Schedule VIScon 2019

Opening Ceremony

Opening Ceremony

Keynote

Keynote

Advanced Analytics for Power Grids

Today we are experiencing a digital revolution; almost all different types of business are being disrupted through new solutions based on the analysis of huge amount of data. This disruption is also reaching the energy sector, but for traditional ...

Pricing climate change for investors

Carbon Delta is a climate change research firm that specializes in identifying and analyzing the climate change resilience of investments. This allows our clients to protect assets, optimize performance and reach sustainability goals. Our startup ...

The goal of this presentation is to share with you first-hand experience of a start-up on the intersection of Climate Change and Fintech. Since 2 of the 3 co-founders hold a Ph.D. from the ETH, and multiple colleagues also graduated from the ETH, our experience might inspire you in your own endeveaours and give insight into how the development of a start-up based in Zurich might develop. For that purpose, the presentation is structured in the following 3 main parts.
Firstly [ 5min ], you will receive a brief introduction to the company Carbon Delta. I will lay out our main cornerstones and milestones, give an overview over our development over the last 4 years, and explain our team development.
Secondly [10 min], I will give you an introduction to our main product we developed: Climate Value-at-Risk. I will elaborate on how we developed our methodology over time, what priorities we had to set, and how we overcame various difficulties. In addition, the role of our colleagues with computer science background in the development of the model will be explained as well.

Thirdly [30 min], I would like to share with you our so-called start-up experience and answer any open questi...

All Code Sucks

All code is always work in progress - here’s why

Spend any amount of time in the Software Engineering industry, and you will know: all code sucks. People take shortcuts, frameworks get outdated, old code is haphazardly adapted to support a new feature. The quick dirty fix is chosen before the redesign that would solve a fundamental problem.

But does it really have to be this way?

To keep your code base clean takes effort. It requires an understanding of why things get bad, and what can be done to prevent it. This talk explores the pitfalls of code maintenance, discusses countermeasures and best practices, and tries to give an insight into the challenges of maintaining a large code base.

A guide to mobile location data

FAIRTIQ lets people get a valid public transport ticket with a single swipe, and the best fare is computed automatically at the end of the trip. While developing and operating our product, we have gathered insights on building systems that work wi...

FAIRTIQ built the first hardware-free solution for public transport ticketing in Switzerland based on the Check-In — Check-Out principle. Users can check-in and receive a valid ticket with a single swipe in the app, and the best ticket is computed automatically at the end of the trip. In the spring of 2018, FAIRTIQ became available across the whole Switzerland. We now also have ongoing pilot projects in Austria, Germany and the UK.

While developing and operating our product, we have collected many insights building systems that work with mobile location and other sensor data. Working with sensor data is always a challenge. One must answer questions such is the solution robust enough for a variety of inputs? Does it need a constant stream of new data to make improvements? And what metrics need to be implemented to ensure the high quality of the solution?

This talk focuses on guidelines for building robust systems to process mobile location, activity and other sensor data and monitor their performance. We also discuss the design choices for data collection and data annotation processes, and how they affect the further development of the system.

Collaboration between architects and software engineers: Potential and opportunities for computational design workflows

The shift in architectural practice towards workflows that use computation as an integral part from conception to realisation of projects reveals new creative potential. In this context, the symbiosis and collaboration of architects / designers an...

See attached file for Lecture content

IoT in Practice

In this talk, we show how three of our customers leverage the Internet of Things to gain competitive advantage.

In this talk, we show how three of our customers leverage the Internet of Things to gain competitive advantage: Coop turned their inventory management into real-time, Bossard provides smart logistics for factories, and Belimo devices are now part of an IoT ecosystem that enables novel use cases. We present practical examples of the challenges we encountered and outline the technical solutions.

Visual-Inertial SLAM: From Flying Robots to Virtual Worlds

Visual-Inertial SLAM ("Simultaneous Localization and Mapping") is a technology with major contributions from several research labs at ETH and UZH over the last decade. This technology enables machines to localize similar to GPS but much more preci...

The humans in the loop

In this talk we'll learn the various (and sometimes surprising) ways humans interact with Machine learning models. At GetYourGuide, we use Machine Learning to help travellers find and book activities when they travel to a destination, and develop ...

Machine Learning algorithms are only as good as the data they are built on. In data science, we tend to emphasise and promote the algorithms but forget to give credit to the data they are built on, and often to the humans generating that data! Directly or indirectly humans influence the products and algorithms we built, and sometimes that explains why things go wrong or not in the direction we expect. Via examples and experiences picked in our own domain, we'll aim to find answers to the following questions
* What could go wrong? When not to trust humans to train ML models?
* How to take messy user behavior data and build a strong training dataset?
* What techniques can you use to leverage the smallest amount of data possible?

From Scalevo to Scewo Bro

Roadmap of the Scewo Spin-off. A short talk about how it all started, what our MVP was, how we got the first customers and our first funding.

Scewo is an ETH Spinoff that develops an electric stariclimbing wheelchair. It balances on two big wheels to drive around on even ground and it has a set of tracks to climb over obstacles. The device is very software heavy and allows for a lot of automation. With the Scewo Bro want to sent a new standard in the wheelchair industry. In this talk you will hear about how it all started and where we are now.

Do you need a Blockchain?

Blockchain is being praised as a technological innovation which allows to revolutionize how society trades and interacts. This reputation is in particular attributable to its properties of allowing mutually mistrusting entities to exchange financi...

Talk of Karl Wüst based on his paper: https://eprint.iacr.org/2017/375.pdf

Hello World! with HoloLens 2

Augmented Reality (AR) has become a buzzword in today's industry. Many of us have seen it on TV, on mobile applications or experienced it in some other form. As an ETH CS graduate I'm always curious how difficult it is to get started with new tech...

My own custom Kafka Cluster in 60 seconds

Speed is key and having access to large data streams with minimal latency is a mission-critical component in most modern application architectures. In this hands-on workshop, participants will learn how to leverage cloud-native Kubernetes features...

Event streaming has become the dominant integration pattern for modern, near real-time applications and is applied for a variety of tasks across industries. Be it an E-Commerce shop looking to recommend new products to users based on their behavior or IoT edge devices continuously sending sensory data for continuous risk modeling, a high throughput, low latency event streaming architecture plays a crucial role in today's IT infrastructure. Apache Kafka was originally developed at LinkedIn and has since become one of the most popular event streaming frameworks. It has proven to be able to scale up to multiple millions of messages per seconds on modest hardware requirements. Setting up a fully operational Kafka cluster used to quite a challenging task because of the various components that are required and the security that had to be implemented on top of it. With the emergence of the open source Strimzi framework, deploying Kafka on a scalable container-based infrastructure has become a lot easier to set up and manage using state of the art DevOps and infrastructure-as-code principles.

In this hands-on workshop, students will learn how to leverage Kubernetes custom resource ...

From data to forecast

During this workshop, we would like to showcase end to end scenario of time series forecasting.

Data science is a complex domain and one of the fastest-growing right now. Although it is one of many pieces in the process of applying it to real-...

Prerequisites to enter the world of computer science as a non-computer scientist… and stay happy!

In this talk, I would like to use my daily experience to show how to get into the IT sector as a non-computer scientist, what it means to work there and how to stay happy.

In my talk, I plan to show what it means to work as a non-computer scientist in a computer science field based on my everyday experience. To illustrate this, I will mainly refer to a project our company, EBP, is developing for the Mobiliar Lab for Natural Risks: A web application for visualizing flood exposure and flood risks throughout Switzerland. Although I had only a minor IT-background when I started working at EBP, I currently work as a software developer in this project. How it came to this I will elaborate in my talk.
In my talk, I will also show that often IT expertise is not the only one required but, especially in the case of this specific project, it needs to be coupled with expertise in environmental sciences, safety and risk as well as communication. It is precisely the interdisciplinary cooperation which makes this work exciting and enriching to me.
In addition to providing an insight into my everyday work life, I will also give the audience a kind of guideline to find out whether a computer science-related job will make them happy in the long run and what to consider when looking for a job in this area as a non-computer scientist.
Furthermore, my talk will dis...

The Good, The Bad And The Cybeer

The InfoGuard Red and Blue Team show possible attack paths usually taken by an advanced persistence threat (APT) during an ongoing attack and also the capabilities of the defending side.

The talk is going to highlight possible attack paths usually taken by an advanced persistence threat (APT) during an ongoing attack. Additionally, we demonstrate the capabilities of the defending side. The talk is structured in a dialog between Red Team (Offensive) and Blue Team (Defensive) member of InfoGuard AG. We will discuss various possible attack scenarios along the Martin Lockheed Cyber Kill Chain and how to detect and analyse them.

The following subjects will be part of the talk (not conclusive):
• Open Source Intelligence
• Ways for delivery (Phishing, Social Eng, Malware Spam)
• Pwn all the things (Exploitation, Privilege Escalation & Installation)
• How to call home (command & control channels)
• Focusing on the goal (getting valuable data out of the network)

Automatic Slide Layout

Since Microsoft PowerPoint's release in 1987, the process of creating slides has remained essentially unchanged. Every day, millions of users waste much of their time pushing text boxes around to create legible, good-looking slide layouts.

It would be very useful to have an interactive, easy-to-use tool that could automatically perform most of those layout tasks. This is a hard problem to solve for several reasons:

1. We need an algorithm that can handle arbitrary user input and produce meaningful results.
2. Its decisions have to be predictable.
3. The problem input is purely geometrical and lacks any semantic information.

We have developed a general-purpose layout algorithm that can solve this kind of problem.
We transform the user input into a system of linear constraints and solve a sequence of linear optimization problems: a so-called lexicographic optimization problem. First, we optimize the degree of constraint satisfaction to deal with over-constrainedness and second, we optimize the space distribution. We use the same layout algorithm in our cross-platform UI library to layout resizable dialogs.

The talk will delve into the mathematical details, and it...

From a start-up diary: Dealing with uncertainty

During this talk, we will present a different side of Software Engineer work from an early stage tech start-up.

Instead of coming to work and pick up the most interesting task from a Kanban board, imagine that you are coming to a meeting with ...

Virtual Reality on Oculus Go

Oculus Go is a standalone wireless VR headset. It brings experiences to life with high detail, vibrant colours and a wide field of vision. There are many tools available to design scenes for the oculus go. We will focus on unity, The world’s leadi...

Launch your startup in Space

TOPIC:
- Space: The European Space Agency (ESA) is Europe's gateway to space. Its mission is to
shape the development of Europe's space capability and ensure that investment in space
continues to deliver benefits to the citizens of Europe and t...

SPEAKER:
Cyril Kubr started his career in asset management and worked in London, Singapore and Hong Kong. Later he joined the team of Impact Hub Zurich and founded three companies. In 2016, Cyril co-launched ESA BIC Switzerland, an incubator focusing on high tech startups. He studied Media Science in Fribourg and Innovation in Oxford.

SCION Future Internet

Experience the Next Generation Internet Architecture SCION in our workshop. Find the best paths to the destinations; go around undesired ones. Win the competition by coding the best algorithm.

In our VISCon workshop, students not only learn about state-of-the-art research about Path-aware Networks, but get the chance to experience this technology hands-on.
The two hour session starts with a quick overview of SCION, a next-generation Internet architecture developed at ETHZ. We highlight on some of the problems and shortcomings of the current Internet, while showing at the same time how SCION can solve them. This introduction is expected to take 20 minutes.
The remaining hour and a half is devoted to a practical, fun competition. Students have to code a client application (python) to communicate with servers deployed in our testing infrastructure. The goal is to solve networking problems such as unavailable routers, extremely low bandwidth, heavy packet loss at the links etc., by leveraging the cutting-edge features of SCION. The more content students are able to get across the network and the fastest, the higher their final score. This game is repeated multiple times, so that they can learn from their mistakes and improve the performance of their application.
To ease the bootstrapping and speed up development, we provide some basic examples of clients. Students can ...

How the Cloud evolves: Developments and Technological Trends in Practice

Driven by digitalization, many companies have started their journey to the cloud. Cloud Computing promises to simplify IT operation at a much lower cost. Can these promises be fulfilled in practice? In this presentation we would like to look behin...

Among other, we will dive into the following topics:

From legacy workloads to cloud native applications: Many companies have built over time complex application ecosystems that are difficult to maintain and manage. To address these challenges, we will highlight the right use of cloud services for different application architectures. As part of this journey, we will cover a variety of alternative building blocks covering virtual machines, orchestrated containers or fully managed application platforms.

Technological trends becoming reality: With the advent of new technologies such as IoT, 5G and edge computing the cloud landscape is going through an evolution. We will discuss what innovations and use-cases result from these new technologies and dive into the technical and architectural consequences.

Airlock: Defeating the unknown: Machine Learning in Web Application Security

In the continuous arms race between hackers and security vendors in the field of web application security, some suppliers promise that machine learning is the next Swiss Army Knife to defend against web security threats.

In the continuous arms race between hackers and security vendors in the field of web application security, some suppliers promise that machine learning is the next Swiss Army Knife to defend against web security threats. As a Swiss vendor of the web application firewall Airlock, we analyzed machine learning-based techniques to detect web attacks. Our proof of concept with productive web traffic shows that machine learning is useful to detect certain threats that cannot be well described by predefined static rules. On the flip side, the approach brings new operational challenges. In this talk, I will present our ideas, experience and results of the proof of concept.

Towards Graph-based Machine Learning for Automated Health Care Services

In this talk, we demonstrate how we tackle the problem of diagnosis prediction in healthcare domain by employing a combination of Knowledge Graph and Machine Learning techniques, namely, Recurrent Neural Networks. Our approach shows significant pr...

We tackle the problem of predicting diagnoses for patients staying in critical care units. To this end, we employ healthcare data of 46 thousand patients with multiple admissions per patient. We consider multiple events as features per admission like fluids (e.g., insulin), lab tests (e.g., pH) and drugs (e.g., aspirin) which represent the evolving state of an admission. We employ graph data model to integrate additional information, per admission, from external sources (e.g., disease-symptom relations) and feed this enriched admission to a Recurrent Neural Network to predict diagnoses. Our approach shows significant results due to the relational information from previous admissions (connected via the external disease-symptom information).

Enzian: a research computer

Academic research in rack-scale and datacenter computing today is hampered by lack of hardware. Cloud providers and hardware vendors build custom accelerators, interconnects, and networks for commercially important workloads, but university rese...

From research code to industry grade with SCION

What started as a research project is now trying to take over the world!

At Anapaya Systems, we are leveraging the SCION Internet architecture to offer next-generation Internet services to our customers. This entails further developing and tran...

Design Thinking

Millennials are rewriting the Swiss IT landscape

What do you need to keep in mind when developing software for us, the millennials? We are becoming the most valued customers - but how can the business world connect with us? Why do Swiss companies constantly fail to deliver products we want?

A...

When sh!t hits the fan

Assume breach is a way of thinking about your information and IT security strategy. It is the assumption that your organization has already been successfully attacked by an adversary or could easily be compromised by a dedicated attacker. W...

Talos, the Deep Learning solution of Credit Suisse

Talos, the Deep Learning solution of Credit Suisse by Angeliki Davourli

Credit Suisse has implemented a Deep Learning solution “Talos” that is used in production in the eCommunications surveillance space. This solution operates on the unstructured data space and leverages GPU technology. Our solution reduces the noise and false positives by leveraging cutting-edge Natural Language Processing techniques to surface potential malicious activity.

Talos works on top of a Big Data analytics platform that ingests and processes millions of eComms a day. We run the data against behavioural models, to generate signals that indicate potential misconduct and market abuse for investigation. The surveillance platform is used for supervisory and compliance review to protect the firm’s reputation.

Talos enables supervisors to prioritize their attention on alerts that require deeper investigation, and act on potential misconduct activities quicker. In more technical detail, the first Talos model (“Talos 1.0”) has been a Convolutional Long Short-term Memory architecture utilizing internal word embeddings, which enabled us to reduce false positives by 37%. The second model of Talos (“Talos 2.0”) has been using BERT language model in a Feedforward Neural Network...

Exploring Graph Analytics and Visualization with Oracle Technologies

This workshop would demonstrate the benefits of using Graphs for different use-cases starting from CyberSecurity (e.g., Threat hunting) to Retail (e.g., Recommender Systems). We would leverage technologies developed at Oracle Labs for this worksho...

The objective of this workshop would be to demonstrate the benefits of using Graphs across multiple domains starting with CyberSecurity (e.g., Threat hunting) to Retail (e.g., Recommender Systems). We would leverage technologies developed at Oracle Labs, namely Parallel Graph AnalytiX (PGX), and Oracle Labs DataStudio (OLDS). PGX (https://www.oracle.com/technetwork/oracle-labs/parallel-graph-analytix/overview/index.html) is a toolkit for graph analysis - both running algorithms such as PageRank against graphs, and performing SQL-like pattern-matching against graphs, using the results of algorithmic analysis. On the other hand, OLDS is a JVM-based notebook interface for data analysts built on Apache Zeppelin and Oracle JET with a focus on graph visualization. With the use of these two tools, we will explore multiple graph use-cases in the above-mentioned domains.

Container Orchestration with Kubernetes

Deploy your applications like a boss.

Docker has revolutionized the way we work at Beekeeper AG. It allows us to easily deploy our software and gives us a strong guarantee that the container images we build and run locally are identical to the ones we deploy to production. However, this is only half the story. When using docker in production you run into a number of problems almost immediately. For example, how do we know our containers are healthy and running? Is the latest version deployed? How can we perform upgrades without impacting our customers?

In order to solve these orchestration issues, Google created Kubernetes, an open-source system for automating deployment, scaling and management of containerized applications. In this workshop we will cover its basic concepts and take you through all the steps you need to follow in order to get your applications running on Kubernetes.

If you want to participate, please install the Minikube application: https://kubernetes.io/docs/tasks/tools/install-minikube/
We recommend to use Linux, though all operating systems should work in theory.

Experiencing Web of Things

One of the challenges faced by implementations of Internet of Things is that they lack interoperability. Systems turn out to be vertical silos of applications which are tightly-coupled not only because of communication interfaces, but also due imp...

The talk will give a brief introduction to the problems in the field of Building Automation in specific, and then go on to explain how the paradigms behind Web of Things promise to offer a solution. We
then examine some real-life problems faced when dealing legacy systems. Finally we will highlight the potential research areas in this field.

Implementing a privacy-compliant software - practical experience

Many cloud services nowadays collect and process a lot of photos, documents, ratings and other personal data. Data protection has changed a lot over the course of the past 2 years - not only in Europe. It became a tough job for programmers to buil...

In this talk, we will give you a brief overview about:
- What has changed in Europe in the past 2 years in the area of data protection?
- How does Switzerland fit into this picture?
Then we will focus on:
- Which types of personal data do exist?
- How are you allowed to process this data in your software?
- What is profiling and when is it allowed?
- When and how should you anonymise the data?
- Can you keep the data forever? If not, when should you delete it?
- What you should notify your users about and when?
We will support these topics with the newest examples that popped up in the media in the resent months.

From Zero to Singularity - Designing the AI that humans need

We have tried many things. All of those work to a certain extent. Some make us stronger, some make us faster, some make us sick. The ability to think always has been a major challenge. Consciousness is the state or quality of awareness or of being...

Defending against Wi-Fi Hacks

The man-in-the-middle poses a real security threat to users of ubiquitous communication devices such as notebooks and smartphones.
In this talk we assume the role of an attacker and demonstrate the capabilities of an inexpensive, off-the-shelf ro...

Why you should care about your internet connected toaster

More and more technologies are becoming part of our daily life and everything is increasingly connected through the Internet. Despite some major incidents lately gained a lot of mediatic attention, security awareness seems to stop at the point whe...

The goal of this talk is to show the importance of considering security aspects when developing and deploying new technologies. Through real-life examples that span over different engineering fields we’ll show how particular poorly engineered or deployed systems have been or could have been misused to wreak havoc in the world.

Computer Vision and Machine Learning for Sports Broadcasting

In our talk, we will introduce practical aspects related to the development of Viz Eclipse, one of the latest Vizrt’s sports broadcast software, which is primarily developed in our Zurich R&D center. Through state of the art computer vision and ma...

In our talk, we will introduce practical aspects related to the development of Viz Eclipse, one of the latest Vizrt’s sports broadcast software, which is primarily developed in our Zurich R&D center. Viz Eclipse replaces the field-side advertising boards in football game broadcasts with photo-realistic virtual signage in real-time and throughout live gameplay, allowing rights holders to unlock further revenue by creating unlimited feeds for regionally-tailored marketing. Viz Eclipse combines state of the art computer graphics, vision and machine learning techniques, operates directly on video feeds, thus not requiring any specialized hardware, and seamlessly integrates with existing production pipelines. Among other state of the art computer vision technologies, Viz Eclipse employs a machine learning pipeline which is able to seamlessly key players, officials and other objects placed in front of the field-side advertising boards, to then replace said advertisements with novel virtual signage. Crucially, Viz Eclipse solely rely on live television feeds to produce virtually unlimited streams, and can be applied to any kind of backgrounds, including animated LED boards.

Developi...

Visual Inertial System for 3D Laser Scanners

Visual Inertial Odometry (VIO) is a key technology in most augmented reality or autonomous systems. It is the process of determining the position and orientation of a camera and Inertial Measurement Unit (IMU) by analyzing accelerations, angular v...

Site Reliability Engineering at Google

This talk is about Site Reliability Engineering at Google and how we apply the SRE principles in practice.

What is Site Reliability Engineering? It’s all about scale! Things fail all the time: Machines fail, switches fail, fibers fail, power fails. How can we build reliable large scale systems that can survive even large scale disasters?
It is all about managing risk. Hope is not a strategy! This talk is about how we implement distributed and highly-available systems at Google and introduces the SRE principles in general.

Closing Ceremony

Closing Ceremony