Chapter 14 Probabilistic Model Formulation (with Time Warping and Genetic Stickiness)
Conceptual Description
- Gaussian Process (Topic Evolution):
- The Gaussian process models how topic proportions naturally change over time, assuming a shared timeline for all individuals.
- Individual Time Warping:
- The warping function W(t, ρ_d) stretches or compresses the timeline for each individual based on their genetic and/or environmental factors (modeled through the warping parameter ρ_d).
- Topic Distribution Update with Stickiness:
- At each time point, an individual’s exposure to topic distributions depends on their warped time index.
- New diagnoses update the topic distribution (θ{d,t}_) based on the relevant disease probabilities at that warped time and the genetic stickiness factor (γ_d).
Illustrative Example
- Consider a topic representing cardiovascular diseases. Individuals with high warping factors will reach disease probabilities associated with this topic earlier compared to those with low warping factors.
- An individual with high genetic stickiness will be less likely to shift away from this topic distribution even in the presence of new diagnoses unrelated to cardiovascular disease.
Notes
- Expressing the interaction of warping and disease likelihoods in a closed-form mathematical expression within markdown can be difficult.
- Computational implementation might involve procedural steps to determine the appropriate topic distribution for an individual at a given time, based on their warped timeline.
- Consider incorporating a pseudocode-like representation to illustrate the procedural nature of the time warping and its influence.
- Specify your choices for the mean function \(\mu(t)\) and any priors on Gaussian Process parameters.
- Clarify how the time warping function W(t, ρ_d) influences disease probabilities π{v,k,t}_
- Consider alternative inference techniques (e.g., variational inference) due to potential computational complexity.