One Transformer for All Time Series: Representing and Training with Time-Dependent Heterogeneous Tabular Data
This addresses the problem of handling mixed categorical and numerical data in time-series for researchers and practitioners in fields like finance, though it appears incremental as it adapts existing Transformer methods to a specific data type.
The paper tackles the challenge of applying deep learning to heterogeneous time-dependent tabular data, such as financial transactions, by proposing a Transformer architecture that represents numerical features with frequency functions and trains uniformly with a single loss function.
There is a recent growing interest in applying Deep Learning techniques to tabular data, in order to replicate the success of other Artificial Intelligence areas in this structured domain. Specifically interesting is the case in which tabular data have a time dependence, such as, for instance financial transactions. However, the heterogeneity of the tabular values, in which categorical elements are mixed with numerical items, makes this adaptation difficult. In this paper we propose a Transformer architecture to represent heterogeneous time-dependent tabular data, in which numerical features are represented using a set of frequency functions and the whole network is uniformly trained with a unique loss function.