The following is the problem I wish to solve.
[[🎥estimator]] and [[generator]]
1. What: Solve non-identifiable problem with implicit prior knowledge
- e.g. hierarchical ODE described [here](https://discourse.mc-stan.org/t/the-monster-model-hierarchical-ode-models/22683/12?u=hyunji.moon) i.e. the Monster problem
2. How: Find representation for structural knowledge via algorithmic convergence.
Five tribes (symbolist, bayesian, connectionist, evolutionaries, analogizer) introduced in the book "[Master algorithm](https://en.wikipedia.org/wiki/The_Master_Algorithm)" will converge to the best form of knowledge represention such as the following.
- As Bayesian posess explainable structure of distributional update with Bayes rule (i.e. change of measure), it is optimal to be used as a basic sauce (e.g. Bayesian deep learning, Bayesian reinforcement learning).
- Connectionists, symbolists are being combined on top of Bayesian (as was predicted in the book). Analogizer (learning with kernel; Fourier?) and evolutionaries (algorithmic reasoning) are in line to be unified.
3. Why: To estimate parameter and predict states
![[Pasted image 20220619123843.png]]
![[Pasted image 20220619123916.png]]
![[Pasted image 20220619123932.png]]