Approaching Scientific Question & How to Model

Marr’s 3 levels of analysis

Brain: hierarchy of complexities

Computational level - 1
- What is the objective of the system?
- How close is it to optimal?
Algorithmic level - 2
- What are the data structures?
- What are the approximations?
- What is the runtime?
Implementation level - 3
- What is the hardware?
  - Neurons? Synapses? Molecules?

Useful: good at solving real-world problems?
Normative: provide the optimal solutions to problems that exist in the real world?
Clinically relevant: produce insights relevant for developing or evaluating clinical interventions?
Inspire experiments: change the way we think about a problem, raising interesting new hypotheses & experiments?
Microscopic realism: describe the microscopic properties of the brain?
Macroscopic realism: describe properties of brain areas and networks?
Behavioral realism: can faithfully describe and explain behavioral phenomena?
Representational: use representations of info that are similar to representations in the brain?
Compact: can be succinctly expressed in math or code, making them easy for humans to understand?
Analytically tractable: understandable through equations instead of numerical simulations, therefore generalizable?
Interpretable?
Beauty: symmetrical, balanced, or resonate well with the way we think?

What exact aspect of data needs modeling?
- Answer this question clearly and precisely! Otherwise you will get lost (almost guaranteed)
- Write everything down!
- Also identify aspects of data that you do not want to address (yet)
Define an evaluation method!
- How will you know your modeling is good?
- E.g. comparison to specific data (quantitative method of comparison?)
For computational models: think of an experiment that could test your model
- You essentially want your model to interface with this experiment, i.e. you want to simulate this experiment