Quick navigation

Main navigation

Artificial Intelligence in Industry and Finance

We would like to welcome you to the «3rd COST Conference on Mathematics for Industry in Switzerland» on Artificial Intelligence in Industry and Finance, hosted by the Institute of Applied Mathematics and Physics and the Institute of Data Analysis and Process Design at the Zurich University of Applied Sciences ZHAW in Winterthur, Switzerland.

Artificial Intelligence in Industry and Finance (3rd European COST Conference on Mathematics for Industry in Switzerland)

September 6, 2018, 9:30-16:30 - ZHAW Winterthur, Technikumstrasse 71, 8401 Winterthur


  • Logo ZHAW School of Engineering
  • Logo School of Management and Law
  • Logo Data Service


  • Logo Swiss Alliance for Data-Intensive Services
  • Logo COST
  • Logo Fintegral
  • Logo InCube

Information at a glance

Aim of the conference

The aim of this conference is to bring together European academics, young researchers, students and industrial practitioners to discuss the application of Artificial Intelligence in Industry and Finance within the COST research network.

The conference is supported by the COST Network MI-NET and by the Swiss State Secretariat for Education, Research and Innovation SERI. COST is the longest-running European framework supporting transnational cooperation among researchers, engineers and scholars across Europe.  The 1st COST Conference on this topic was held on September 15, 2016, and the 2nd COST Conference was held on September 7, 2017.

All lectures are open to the public. Registration is now open.

Keynote talk in 2017

Scientific topics

  • Artificial Intelligence challenges for European companies from the mechanical and electrical industry, but also life sciences.
  • Artificial Intelligence challenges  for the European Fintech industry, banks and insurance companies


We have invited 20 speakers both from within Switzerland as well as abroad, working on AI in Finance and Industry.

One track will focus on financial mathematics and its applications of machine learning, whereas the other one will tackle the implications for industry.

Session progress in 2017


In September 2017, we have had more than 190 participants, both from Academia and Industry. The latest installment of the COST conference also saw a large number of international guests and speakers, travelling to Switzerland from destinations such as the UK, Germany, the United States and Bulgaria.

The largest proportion of participants come from the industry complemented by a significant number of academic researchers. This mirrors our unique approach of connecting the academic world to their respective fields of application, putting new exciting concepts to work in industrial frameworks, where they can open up new opportunities.

Participants in 2017

Key Note Presenter

Prof. Dr. Marcello Pelillo, Università Ca'Foscari Venezia/ European Centre for Living Technology: "Opacity, Neutrality, Stupidity: Three Challenges for Artificial Intelligence"

Invited Speakers

Financial Mathematics

  • Anna Maria Nowakowska, InCube: "Recommender Systems for Mass Customization of Financial Advice"
  • Dr. Daniel Egloff, Flink AI/QuantAlea: "Trade and Manage Wealth with Deep Reinforcement Learning and Memory"
  • Prof. Dr. Markus Loecher, Berlin School of Economics and Law: "Pitfalls of Variable Importance Measures in Machine Learning"
  • Dr. Jürgen Hakala, Leonteq Securities AG: "Machine Learning applied to SLV Calibration"
  • Dr. Yannik Misteli, Julius Bär: "Decision Trees in Machine Learning"
  • Prof. Dr. Paolo Giudici, University of Pavia: "Scoring Models for Roboadvisory Platforms: A Network Approach"
  • Prof. Dr. Marc Wildi, ZHAW School of Engineering: "FX-trading: challenging intelligence"
  • Alla Petukhina, Humboldt University of Berlin: "Portfolio allocation strategies in the cryptocurrency market"
  • Dr. Damian Borth, DFKI Kaiserslautern: "Deep Leaning & Financial Markets: A Disruption and Opportunity"
  • Prof. Dr. Andreas Hoepner, UCD/MFS/Henley Business School/University of Hamburg: "Finance Keynote Talk: Embracing AI Opportunity = (Humans*Teamwork)^Machine -1"



Industrial Mathematics

  • Dr. Christian Spindler, PWC: "Trust in AI: explainability and compliance"
  • Dr. Martin J. Fengler, Meteomatics: "Big Data meets weather: How real-time access to weather data enables a rapid development of business applications"

Prof. Dr. Marcello Pelillo

Marcello Pelillo is a Professor of Computer Science at the University of Venice, Italy, where he directs the European Centre for Living Technology and leads the Computer Vision and Pattern Recognition group, which he founded in 1995. He held visiting research positions at Yale University (USA), McGill University (Canada), the University of Vienna (Austria), York University (UK), the University College London (UK), and the National ICT Australia (NICTA) (Australia). He serves (or has served) on the editorial boards of IEEE Transactions on Pattern Analysis and Machine Intelligence, IET Computer Vision, Pattern Recognition, Brain Informatics, and is on the advisory board of the International Journal of Machine Learning and Cybernetics. He has initiated or chaired several conferences series (EMMCVPR, IWCV, SIMBAD, ICCV). He is (or has been) scientific coordinator of several research projects, including SIMBAD, a highly successful EU-FP7 project devoted to similarity-based pattern analysis and recognition. Prof. Pelillo has been elected a Fellow of the IEEE and a Fellow of the IAPR, and has been appointed IEEE Distinguished Lecturer (2016-2017 term).

Anna Maria Nowakowska

A Brief Biography


Anna Maria Nowakowska leads the Data Analytics team at InCube, which focuses on delivering data consulting services to clients within the Swiss financial sector. She holds a Master of Engineering degree in Electronics and Electrical Engineering from the University of Edinburgh, as well as the Chartered Financial Analyst® designation from the CFA Institute. She has over 7 years of experience in the software and financial services industries and has worked in the UK, US and Switzerland.



Recommender Systems for Mass Customization of financial Advice


Recommender systems have been widely adopted in areas such as online shopping and movie streaming. They automatically suggest new items to users based on their characteristics and previous behaviour. Despite the support that recommender systems can bring to decision making in finance, their application to banking data is an underexplored field, and our research is focused on filling this gap. We build recommenders for private and retail banking use cases, following the growing push for digitization and mass customization of financial advice. The vision is to enhance the quality of personal financial advice and to make it accessible to a wider client base, by automating a large part of the process.



Dr. Daniel Egloff

A Brief Biography


Dr. Daniel Egloff is the founder of Flink AI and QuantAlea. Flink AI is developing new AI solutions using Reinforcement Learning and is advising banks, hedge funds and eCommerce companies on practical applications of AI and Deep Learning. QuantAlea is a Swiss based software engineering company specialized in GPU software development and high performance numerical computing. He studied mathematics, theoretical physics and computer science and worked for more than 15 years as a quant in the financial service industry before he started his entrepreneurial career in 2007.


Trade and Manage Wealth with deep Reinforcement Learning and Memory


In this session we present how Deep Reinforcement Learning (DRL) and memory extended networks can be used to train agents, which optimize asset allocations or propose trading actions. The memory component is crucial for improved mini-batch parallelization and helps to mitigate catastrophic forgetting. We also address how concepts from risk sensitive and safe reinforcement learning apply to improve the robustness of the learned policies. The DRL approach has several advantages over the industry standard approach, which is still based on the Mean Variance portfolio optimization. The most significant benefit is that the information bottleneck between the statistical return model and the portfolio optimizer is removed and that the available market data and trade history is used much more efficiently




Prof. Dr. Markus Loecher

A Brief Biography

Prof. Dr. Markus Loecher has been a professor of mathematics and statistics at the Berlin School of Economics and Law (HWR Berlin) since 2011. His research interests include machine learning, spatial statistics, data visualization and sequential learning. Prior to joining HWR Berlin he worked as principal and lead scientist at various data analytics companies in the United States. In 2005, he founded a consulting firm, DataInsight, which focused on applying novel statistical learning algorithms to massive data sets. Prior to DataInsight, he worked at Siemens Corporate Research (SCR) in Princeton, NJ for 5 years, where he focused on failure prediction. Markus Loecher completed his postdoctoral research in physics at the Georgia Tech University in which he studied the spatiotemporal chaos. He holds a PhD in physics and a master degree in statistics.


Pitfalls of Variable Importance Measures in Machine Learning


Random forests and boosting algorithms are becoming increasingly popular in many scientific fields because they can cope with "small n large p" problems, complex interactions and even highly correlated predictor variables. The predictive power of covariates is derived from the permutation based variable importance score in random forests. It has been proven that these variable importance measures show a bias towards correlated predictor variables. We demonstrate the fundamental dilemma of variable importance measures as well as their appeal and wide spread use in practical data science applications. We address recent criticism of the reliability of these scores by residualizing and deriving analogous procedures to the F-test.

Dr. Jürgen Hakala

A Brief Biography


Jürgen works for Leonteq Securities AG, where he is involved in modelling and financial engineering for all asset classes. His interests are numerical methods in mathematical finance, in particular multi-asset and hybrid modelling, as well as the impact of regulation onto markets and models. Backed by his PhD was on Neural Network he recently rekindled his interest in machine learning methods, now applied to problems in financial engineering. Initially he worked on foreign exchange, where he is co-editor of a textbook about FX derivatives.


Machine Learning applied to SLV Calibration


We calibrate local stochastic volatility using the particle method developed by [Guyon, Henry-Labordere]. A critical step in this method is an estimation of the conditional expectation of the stochastic volatility process, given the realized spot. We reformulate this estimation as a non-linear regression at each time-step of the discretized process, which allows us to apply machine-learning (ML) techniques. We review appropriate ML techniques, compare results and Performance.

Dr. Yannik Misteli

A Brief Biography


Dr. Yannick Misteli leads the Advisory Analytics Team at Julius Bär in Zürich. The team develops new data based approaches with the aim of better supporting the relationship managers who serve the clients. During his PhD he gained experience in numerical high-performance computing, multi-objective optimisation algorithms as well as statistical mechanics. He was always keen to transfer his insights gained from academic research over to industry and is now enjoying working to shape the future of the financial sector.


Decision Trees in Machine Learning


Decision trees neither are the most sophisticated classifier nor are they accurate predictors. However, they are a powerful instrument to establish machine-learning techniques within a company as they are easily and intuitively interpreted. They can be used to gradually build up understanding and confidence in using machine learning techniques amongst (senior) management, introducing more sophisticated methods once this confidence has been achieved. We demonstrate the use of decision trees in the context of identifying clients that should be moved to a different service model. A simple partition tree classifier is used to model the different customers and hence the leaf nodes are investigated for misclassified clients.




Prof. Dr. Paolo Giudici

A Brief Biography


Professor of Statistics and Data Science at the Department of Economics and Management of the University of Pavia. His current research interests are: Financial networks, Financial risk management, Systemic risk, and their application to Cryptocurrencies and Fintech platforms. Director of the University of Pavia Financial Technology laboratory (formerly Data Mining laboratory) which, since 2001, carries out research and consulting projects, for leading financial institutions.


Scoring Models for Roboadvisory Platforms: A Network Approach


Due to technological advancement, robo-advice platforms have allowed significant cost reduction in asset management. However, this improved allocation may come at the price of a biased risk estimation. To verify this, we empirically investigate allocation models employed by robo-advice platforms. Our findings show that the platforms do not accurately assess risks and, therefore, the corresponding allocation models should be improved, incorporating further information, through clustering and network analysis.



Prof. Dr. Marc Wildi

A Brief Biography



Marc Wildi holds an M.Sc. in Mathematics from the Swiss Federal Institute of Technology (ETH) in Zurich; he obtained his PhD from the University of St-Gallen, Switzerland. After being lecturer in Statistics at the Universities of Fribourg and of St-Gallen, he began his current position as a Professor in Econometrics in 2002 at the Zurich University of Applied Sciences. His novel forecast and signal extraction methodology, the so-called Multivariate Direct Filter Approach (MDFA), won two international forecast competitions in a row. His current research interests involve applications of the MDFA to mixed-frequency (daily) macro-economic indicators and to algorithmic trading.  


FX-trading: challenging intelligence


FX-trading is widely recognized as one of the most challenging forecast applications with the range of methodological complexity reaching from appalling simplicity to frightening complexity. We here sweep through this methodological range by proposing a series of novel and less novel, linear and non-linear, intelligent and less so approaches, either in isolation or in combination. Empirical results are benchmarked against plain-vanilla approaches, based on the most liquid and therefore most challenging (FX-)pairs. R-users will be pleased to replicate results.

Alla Petukhina

A Brief Biography


Alla Petukhina holds a M.Sc. in economics from the Ural state university, Russia.  Since 2014 she has joined the Ladislaus von Bortkiewicz chair of statistics at the Humboldt-University in Berlin as a Ph.D. candidate. Her research interests are focused on asset allocation strategies and risk modelling for high-dimensional portfolios, investment strategies in crypto-currencies market.


Portfolio allocation strategies in the cryptocurrency market


Current study aims to identify pro and con arguments of crypto-currencies as a new asset class in portfolio management. We investigate characteristics of the most popular portfolio-construction rules such as Mean-variance model (MV), Risk-parity (ERC) and Maximization diversification (MD) strategies applied to the universe of cryptocurrencies and traditional assets. We evaluate the out-of-sample portfolio performance as well as we explore diversification effects of incorporation of crypto-currencies into the investment universe. Taking into account a low liquidity of crypto-currency market we also analyze portfolios under liquidity constraints. The empirical results show crypto-currencies improve the risk-return profile of portfolios. We observe that crypto-currencies are more applicable to target return portfolio strategies than minimum risk models. We also found that the MD strategy in this market outperforms other optimization rules in many aspects.



Dr. Damian Borth

A Brief Biography


Dr. Damian Borth is a computer scientist and head of the centre of competence for deep learning at the German research center for artificial intelligence in Kaiserslautern (DFKI). He was awarded his PhD in Computer Science at the TU Kaiserslautern and the center of competence for multimedia analysis and data mining (MADM). For his achievements Borth and his team  received various prizes, such as the McKinsey business technology award and the google research award.


Deep Learning & Financial Markets: A Disruption and Opportunity


Learning to detect fraud or accounting irregularities from low-level transactional data e.g. general ledger journal entries is one of the long-standing challenges in financial audits or forensic investigations. To overcome this challenge we propose the utilization of deep autoencoder or replicator neural networks. We demonstrate that the latent space representations learned by such networks can be utilized to conduct an anomaly assessment of individual journal entries. The representations are learned end- to-end eliminating the need for handcrafted features or large volumes of labelled data. Empirical studies on two accounting dataset support our hypothesis. We evaluated the methodology utilizing two anonymized and large scaled datasets of journal entries extracted from Enterprise Resource Planning (ERP) systems.



Prof. Dr. Andreas Hoepner

Prof. Dr. Andreas Hoepner


Professor Andreas G. F. Hoepner, Ph.D., is a Financial Data Scientist working towards the vision of a conflict-free capitalism. Formally, Dr. Hoepner is Full Professor of Operational Risk, Banking & Finance at the Michael Smurfit Graduate Business School and the Lochlann Quinn School of Business of University College Dublin (UCD).

Andreas is also heading the ‘Practical Tools’ research group of the Mistra Financial Systems (MFS) research consortium which aims to support Scandinavian and global asset owners with evidence-based tools for investment decision making.

Furthermore, he is currently a visiting Professor in Financial Data Science at the University of Hamburg, as well as at the ICMA Centre of Henley Business School, were he was an associate Professor of Finance.


Finance Keynote Talk: Embracing AI Opportunity = (Humans*Teamwork)^Machine -1


Prof. Hoepner argues that AI and augmented Intelligence (AugmI) can both have huge potential, with the use case suggesting which one to deploy, and how to organize ones team. He cites Gary Kasparov's observation that the Freestyle Chess Championships were won neither by the best grand master nor the best machine but by the best human-machine team: two amateur chess players using three machines simultaneously. Based on an immediacy, confirmability, population size and time-series attributes of use cases, he argues that AI is superior for real-time repeated recognitions of static objects, while AugmI is likely to remain preferred choice for quite a while in regular but not real-time predictions of reactive processes. AugmI also demands a strong focus on creating an excellent teamwork between among all involved humans and them and their machine(s).

Lastly, Prof. Hoepner connects both AI abbreviations with the pressing need of climate change mitigation based on the use case of corporate GHG emissions reporting.

Dr. Christian Spindler

A Brief Biography


Christian Spindler is Data Scientist at PwC’s Artificial Intelligence Center of Excellence in Zürich. Educated in physics, Christian Spindler accumulates 10+ years of experience with machine learning and deep learning modelling for various applications in financial services, insurance, manufacturing and robotics.


Explainable and responsible AI for financial Services and Insurance


Decisions taken by machine learning / artificial intelligence systems cannot be assessed using traditional source code analysis. The final behavior of such algorithms is defined during training, not during programming. PwC develops services based on LIME, QII and other assessment techniques for black-box systems to build trust in AI. We analyze the influence of individual input factors and search for bias, possibly leading to overreactions on specific inputs. We determine whether algorithms are fully trained and identify potential risks of developing an algorithm into production.

We demonstrate a case study for responsible AI in Financial Services, particularly the insurance of autonomous, driverless vehicles. We cover various aspects of trustful AI systems and show how to integrate them in daily business operation.

Dr. Martin J. Fengler

A Brief Biography


Martin Fengler studied at TU Kaiserslautern (Germany) where he received his PhD in applied mathematics. After his studies he developed several numerical weather prediction codes for Meteomedia AG (now MeteoGroup Switzerland) where he became responsible for the technology & innovation department. In spring 2012 he decided to found Meteomatics AG in St. Gallen. Meteomatics focuses primarily on weather solutions for industry. One of its USPs is beside the Meteodrone technology an in-house developed in-memory caching solution which enables a simple real-time weather data access to a huge variety of data: This weather API enables industry to analyze business data and to optimize production. 



Big Data meets weather: How real-time access to weather data enables a rapid development of business applications


The availability of quality weather data has improved dramatically over the past decade. At the same time the number of big data analytics businesses delivering sector-specific solutions and business insights has also grown dramatically. However timely access to quality weather data, as cut outs that are suited to specific business requirements, delivered in formats that users can simply apply to new and existing in-house systems and models has remained a challenge. Meteomatics is a commercial weather data provider that is working collaboratively with National Met Services (NMSs), Academia and Scientific communities. We bring together historical, nowcast and forecast weather data from global models such as the ECMWF model, satellite operations and station data. By applying in-house modelling and downscaling capabilities, Meteomatics is able to deliver weather data for any lat / long and time series to use in 3rd party models via an industrial scale robust Weather API. Weather data enabled insights are relevant to many sectors in industry. For instance, aviation, automotive, energy, logistics, insurances and others, both public and private: In the field of agriculture, weather risk management solutions are already protecting the crops of farmers across Africa from drought and innovative start-ups around the globe are applying weather data to a variety of models to meet precision farming challenges. Energy companies, both in the traditional and renewable sectors, are extensively using these solutions to forecast demand, power output, inform energy trading, protect themselves against unfavourable seasons and safeguard revenues. Meanwhile, wind farm operators seek protection against low or excessively strong wind to secure cash flow and underpin their financing. Marine insurers are combining vessel tracks and crew behaviours in differing weather conditions to influence their view of risk, and Lloyd’s of London are using historical weather data and ship tracks to identify fraudulent marine claims. Water utilities are enhancing demand and leakage models, better managing system capacity and ensuring regulatory compliance through weather-enabled automation of alarms, catchment modelling and enhanced workforce management. So, in summary, simple API access to quality weather data is extending the understanding of weather risk for a broad range of sectors. The speed of development of new products and services underpinned by quality weather data, indices, benchmarks and parametric triggers is growing rapidly.


Please find here the form to register for the «3rd European COST Conference on Artificial Intelligence in Industry and Finance».


Travel to ZHAW School of Engineering, Winterthur

From the Zürich airport local transport brings you to the centre of Winterthur in just 20 minutes.

The conference takes place in a short walking distance to the train station. Numerous hotels in the immediate vicinity will ensure a pleasant stay in Winterthur

By train
The trains run as often as every 15 to 20 minutes from Zürich Airport and Zürich City and take about 15 minutes to arrive in Winterthur.

From the main railway-station, the conference location can be reached within a walking distance of less than 5 minutes.

By Car

Organizing Committee

Program Comittee