The recent ReWork Deep Learning in Finance Summit in London, UK, featured 46 top scientists and professors from world-leading institutions, who presented their research progress and provided a glimpse into emerging trends in the field of artificial intelligence and fin tech.
The first speaker was Jackson Hull from British financial services company GoCompare Group, a financial services, utilities and home services comparison platform. Hull was also CTO at Plum District in San Francisco, Global Student Accommodation Marketplace, and the upmarket Airbnb-like accommodation service Onefinestay. He received his Bachelor’s degree in Mechanical and Material Science Engineering at University of California, Davis, and a Master’s in Information Systems from the University of California, Berkeley.
Hull spoke on the use of deep learning to improve customer experience, leveraging transactional datasets to provide customers with more value from financial services, and why AI will be transformative in fin tech. He explained how GoCompare is partnering with fin tech innovators, sharing best practices in machine leaning and data science, and developing AI-enabled APIs.
Hull believes “open learning” is the new open source. Open collaboration in machine learning is a natural step: data scientists contribute useful data, share tested and innovative approaches to artificial intelligence, and form communities and create synergy between technical specialists.
The next speaker was Huma Lodhi, a principal data scientist from Direct Line Group. Lohi has 15 years of experience in the field of AI and machine learning across both industry and academia. She is an expert with hands-on experience in development and application of deep learning, kernel methods, relational learning and ensemble methods for areas ranging from insurance to health care. Huma has a PhD in Machine Learning from the University of London.
Huma shed light on the importance of scene understanding technologies for insurance and risk management, and talked about the successful applications of deep learning for different tasks ranging from risk modeling to claim settlement. She sees intelligent methodologies driven by deep learning as a key focus of academic researchers and industry practitioners.
Huma compared the advantages of classical statistic models and machine learning. Statistical methods are usually based on parametrics, require prior knowledge, and perform well on small datasets and few variables. Machine learning meanwhile involves non-parametric methods, needs no prior knowledge, and is fed with big data and many variables. Most importantly, machine learning is able to generalize well on new data compared with statistical methods.
The third speaker, Manuel Proissl is the head of predictive analytics at UBS. He was a senior advisor and the AI cloud platform lead at Ernst and Young with a focus on AI-driven business solutions for international companies. He has also taken managing roles in cross-border auditing and advisory engagement, and leads international research collaborations with contributions to AI research, cognitive systems and particle physics.
Proissl focused on human-augmented training of domain-specific neural networks, and discussed use cases and recent advances in methods to address model transparency, adversarial robustness, algorithmic bias and fairness. He spoke on how most applications based on neural networks aim to extract modellable features, which are simulated or solved by constructing a mathematical model from the data to enable precise predictions.
The final speaker was Rich Radley, a customer engineer from Google Cloud. Radley joined Google in 2016 after working in management consulting where he advised clients on technology outsourcing.
Radley showed how Google is partnering with financial services organizations to apply deep learning techniques to problem domains such as forecasting, risk, and financial crimes, and how AI can improve customer experience. He stressed that model interpretation is important because it increases understanding and trust, identifies and mitigates bias, and boosts performance. He also spoke on the importance of building a new, Bayesian Case Model (BCM), which can identify relevant features to accelerate evaluation time and interpretability.
The ReWork Deep Learning in Finance Summit was held March 19 and 20 at the Etc.Venues St Paul’s in London.
Author: Shao Zi | Editor: Michael Sarazen