Modeling Mood Polarity and Declaration Occurrence by Neural Temporal Point Processes.

Area of Science:

• Computational neuroscience
• Machine learning
• Time series analysis

Background:

• Real-world data often comprises heterogeneous time series from diverse sources.
• Traditional point process models struggle with data exhibiting varied distributions and sampling rates.
• Integrating sparse subjective declarations with dense sensor data presents a significant challenge.

Purpose of the Study:

• To introduce and validate a neural point process framework for analyzing complex health and behavioral data.
• To explore the impact of incorporating multiple, heterogeneous data sources into the model.
• To assess the predictive performance of neural point processes for event type and timing.

Main Methods:

• Development and empirical validation of novel neural network architectures for point processes.
• Integration of sparse event data (subjective declarations) and dense time series data (wearable sensors).
• Utilizing a novel, challenging dataset collected from a real-world, uncontrolled experimental setting.

Main Results:

• Neural point processes demonstrate strong performance in predicting the next event type.
• The models successfully predict the time until the next user interaction.
• Including diverse input sources enhances the predictive capabilities of the neural point process models.

Conclusions:

• Neural point processes offer a flexible and robust framework for modeling heterogeneous time series data.
• This approach is highly effective for health and behavioral analytics, leveraging multi-modal data.
• The findings highlight the potential of neural point processes for real-time event prediction in complex environments.