Skip to article frontmatterSkip to article content

Computational Model

Computational and Mathematical Modeling

CNRS
MEOM

Overview

Sunset at the beach

Figure 1:A schematic for Modeling and Simulating Physical Systems. Source: Book (Modeling and Simulation in Python)

We have a system which represents the state, zz. We never observe the state. Never. We only have measurements, yy, which are a sparse representation of our system. i.e., they don’t cover the whole globe at every given point in space and time.

So how do we learn? One way to do so is through mathematical modeling:

Note: that second step is what I would separate learning from estimation/predictions. I can easily construct a data-driven model which can emulate something given data. But I don’t consider it really learning unless I have asked a question first and I feel happy with my answer.

Example:

θ=arg maxp(θD)\theta^* = \argmax \hspace{2mm} p(\theta|\mathcal{D})

What can go wrong

Data, D\mathcal{D}

Domain, Ω\Omega

Transformation, T,ΘT,\mathcal{\Theta}

Learning, p(MD)p(\mathcal{M}|\mathcal{D}).