ATD: Multimodal Transformer-based Model for Time-series Prediction and Spatiotemporal Analysis

Consistency and reliability are necessary for algorithms used for data analysis and high-consequence decision-making. The project will contribute to this goal by developing algorithms for time-series analysis and anomaly detection that expand capabilities of understanding subtle features from multiple distinct data sources. In addition, this work will advance the spatial reasoning abilities of artificial intelligence algorithms which could be applied to other engineering or scientific problems. The project will train PhD students through involvement in the research.

The aim is to develop mathematical algorithms for forecasting time-series, predicting spatial dynamics, and detecting anomalies using a multimodal transformer-based model. The project will construct methods for analyzing systems that switch dynamics or change behaviors. This can be applied to downstream tasks such as data analysis and anomaly detection.

The research will address how to utilize contextual information with time series for more reliable predictions and how to consistently incorporate multiple pieces of information and observational modalities into prediction and anomaly detection.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.