Last week’s RE•WORK AI in Finance Summit featured 50 speakers and drew 250 technologists to the Westin New York at Times Square to explore the intersection between AI and Fintech. The two-day event hosted various sessions where leading innovators from academia and the financial sector discussed their discoveries, advances and insights in machine learning tools and techniques for finance.
Topics covered included fraud detection, AI security in FinTech, available advanced AI technologies, AI implementation in financial services, and financial forecasting and regulations.
The room was packed for Manuela Veloso’s talk. Head of the Machine Learning Department at Carnegie Mellon University and recently appointed Head of AI Research at J.P. Morgan, Veloso shared her experiences with AI and autonomous agents, and outlined how J.P. Morgan is developing AI in the financial sector. “J.P. Morgan is a fantastic environment for a researcher and a developer and it is fascinating for me to see how AI can make a difference there. The main problem though, as always, is data. How do we deal with data in the most effective way and harness it?”
Veloso said data and cryptography are a central focus of J.P. Morgan’s AI research. She noted however that data of relevance to financial services is often inconsistently defined, sparse and noisy. Regarding clients’ concerns about data protection, Veloso said data regulation also needs to be weighed in.
Veloso reminded the audience that powerful technologies such as reinforcement learning need to be handled carefully: “what makes it powerful also makes it dangerous when applying it to the real world.” With regard to security, she said researchers have encountered uncertainty problems both in AI and financial services, and that these uncertainties enabled researchers to learn from patterns and provide solutions. She said researchers should carefully consider uncertainty problems when predicting market direction and value credit quality.
Veloso noted that her CMU office has a cute and popular robot receptionist, and suggested that AI can bring automated human-machine interaction to financial services via call centers and trade processing. Veloso also reminded the audience that dynamic markets are a complex environment with both collaborative and adversarial players.
Synced noted that a number of keywords repeatedly popped up in various sessions:
Enhancing customer services
- Credit Suisse Head of Process Automation and Robotics Uday Singh said his firm has broadly adopted AI and ML to automate its IT infrastructure. He said the organizational setup — including an AI implementation center, data modeling, and AI tools management — had improved the customer experience.
- Ambika Sukla from Morgan Stanley suggested that AI offers companies many opportunities to boost customer services, but warned “we need to fully understand how models work and it is vital that we continue involve humans in the loop.”
- In the AI in Enterprise session, AIG’s Yuanyuan Liu discussed his firm’s encounters with AI. He said AIG has realized that AI can speed up the underwriting process, and his Nine Killer Applications of Digital Technology in General Insurance presentation focused instead on the importance of enhancing customer experience.
DATA DATA DATA
- NYU Research Professor of Financial Machine Learning Igor Halperin spoke on the problem of insufficient data. “Challenges that I have faced in my teaching have made me focus on what we do modeling wise when we don’t have enough data” His talk on Reinforcement Learning for Portfolio Optimization and Market Modeling posed the question: How can academia deliver advanced research if the required data is not accessible?
- While some wondered where to obtain data, others were more concerned with how to process data. Yvonne Li of Wall Street regulator FINRA (Financial Industry Regulatory Authority) said FINRA is using NLP to mine text fields in the many regulatory forms they receive each day.
- Laura McKiernan Boylan from Haven Life said that a combination of tree-based and deterministic datasets have proven the best solution during her company’s algorithmic underwriting processes for life insurance.
- Andrew Clark of Capital One explained that because data is distributed in many different places, it is important to establish strong collaboration between different departments. He introduced a Capital One’s “unsupervised learning data exploration environment” as a way to tackle the problem.
- World Bank Data Scientist Anton Prokopyev discussed AI’s application in international trade, specifically the use of machine learning for identifying non-tariff measures. He recommended one of the takeaways from his speech should be “Engineers need to train data properly from ground zero.”
- MIT-IBM Watson AI Lab researcher Mark Weber introduced the FastGCN research project, which aims to develop new methods for deep learning on huge, dynamic graph datasets.
- Turkish startup FinBrain Technologies presented their deep learning “One Algorithm to Predict Them All” model, which predicts how financial assets will move in the future.
- Israel’s I Know First is another startup taking full advantage of algorithms. Company CEO Yaron Golgher told the audience “the historical correlation between past algorithm predictions and accrual market movement for each asset is the key to identify and focus on the most predictable markets and securities, enhancing the overall performance.”
RE•WORK AI in Finance New York attracted a significant number of young people eager to stay abreast of AI developments in Fintech. Worcester Polytechnic Institute student Mukund Khandelwal told Synced: “As a data science graduate student, it was amazing to learn about advancements in AI and its application in Finance.”
Journalist: Fangyu Cai | Editor: Michael Sarazen