Artificial Intelligence In Finance Keynote at RFinance Conference

Artificial Intelligence In Finance Keynote at R/Finance Conference, Chicago (based on my invited presentation at MIT Tech. Review Em. Tech Digital, San Francisco, March 2018) Li Deng, Chief AI Officer May 2018

Why I Joined Citadel The Opportunity for AI in Investment Management

Role of Investment Firms Social Benefits of Investment Management + Data + Talent = Technology Investment Firm Who Benefits: Universities Governments Hospitals Museums Retirees Research

Outline of the Main Topics 1 2 3 Will AI transform the financial markets? § Speech § Computer vision § NLP § Finance Three technical challenges unique to financial investment industry Other constraints in applying AI to financial investment management

Will AI Transform the Financial Markets? Learning From Other Industries What can we learn from successful AI applications in other industries: § AI disrupting speech industry (2009 -present) – (Small) similarities to finance industry – (Large) differences from finance industry § AI disrupting computer vision industry (2012 -present) § AI disrupting NLP (2014 -present)

Disrupting the Speech Industry Launch of Deep Learning in Speech was at NIPS in 2009

Disrupting the Speech Industry Deep Learning practically solved the speech recognition problem by 2012 By John Markoff Tianjin, China, October 25, 2012 Voice recognition and translation program translated speech in English given by Richard Rashid, Microsoft’s top scientist, into Mandarin Chinese.

Disrupting the Speech Industry Deep Learning for Speech Named Best Breakthroug h Technology by MIT Tech Review in 2013

Disrupting the Speech/Vision/NLP Industry by Deep Learning The success has no controversy

Disrupting the Speech/NLP Industry Separate Speech Recognition Models Unified by End 2 End Deep Learning Training Data Applying Acoustic Models Constraints Lexical Models Language Models Recognized Words Speech Signal Representation Search

Three Challenges Unique to Investment Management 1 2 3 Very low signal-tonoise ratio Strong nonstationarity with adversarial nature Heterogeneity of big (alternative) data

Three Challenges Unique to Investment Management 1. Very low signal-to-noise ratio AI problems outside finance generally have lower noise levels, for example: The technology used to combat noise shares characteristics with the technology used to handle small data in training large AI systems, including: § Speech § Ability to exploit structure in data § Machine translation § Reliance on prior knowledge § Language understanding § Use of data simulation/augmentation § Image/video classification & detection § Smart model regularization § Medical diagnosis § Etc.

Three Challenges Unique to Investment Management 2. Strong non-stationarity with adversarial nature

Three Challenges Unique to Investment Management 3. Heterogeneity of big (alternative) data

Structural Constraints Applying to AI in Investment Management What still needs to be done to ensure success? Data Access Respect for Privacy Scarcity of Talent Tailored Algorithms

Q&A