Vertically-Advised Federated Learning for Multi-Strategic Stock Predictions through Stochastic Attention-based LSTM
Danilo Menegatti, Emanuele Ciccarelli, Michele Viscione, Alessandro Giuseppi · 2023
In recent years, stock price forecasting has become a challenging task commonly used to evaluate the performance of various machine learning solutions. This work explores a Federated Learning (FL) framework within a competitive collaboration scenario with the aim of training a centralised model advised by non-recoverable decentralised strategies so that no exchange of private data is required. The proposed Vertically-Advised Federated Learning (VAFL) framework combines elements from both horizontal and vertical FL, as each client trains two independent models. Furthermore, a novel forecasting architecture, based on a stochastic variant of an Attention-based Long Short Term Memory (LSTM) network, is proposed and validated on a simulated scenario based on real data from the stock market.