Deep Learning for Financial Data Analysis

Motivation

The recent developments in Artificial Neural Networks and Deep Learning methods delivered robust and scalable methods for the analysis of a wide class of heterogeneous and complex data-driven problems. Whilst Deep Learning methods have been early-applied and are widespread in Computer Vision, Natural Language Processing, Robotics, and Signal Processing for in-sample analysis and prediction, they have been only very recently adopted for data-driven financial analysis and forecasting.

The electronic high-frequency trading characterizing modern financial markets generates massive amounts of data, where the complex interaction between algorithms and the trading strategies they implement lead to a unique class of highly complex, unstructured and multidimensional datasets embedding high potential for Machine Learning applications.

Indeed, the actual complexity of the typical modern high-frequency markets goes beyond the modeling capabilities of classic econometric methods, which are often found inappropriate. On the other hand, recent applied multidisciplinary research across Machine Learning and finance has shown Deep Learning methods to be capable of extracting information and meaningful features from market data, providing a feasible tool for a wide number of regression and classification problems. While Deep Learning methodologies provide a strong candidate for a paradigm shift in financial data analysis, the financial academic community has been slow to adopt them and propose further developments addressing challenges created by the unique properties of financial data

Scope

The relevant literature on Deep Learning and, in general, Machine Learning applications in financial data analysis and prediction is growing at a very high pace, showing a high research potential for this emerging topic, and a widespread interest not limited to the Machine Learning community. Whereas Deep Learning methods have been extensively applied in different domains, in financial applications the research is still at its early stage requiring further research and investigation. The focus topic of the Special Session is of high impact, timely and trending.

The proposed Special Session is aligned to the Neural Networks rationale of IJCNN aiming at bringing researchers in Machine Learning to exchange recent advances and uncover future trends in the rich domain of Deep Learning and Artificial Neural Networks techniques and methodologies for financial applications.

Topics of Interest

Topics of interest include (and are not limited to):

• Optimized methods for financial regression and classification problems

• Nonlinear financial time-series forecasting

• Efficient inference

• Big-Data methods for ultra-high frequency financial time-series market data

• Automated feature extraction for Representation Learning and classification

• Supervised and Unsupervised Deep Learning for financial forecasting

• Comparisons of Deep Learning methods with their corresponding econometrics approaches

• Reinforcement Learning for financial data analysis

• Graph Neural Networks for investor networks analysis

• Deep multi-agent systems for finance

• Interpretable machine learning for finance market research

• Bayesian machine learning for finance