SyntAIr - IMPROVED ATM AUTOMATION AND SIMULATION THROUGH AI-BASED UNIVERSAL MODELS FOR SYNTHETIC DATA GENERATION

SyntAIr - IMPROVED ATM AUTOMATION AND SIMULATION THROUGH AI-BASED UNIVERSAL MODELS FOR SYNTHETIC DATA GENERATION

The main objective of SynthAIR is to explore and define AI-based methods for synthetic data generation in the domain of ATM system due to the limitation of AI-based tools development by the lack of enough data available (e.g., safety-related data) and the problem of generalization of those AI-based models. We want to explore data-driven methods for synthetic data generation, since they require 1) less user knowledge expertise (no need to derive the explicit model of the distribution), 2) better generalization capabilities. More in detail, inspired by recent advancement in Computer vision and Language Technology, we propose the concept of Universal Time Series Generator (UTG). A UTG, is a model trained on several different time series, and able to generate a synthetic dataset representing a new dataset, simply conditioned by a compressed representation of it. In aviation domain, this generator can be trained on a certain set of data related, for example to few airports, and be used to generate synthetic data from a new airport. The same principle can be applied to define a universal time series forecaster (UTF) able to do prediction to a new environment (I.e., data from a new airport) without any new training.