satisfiability checking with 3 variables takes 1 second, exponential time then means that 13 variables require 1024 seconds, and 35 variables require 4 billion seconds or 126 years. Despite this, modern satisfiability checkers deal with 500.000 variables in less than 10 minutes. This talk reveals some key techniques that enable such performances.