Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Synthesising Activity Participations and Scheduling with Deep Generative Machine Learning

Fred Shone, Tim Hillel

TL;DR

The paper tackles synthesising realistic human activity schedules by learning the joint distribution of participations and timings with Variational Auto-Encoders. It introduces a continuous duration encoding and systematically compares discrete vs continuous encodings across FF, CNN, and RNN architectures, using a comprehensive evaluation framework that includes density estimation, validity, and creativity metrics. The Continuous RNN model emerges as the strongest performer, delivering superior distributional fidelity, disaggregate realism, and novel schedule generation, while maintaining efficient generation speeds and avoiding severe overfitting. The work demonstrates the practicality of pure generative synthesis for non-conditional schedule generation, with clear implications for fast upsampling, anonymisation, and scenario analysis in transport, energy and epidemiology contexts, and outlines directions for incorporating conditionality in future work.

Abstract

Using a deep generative machine learning approach, we synthesise human activity participations and scheduling; i.e. the choices of what activities to participate in and when. Activity schedules are a core component of many applied transport, energy, and epidemiology models. Our data-driven approach directly learns the distributions resulting from human preferences and scheduling logic without the need for complex interacting combinations of sub-models and custom rules. This makes our approach significantly faster and simpler to operate than existing approaches to synthesise or anonymise schedule data. We additionally contribute a novel schedule representation and a comprehensive evaluation framework. We evaluate a range of schedule encoding and deep model architecture combinations. The evaluation shows our approach can rapidly generate large, diverse, novel, and realistic synthetic samples of activity schedules.

Synthesising Activity Participations and Scheduling with Deep Generative Machine Learning

TL;DR

The paper tackles synthesising realistic human activity schedules by learning the joint distribution of participations and timings with Variational Auto-Encoders. It introduces a continuous duration encoding and systematically compares discrete vs continuous encodings across FF, CNN, and RNN architectures, using a comprehensive evaluation framework that includes density estimation, validity, and creativity metrics. The Continuous RNN model emerges as the strongest performer, delivering superior distributional fidelity, disaggregate realism, and novel schedule generation, while maintaining efficient generation speeds and avoiding severe overfitting. The work demonstrates the practicality of pure generative synthesis for non-conditional schedule generation, with clear implications for fast upsampling, anonymisation, and scenario analysis in transport, energy and epidemiology contexts, and outlines directions for incorporating conditionality in future work.

Abstract

Using a deep generative machine learning approach, we synthesise human activity participations and scheduling; i.e. the choices of what activities to participate in and when. Activity schedules are a core component of many applied transport, energy, and epidemiology models. Our data-driven approach directly learns the distributions resulting from human preferences and scheduling logic without the need for complex interacting combinations of sub-models and custom rules. This makes our approach significantly faster and simpler to operate than existing approaches to synthesise or anonymise schedule data. We additionally contribute a novel schedule representation and a comprehensive evaluation framework. We evaluate a range of schedule encoding and deep model architecture combinations. The evaluation shows our approach can rapidly generate large, diverse, novel, and realistic synthetic samples of activity schedules.
Paper Structure (56 sections, 10 equations, 14 figures, 23 tables)

This paper contains 56 sections, 10 equations, 14 figures, 23 tables.

Table of Contents

  1. Introduction
  2. Activity Scheduling
  3. Supervised Discriminative Machine Learning Approaches to Activity Scheduling
  4. Activity Schedule Evaluation
  5. Deep Generative Modelling
  6. Evaluating Deep Generative Models
  7. Generative ML in Transport Modelling
  8. Summary of Discriminative and Generative ML Approaches for Activity Scheduling
  9. Generative ML for Sequence Data
  10. Research Contributions
  11. Methodology
  12. Informal Problem Definition
  13. Activity Schedule Definition
  14. Problem Definition
  15. Generative Modelling with Variational Auto-encoders
  16. ...and 41 more sections

Figures (14)

  • Figure 1: UK National Travel Survey (2023) Activity Frequencies: Aggregate activity participation by 10-minute time bin
  • Figure 2: VAE Model Template
  • Figure 3: Example Schedules from the UK National Travel Survey (2023)
  • Figure 4: Discrete Activity Schedule Encoding
  • Figure 5: Sequence Activity Schedule Encoding
  • ...and 9 more figures