A Brief Biography

Peter Mattson is a Google senior staff engineer. He co-founded and is President of MLCommons, and co-founded and was General Chair of the MLPerf consortium that preceded it. Previously, he founded the Programming Systems and Applications Group at NVIDIA Research, was VP of software infrastructure for Stream Processors Inc (SPI), and was a managing engineer at Reservoir Labs. His research focuses on understanding machine learning models and data through quantitative metrics and analysis. Peter holds a PhD and MS from Stanford University and a BS from the University of Washington.

MLCommons: driving ML innovation with benchmarks

MLCommon.org is a non-profit with a mission to make ML better for everyone by building the ML ecosystem, principally benchmarks, datasets, and best practices. The talk will describe several benchmarks with applicability to finance, including the MLPerf speed benchmark, the DataPerf data quality benchmarks, and the MedPerf benchmark approach to federated evaluation. It will also speculate on the potential for benchmarks in the finance industry.

A Brief Biography

Ekaterina is a senior data scientist at NVIDIA specialized in solving computer vision and video analytics problems by means of AI. Her current focus also includes deep learning inference optimization on embedded devices. Previously, Ekaterina was a research engineer applying deep learning to medical image analysis. She has also authored several peer-reviewed journal and conference publications on various applications of image-based 3D reconstruction, localization and tracking. Ekaterina received her Ph.D. in Computer Science and M.Sc in Media Informatics, she also holds a Diploma in Business Informatics.

Synthetic Data and Simulation Workflows for AI Training in Robotics

Developing robots capable of perceiving complex environments in a scalable way is not a trivial task. Relying on AI, robots become more intelligent, flexible, and robust, but AI training requires lots of data. Data collection and annotation are time consuming and expensive. Furthermore, in order to create models which generalize well, data needs to be diverse and balanced. With advancement in simulation tools and generative models, more and more AI practitioners start utilizing synthetic data as a possible alternative to real data. In this session, we will focus on NVIDIA’s approach to synthetic data generation and robotics simulation and will showcase the underlying value of our tools for realistic and accurate data creation.

A Brief Biography

Dr. Martina Perani holds a PhD degree in Physics from the University of Bologna, where she focused on the optoelectronic and morphological characterization of semiconductors for photovoltaic applications. She then moved to the Fernfachhochschule Schweiz (FFHS), where she further developed her competences in the field of data science and artificial intelligence. Currently, she leads the research field “Data Science for Energy, Environment and Materials” at the FFHS. Her research interests lie in the application of artificial intelligence to address research questions in these technical disciplines, with projects in the areas of additive manufacturing, photovoltaics and warning systems.

Deep neural networks for predicting track geometry in additive manufacturing

Laser Metal Deposition (LMD) is an additive manufacturing technology that attracts great interest from the industry, thanks to its potential to realize parts with complex geometries in one piece, and to repair damaged ones, while maintaining good mechanical properties. The complexity of the LMD process has however limited its widespread adoption so far, as different part geometries, strategies and boundary conditions can yield very different results in terms of external shapes and inner flaws. Deep learning based architectures have been used to estimate the size of tracks deposited by LMD using powders of Inconel 718, a relevant material for the aerospace and automotive sectors. An architecture that combines the output of a convolutional neural network (that takes melt pool images as inputs) with scalar variables (process and trajectory data) is used to predict the track geometry. Moreover, preliminary results suggest that long-short memory networks are effective in modeling the temporal evolution of the deposition process using process and trajectory data.

A Brief Biography

Sebastian Welter is leading AI innovation at IKEA. His background is classical IT, being an IT architect by profession. His first steps in AI were taken at IBM, when he was responsible as architect in Germany for the AI that started the hype in 2011 - IBM’s Watson. From there he went into consulting, spending some years in defining AI strategies and solutioning analytics/cloud/data/AI projects as a director at Accenture. Now his focus is technology driven innovation, whether its AI or any other fascinating digital technology.

Innovate a retailer with AI: The life of an innovation manager in IKEA

Everyone “does AI” nowadays, and the larger the company, the bigger the fails and maybe the achievements are. But how do you do it? Follow me on a typical day in innovation in IKEA on our journey to create a better everyday life for the many people. The talk will touch on how AI and innovation can change a company, what challenges to overcome and how to make it all work in grand scale.

A Brief Biography

Dr. Nikola Pascher is the global head of Mettler Toledo's R&D team in the strategic business unit standard industry. She is responsible for the product portfolio, consisting of weighing solutions for industrial applications, controllers and software, covering the entire technology-stack from the sensor to the cloud. She studied physics at the University of Augsburg in Germany. Her phd at ETH Zurich, Switzerland and a consecutive postdoctoral assignment at IBM Research, Zurich were focused on quantum science. After leaving academia, she worked in various positions in technology and innovation management with a focus on high-precision measurement instrumentation and data science for digital solutions and services. She extended her knowledge with an Executive MBA from the University of St. Gallen. Her main interest is to create value with breakthrough technologies.

Dr. Thilo Stadelmann is professor of artificial intelligence and machine learning at the ZHAW School of Engineering in Winterthur, director of the ZHAW Centre for Artificial Intelligence and head of its Computer Vision, Perception and Cognition Group. A computer scientist by training, he received his doctor of science degree from Marburg University in 2010, where he worked on multimedia analysis and voice recognition. Thilo worked in engineering and leadership roles in the automotive industry for 3 years prior to switching back to academia. His current research focuses on robust deep learning to solve diverse pattern recognition tasks such as biometrics, document analysis or computer vision for industrial and medical applications. He is fellow of the European Centre for Living Technology in Venice, Italy, co-founder and board member of Switzerland’s largest innovation network, the data innovation alliance, and of the ZHAW Datalab, one of Europe’s first dedicated research hubs for data science. His book "Applied Data Science - Lessons Learned for the Data-driven Business" was published 2019 by Springer, a pre-print on "A Theory of Natural Intelligence" that informs a future roadmap for machine learning research was released recently.

Data-centric transfer learning as a solution to key-challenges in industrial data-based solutions

To date, most of the applications of machine learning in industrial companies have an inward focus. Methods form data science are used to simplify company processes, place efficient marketing, automatic digitalization of documents and many more. The industrial internet of things instead, is focused outwards, towards the customers with the goal to create true value from their data, which is obtained, e.g., from running industrial applications. IIoT pictures a beautiful perspective, yet many companies struggle to offer data-based solutions, which are based on measured sensor-data from industrial processes. Digital innovators meet a variety of obstacles, which prevent them from efficiently developing solutions which turn data from the customer’s production sites into value. Taking Mettler Toledo’s industry division as an example, we picture the pathway of a very traditional industry company extending their technology stack to the top.

A general problem in industrial applications of machine learning is that (labeled) training data is rarely available. Either the creation of labeled ground-truth data sets is too expensive, or the data belongs to the customers, who might be worried to share their data with a solution provider. On top of that, machine learning models do not generalize efficiently. Model performance is poor, if the pre-trained model is applied to different data sets it was not trained for initially. This means a big pain to the solution provider and to the customer, as extensive model training would have to be included in the customer workflow.

We focus on a data-centric approach in combination with transfer learning as a possible solution scenario for the above-mentioned challenge. We show an example from medical imaging, where we improve the performance of automated image recognition of the spine with an unsupervised domain adaptation (UDA) approach for vertebrae detection and identification based on a novel Domain Sanity Loss (DSL) function.

This example can be abstracted to a variety of different processes and could be a valid solution to facilitate the use of machine learning in industrial processes.

A Brief Biography

As a financial economist and data scientist I focus on computational finance, financial risk management, impact investing, natural language processing, extreme value theory, econlinguistics and systemic risk. As a lecturer at Frankfurt School of Finance & Management, I teach courses in financial economics and derivative pricing. I am a co-founder of two companies, namely Axiomatec AG, a fin-tech company that develops quantitative investment and risk management solutions for institutional clients. As well as Hiveanalytica UG that focuses on business analytics.

"Let me get back to you" - A machine learning approach to measuring non-answers

Using a supervised machine learning framework on a large training set of questions and answers, we identify 1,364 trigrams that signal non-answers in earnings call Q&A. We show that this glossary has economic relevance by applying it to contemporaneous stock market reactions after earnings calls. Our findings suggest that obstructing the flow of information leads to significantly lower cumulative abnormal stock returns and higher implied volatility. As both our method and glossary are free of financial context, we believe that the measure is applicable to other fields with a Q&A setup outside the contextual domain of financial earnings conference calls.

A Brief Biography

Professor Dr. Blanka Horvath's current research interests focus on asymptotic and numerical properties of stochastic volatility models, in particular models with rough paths (rough volatility), as well as using generative methods -across models- to generate synthetic financial market data. For her contributions to the calibration of (rough) stochastic volatility models with deep neural networks, she received the Rising Star Award from Risk Magazine in 2020. Blanka Horvath was appointed Professor at the Chair of Financial Mathematics at the Technical University of Munich in April 2021.

Machine Learning - Market Simulation Generating Financial Markets with Signatures

Market Simulation and Deep Hedging: Neural network based data-driven market simulation unveils a new and flexible way of modelling financial time series, which has recently inspired a surge of research activity in the quantitative finance community. While most generative models tend to rely on large amounts of training data, we present here a generative model that works reliably even in environments where the amount of available training data is small, irregularly paced or oscillatory. We show how a rough paths-based feature map encoded by the signature of the path outperforms returns-based market generation both numerically and from a theoretical point of view. Finally, we also propose a suitable performance evaluation metric for financial time series and discuss some connections of our signature-based Market Generator to deep hedging.

A Brief Biography

Linus Grob is a research associate at the Institute for Wealth and Asset Management at the ZHAW School of Management and Law and a business analyst at the Bank Vontobel AG. He graduated in Banking and Finance at the ZHAW School of Management and Law and is now pursuing his Ph.D. in Banking and Finance at the University College Dublin. His research focuses on machine learning applications in capital markets and the application of alternative data in Finance.

Dominik Boos is a lecturer at the Institute for Wealth and Asset Management at the ZHAW School of Management and Law. His teaching focuses on active and quantitative investment strategies as well as on sustainable investing. Dominik has more than twenty years of industry expertise in strategy research and active investment management. He is also an investment advisor for Serafin Asset Management and founded Avena Capital in 2017. Dominik worked as Head of Global Macro Research at SSGA where he was a member of the global macro investment committee and directed the development and the implementation of the relative value book. Before, Dominik worked as a Senior Investment Strategist at the Swiss National Bank, responsible for the strategic asset allocation of the foreign reserves. Previously, he was Head of Quant Research at Zurich Cantonal Bank. Between 2004 and 2009, Dominik worked at Kraus Partner, a Zurich-based asset management company. As Partner and Head of Investment Management he was responsible for research, portfolio management and trading.

Dominik has earned a PhD in Finance and Economics from the University of St. Gallen and a Masters in Mathematics from the Swiss Federal Institute of Technology.

Tracking Speculative Trading

Managed futures funds are predominantly trend-followers. By analyzing positioning data, we provide novel evidence for this claim and estimate signals applied by these funds. We write trend-followers aggregate position as a weighted sum of past daily returns and use a generalized ridge regression for regularization and parameter estimation. This procedure prevents overfitting but remains flexible enough to capture various patterns. For the 23 commodities considered, trend-following can explain speculators' position changes with an average $R^2$ of more than 40 percent. Finally, we document that producers act as contrarians in a way that closely mirrors the behavior of momentum traders.

A Brief Biography

Thomas Krabichler is a lecturer and quant researcher at the OST – Eastern Switzerland University of Applied Sciences. He holds a doctoral degree from ETH Zürich in mathematics. His research focuses on applications of machine learning in finance. Previously, he worked as a quant specialist for the pricing and hedging of financial derivatives. In this role, he provided technical model validations of valuation models for major investment banks in London and Paris.

Deep Asset-Liability-Management

Recent advances inspired by reinforcement learning support well-balanced risk-taking and unprecedented risk-adjusted pricing. In ongoing initiatives with several colleagues and partners, we are utilising these techniques in order to tackle high-dimensional optimisation problems in the presence of constraints and frictions. The range of applications involves optimal decision making for treasurers, optimal procurement of commodities, or the optimisation of hydroelectric power plants.

A Brief Biography

Matthias Niggli is a research data scientist and project manager at the Center for International Economics and Business | CIEB of the University of Basel. He holds a PhD from the University of Basel, has a special interest in experimenting with all kinds of machine-learning methods and is enthusiastic about discovering novel approaches to visualize data.

Christian Rutzer is deputy director and project manager at the Center for International Economics and Business | CIEB of the University of Basel. He holds a PhD in economics from the University of Basel and likes to apply new methods from data science to analyze economic issues.

What is it worth? Predicting the Importance of Patents Using Supervised Learning

We demonstrate that a combination of deliberate feature engineering and deep learning methods can be used to predict the technological importance of recently filed patents. This provides a scalable framework for early assessments of companies’ current innovation pipelines.

A Brief Biography

Prof. Dr. Marc Wildi is Professor of Econometrics at the Zurich University of Applied Sciences. His research interests are about forecasting, real-time signal extraction, business-cycle analysis, algorithmic trading and risk management. His recent work emphasizes hybrid approaches (mixing real-time filter designs and generic trading concepts) as well as explainability (XAI) of computationally intensive approaches (NN) in the context of longitudinal data (time series).

Prof. Dr. Branka Hadji Misheva is a Professor in Applied Data Science and Finance at BFH, working on AI applications in finance, XAI methods, network models and fintech risk management. She holds a PhD in Economics and Management of Technology with a specific focus on network models as they apply to the operation and performance of P2P lending platforms, from the University of Pavia, Italy. In the past, she has led several research and innovation projects on Artificial Intelligence and Machine Learning for Credit Risk Management (Innosuisse, EU Horizon 2020, COST Action). She is a research author of over 17 papers in the field of credit risk modeling, graph theory, predictive performance of scoring models, lead behavior in crypto markets and explainable AI models for credit risk management.

A Time Series Approach to Explainability for Neural Nets with Applications to Risk-Management and Fraud Detection

Artificial intelligence (AI) is creating one of the biggest revolution across technology-driven application fields. For the finance sector, it offers many opportunities for significant market innovation and yet broad adoption of AI systems heavily relies on our trust in their outputs. Trust in technology is enabled by understanding the rationale behind the predictions made. To this end, the concept of eXplainable AI (XAI) emerged introducing a suite of techniques attempting to explain to users how complex models arrived at a certain decision. For cross-sectional data classical XAI approaches can lead to valuable insights about the models' inner workings, but these techniques generally cannot cope well with longitudinal data (time series) in the presence of dependence structure or non-stationarity. In this talk I will present common work with Branka Hadji about a novel XAI-technique for deep learning methods (DL) which preserves and exploits the natural time ordering of the data. After a brief introduction to the main concepts, simple applications to financial data illustrate the potential of the approach in the context of risk-management and fraud-detection.

A Brief Biography

Ricardo Chavarriaga, PhD, is a neuroscientist passionate about responsible development and societal implications of technology. His work is focused on fostering research and innovation in emerging technologies towards the benefit of society.

Dr Chavarriaga is senior researcher at the Zürich University of Applied Sciences (ZHAW) and Polymath fellow at the Geneva Center for Security Policy (GCSP). He is the head of the Swiss office of the Confederation of Laboratories for AI Research in Europe (CLAIRE). CLAIRE is the largest European network on Artificial Intelligence (AI), comprising more than 400 research groups and 3’000 individual supporters. He chair the principles sub-group of IEEE P2863 Working group on Recommended Practices for Organisational Governance of AI.

Operationalizing AI ethics: Challenges for organizational governance

Upcoming regulatory frameworks for Artificial Intelligence (AI) systems will require organizations to certify that the systems they develop comply with requirements based on ethical principles such as transparency, fairness reliability, and safety. Making the bridge between these ethical principles and organizational processes is not an easy feat. This talk will discuss this challenge and present current examples of initiatives aimed at tackling it. 

A Brief Biography

Eleonora Viganò is Postdoctoral Researcher at the Institute of Biomedical Ethics and History of Medicine, at the University of Zurich. She was recently the Executive Manager of the ELSI-Task-Force for the Swiss National Research Programme 75 “Big Data” and developed the moral aspects of the Swiss National Research Programme 77 project “Socially acceptable AI and fairness trade‐offs in predictive analytics”.

Her main research areas are Neuroscience of Ethics and Digital Ethics. She is currently working on the ethical issues of digital technologies aiming to improve people’s well-being.

Her latest publication is the book “Moral choices for our future selves” for the series Routledge Focus on Philosophy.

Ethics of algorithmic prediction: How algorithms threaten our right to be unpredictable 

Predictive algorithms give an unprecedented support for decision-making in high-stakes domains such as the assignment of medical treatments, jobs, loans, and parole, as well as for everyday choices such as the suggestion of what news to read or movie to watch. Yet, such a support comes at ethical costs concerning privacy, lack of transparency, discrimination, profiling, and choice manipulation. In this talk, first, I will discuss an ethical issue of predictive algorithms involving the essence of being a person in the sense of an agent able to make choices autonomously. This ethical issue is due to an aspect of the design of predictive algorithms: a predictive algorithm treats people as mere members of a group it assigns them to, thus overlooking their uniqueness deriving from being autonomous agents and the possibility that their behavior deviates from the group they are assigned to. Then, I will defend the right to be unpredictable, which requires that we consider the possibility that the subject of an algorithmic decision can be an exception to the algorithm’s prediction, and analyze some decisional contexts in which this right is applicable.

A Brief Biography

Jérôme Koller, Head of Arena Steuerung, La Mobilière

Jérôme is a passionate artificial intelligence professional committed to delivering value from data at scale in the organization. He has built and led a team of 25 scientists implementing projects across La Mobilière and aiming to a broad adoption of AI within the company. Jerome has previously held various positions in the field of data at La Mobilière, from risk management to data warehousing, business intelligence and customer analytics. He currently is in charge for the strategic steering of the agile organization of La Mobilière. He studied mathematics and informatics at the University of Fribourg and completed his EMBA degree at IMD Business School in Lausanne in 2022.

You have your ethical framework, so what?

Many companies are incorporating AI and data-driven solutions in their toolkit to create value for their customers. Using more data usually leads to ethical questions and can raise passionate internal debates about what to do or not. Is the ethical questioning coming from the more innovative side of the company, arguing that you must define such policies today to "stay tuned" with society, or is it a way for the more conservative employees to question the transformation?
At La Mobilière, we defined our ethical framework in 2019 and have been using it since then. I will give an overview of the ethical principles of La Mobilière, explain how we use them, and give my lessons learned from this journey. By focusing on which problems the framework can solve and which ones, in my opinion, are still open, I will highlight some underestimated values of clarifying ethical thoughts in Business.

A Brief Biography

Prof. Dr. Patricia Deflorin is Research Director at the Swiss Institute for Entrepreneurship (SIFE), FH Graubünden, and lecturer at the University of Zurich. Within the NTN Innovation Booster Databooster, she is the co-leader of the focus topic Industry 4.0. Her research focus on innovation, supply chain and digital transformation. Her area of experience includes the design and implementation of data-based services, the generation and analysis of IoT business models and the identification of necessary technologies for the successful implementation of digital transformation.

Databooster – Open innovation activities to boost data based innovation

Technologies, data analysis, interfaces and standards. Companies pursuing Industry 4.0 and smart services are facing many decisions on how to pursue their activities. But which investments may lead to a radical innovation? The NTN Innovationbooster Databooster supports companies in their quest to develop and implement databased innovations. Patricia Deflorin shows which topics industrial companies seem as relevant to pursue and how these topics may differ between small and larger companies. In addition, she sheds light into the Databooster activities and how industrial companies can profit from the networks activities.  

A Brief Biography

Hanspeter Bär's background is in Electrical Engineering (Dr. sc. techn./ETHZ in Communications Technologies) and Business Management. He spent his business life in 2 international Swiss companies (Zellweger Uster and Huber+Suhner), in various positions from Sales Engineer, R&D Manager to Division Manager and Member of the Group Management. Since his retirement, he is active as an Innovation Mentor for Innosuisse. In this role, he has coached more than 300 project teams, supporting them in writing a convincing application for Innosuisse funding of an innovation project.

How to be successful with your Innosuisse Application for innovation project funding

I will present the key instruments Innosuisse is offering for project funding. Furthermore, I will talk about the criteria for a successful application.

A Brief Biography

Arman Iranfar received his PhD in electrical engineering from EPFL in 2020 with a focus on applied machine learning in embedded systems. He has published over 20 peer-reviewed papers in top-notch conferences and journals and served as a reviewer in several top-ranked conferences and journals including IEEE TC, IEEE TSUSC, and IEEE TSC. He is currently a senior data scientist at CertX, building the first certification scheme for AI trustworthiness.

Amin Amini is leading the functional safety expert of CertX since September 2018. Based on his strong background in telecommunications and automotive industry, he has developed functional safety certification schemes for supporting customers to meet the State-of-the-Art, as well as to demonstrate their best practices to authorities across different market segments. Amin is an ISO 26262, OPST and IEC 61508 certified engineer and expert member of ISO 26262 2nd edition and ISO/PRF PAS 21448 (SOTIF) standardisation group.

certAInty - A Certification Scheme for AI systems

Certification of AI Systems by an accredited body increases trust, accelerates adoption and enables their use for safety-critical applications. Under the Innosuisse umbrella, CertX is developing a Certification Scheme comprising specific requirements, criteria, measures, and technical methods for assessing Machine Learning enabled systems.

Prof. Dr. Peter Schwendner leads the Institute of Wealth & Asset Management at Zurich University of Applied Sciences, School of Management and Law. He collected 15 years of work experience in the financial industry as a head of quantitative research at Sal. Oppenheim and as a partner at Fortinbras Asset Management after he completed a doctorate in physics. He has been developing analytics for primary and secondary markets and quantitative risk premia strategies with financial industry partners.

Jennifer Victoria Scurrell is a PhD candidate at the Center for Security Studies (CSS) at ETH Zurich. In her dissertation, she examines the influence of AI agents on political opinion formation in online social networks using experimental methods and ML. Jennifer is a political scientist by training and holds a MA and a BA from the University of Zurich.

Pius von Däniken is a research assistant at the ZHAW Centre for Artificial Intelligence. As a member of the Natural Language Processing group, he focuses on malicious behavior in Social Media and the dynamics of online radicalization. His other research interests include the evaluation of dialogue systems and the development of  NLP tools for Swiss German. He holds a BSc in Computer Science from EPFL and is currently a MSE student at the ZHAW.

• Prof. Dr. Patricia Deflorin
• Prof. Dr. Rudolf Füchslin
• Prof. Dr. Branka Hadji Misheva
• Dr. Andreas Henrici
• Prof. Dr. Peter Schwendner
• Prof. Dr. Thilo Stadelmann
• Prof. Dr. Dirk Wilhelm