Deep Learning is picking momentum in Quantitative Finance, outside the obvious application to the prediction of asset prices (where to my knowledge it is not particularly effective) and spreading into the more serious application area of option pricing and risk management.
These two recent papers clearly demonstrate the benefits of DL as a pricing technology alternative to the classical FDM and Monte-Carlo in certain contexts:
These posts also explore the ideas of 'deep calibration' (application of DL to the calibration of financial models in a particularly interesting form) and 'deep analytics' (using DL to resolve the conundrum of revaluation in the context of regulatory simulations):
I should add that those two posts only constitute early explorations. Those are extremely interesting and high potential ideas, but serious implementation challenges must be overcome to put them in production.
Update: This paper by Blanka Horvath, Aitor Muguruza and Mehdi Tomas just hit SSRN and LinkedIn: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3322085.
This paper generalizes the approach of the two papers previously mentioned, among others. It offers a clear description and perspective of the application of DL to pricing and calibration, and offers a practical solution in the context of rough volatility models, where no other viable solution exists (to my knowledge). They also offer practical advise on structuring ANNs for these purposes, and share numerical results.