slides/tws2022/outline.md

In mbjoseph/neuralecology: Fit Neural Hierarchical Models of Ecological Populations

vISEC2020 talk

Main points:

1. Neural hierarchical models combine neural networks with models of ecological states/processes and observations
2. If you already know what a hierarchical model, a neural hierarchical model uses a neural network for function approximation (like Gaussian proceses, GAMs, etc.)
3. If you already know about neural networks, a neural hierarchical model modifies the loss function to account for state dynamics and the observation process
4. e.g., BBS analysis (mention scale afforded by stochastic optimization)

5. e.g., ConvHMM movement model (mention flexibility)

6. This is part of a broader synthesis of science-based deep learning

7. mention universal differential equations

8. physics-constrained neural networks

9. Let's think more broadly about how we use deep learning

10. e.g., how can we couple the output of computer vision models with ecological models?
11. what are the new failure modes that we invite with deep learning?

Hook:

Deep learning revolution

What has ecology done with the deep learning revolution - example applications

These applications are unsatisfying. They don't tell me what I want to know.

I don't just want to predict data. I want to learn about ecology! - population dynamics - occupancy states - historical change - demographic rates

These are all things we study with hierarchical models

mbjoseph/neuralecology documentation built on Nov. 6, 2022, 3:48 p.m.