The top-down and bottom-up approaches are two different epistemological strategies for modeling cognition across different Levels of cognitive theories:
| Name | Starting point | Relevant data | Examples |
|---|---|---|---|
| Top-down | Theoretical frameworks and assumptions about the structure or goals of a cognitive system | Task structure and environment statistics; behavioral responses | Normative: Bayesian models, (resource-)rational analysis Descriptive: theory-theory of concepts, symbolic rule-based models (e.g., ACT-R) |
| Bottom-up | Observed behaviors and neural processes | Physiological | Connectionist models, embodied cognition |
- Where does something like reinforcement learning (normative, yet data-driven) fall?
- How do metaphysics—particularly the multiple realizability of brain functions—change the epistemology (see: @2022caoMultiple)?