We will cover:
In this workshop, led by world-renowned experts, we will explore practical considerations when building artificial intelligence (AI) and machine learning (ML) solutions for financial institutions. We will provide you with a powerful frameworks and tools to become an expert in your organization.
During the rigorous hands-on exercises, we will internalize best-in-class tools and templates for implementation, maintenance and the buy-in of the your peers and the stakeholders for the artificial intelligence solutions of your choice.
We will also learn how to structurally solve complex legal and compliance challenges arising due to the use of automated decision making systems and artificial intelligence systems.
1. Contrast Top Machine Learning Methods
Sticky, insightful and simple ways of explaining to others the conceptual, mathematical and legal contrasts between the top ML methods: logistic regression, gradient boosting and neural networks and the performance metrics behind them
2. Self-Learn, Self-Execute and Self-Deploy Models: How?
The 'holy-grail' of machine-learning: the self-learning, self-executing and self-deploying models: do they exist at all and, if yes, how to maintain them?
3. Model Building: 'Quick & Dirty' vs. 'Slow & Clean'
When you can build 'quick & dirty' and when you need 'slow & clean' model and why it matters - e.g., why mindlessly focusing on the final ROC value is not good and looking at the ROC curve shape function is often equally important
4. Optimizing your Model Beyond GINI
How to turn-off your autopilot of mindlessly optimizing your models on the standard performance metrics (AUCs, GINIs, loss rate, R2, etc.,), how you can cheat them, why understanding confusion matrix and sensitivity/accuracy/specificity/F1-score is more important and why proper validation design through out-of-time/of-sample tests is a must
5. AI/ML Fit into your IT Architecture
Fitting the IT architecture and platforms (databases, software, tools) into the scope of your AI/ML activities and understanding when Excel is enough (you knew that you can run a neural network in Excel, right?)
6. Optimize for KPIs you Understand
How to ensure that your model is optimized for KPIs you understand, (customer-life-time value, cross/up-sell, churn, expected loss, etc.) and not for some typical data science mumbo jambo mathematical constructs which are often three-degrees removed from your quarterly business targets
7. Proof-of-Value for your A.I. Solution
Getting stakeholders buy-in and board/steerco approvals through sculpting a compelling financial impact analysis for a final go/no-go decision:
how to estimate the marginal added value of your AI/ML solution over other competing solutions through backward- and forward-looking what-if-and the NPV (net present value)- scenarios
modelling across number of scenarios such as with vs. without historic financial data (existing/innovative solution) or record- vs. portfolio- level estimations (slow/fast execution)
8. In-House, Buy or Partner for A.I. ?
Deciding between: in-house developments, hiring new staff, in/outsourcing, buying software, partnering up, creating joint ventures, and acquiring start-ups and hence:
covering possible implementation scenarios on - existing infrastructure, micro-services, internal/external cloud, SaaS, etc.
understanding legal remedies and warranties/guaranties on commercial vs. open-source software
9. End-to-end (E2E) cycles of building, deploying and maintaining robust ML model
Best-in-the class practices of end-to-end (E2E) cycles of building, deploying and maintaining robust ML models across your organizations: why everyone wants always to build the models but none ever wants to implement and maintain them
10. Why most pilots and POCs become final solutions?
Why most pilots and POCs become final solutions and how to stagger and phase them correctly at your own organization or on-site at clients
11. De-biasing algorithms and making them fair again
De-biasing algorithms and making them fair again: managing the model's likely discriminatory misjudgments, about such sensitive user-characteristics as gender, ethnicity, religion or age, which mainly stem from the ill-thought mathematical assumptions and lack of explicit rules for tackling model's hidden correlations
12. How to explain your model's decisions in a compelling way ?
How to explain your model's decisions in a compelling way to the end-users in a compliant manner but without revealing the inner workings of the model and what the current legislation actually mandates so you don't throw the baby out with the bathwater
13. Who is end-responsible for the data processing - you or the party you engage?
Who is end-responsible for the data processing - you or the party you engage? The crucial difference between the data controller, data processor and how to identify them. Is possible to indemnify oneself against GDPR fines by holding accountable the data processor or data controller, can you split the fine or buy an insurance against it?
14. Methods of anonymizing data
Best-in-the-class and industry-standard ways of how to anonymize your data and what is a difference between fully anonymized data and pseudo-anonymized data ?
15. Biometric and sensitive data
Is it worth processing biometric or sensitive personal data ?
Any views or opinions presented in this workshop are solely those of the author and do not necessarily represent those of the organizations.
Who Will Deliver this Workshop?
Peter Lewinski, PhD
Practice: Banking, Fintech
Peter helped to found a deep-learning SaaS A.I. startup, worked in startups, banks, fintechs and universities across London, Amsterdam, Madrid, Leuven, Switzerland and Warsaw. He received 3-year long Marie Sklodowska-Curie Fellowship (European Commission) of +250 000 EUR to study research consumer behavior using A.I. solutions.
In banking domain, he gained experience as a Deputy Director of Credit Risk Modelling Department at ING Bank where he led a team of 10 data analysts building in-house machine learning / A.I. solutions from internal and alternative big data sources as well as scouted, chose and implemented fintech start-ups (e.g. twisto.cz, finansowaniefaktur.pl). In fintech domain he gained experience as a Programme Director at Oxford Fintech & SmartLaw Society and as a Summer Associate in a London Mayfair-based alternative credit risk scoring fintech FriendlyScore (friendlyscore.com).
From September 2018, he joined a leading boutique A.I. software house – smr.nl, where in his new role as a management team member, he is responsible for strategy and business development of the proprietary deep learning and machine learning solutions, implemented already at Fortune 500, financial institutions, law enforcement authorities, Ivy League universities and startups.
Peter is a graduate of Oxford University. He received his PhD from University of Amsterdam, was a postdoc at Université de Neuchâtel, and made a tenure-track Assistant Professor at ESCP Europe (Warsaw Campus, Kozminski International Business School). Since 2011, Peter has spoken at 20+ international conferences, co-authored +12 publications, his scientific work on behavioral economics has been cited hundreds of times and received numerous scholarships (European Commission, Council of Europe, Dutch Science Foundation).
Practice: Banking, Fraud, Crime
A veteran business person who is always looking for the next big challenge, Pieter Bison has helped founding our company in 2003. He has 20+ year of experience in data mining. He delivered A.I. / data mining KPIs to clients across banking, insurance and fraud/crime related data analytics cases. He has multiple language skills programming, IT skills, platforms, courses certificate and newspaper mentions and awards.
With his passion and expertise, Pieter has managed to take our company to new highs and achieve growth. Check out our contact details for lectures or press opportunities with Pieter Bison.
Practice: Deep neural networks
Amogh is a machine learning research engineer and he has 5+ years of working experience in deep neural networks. Since 2013 Amogh has been working in a Dutch A.I. start-up, where he has worked on projects such as recognizing semantic features using deep learning or weakly supervised object localization. Before that he was a member of Dutch Nao Team achieving 3rd place at Iran Open 2013 – Tehran and top-16 finish in 2013 - Eindhoven. He was also a systems engineer for Tata Consultancy Services.
Amogh is an accomplished scientists. In 2019 he will finish his PhD on data computational efficiency in deep neural networks from Delft University of Technology, the Netherlands. In 2014 He obtained his Masters in Artificial Intelligence from the University of Amsterdam as well a Bachelor of Engineering from technological university. His work has been cited hundred times and he has presented in dozens of international conferences.
Come Fly with Us
At Sentient, we’re passionate about training executives who will then go on disrupting their industry. Since 2003, we’ve provided a high quality workshops to all of our clients and as we grow, our commitment to your needs remains the key driving force behind our software startup. Our team of professionals is here to inspire you with their unique ideas and abilities - get in touch today to learn more.