*** Matteo ***
(top-down, Probabilistic): 
Induction - How can people learn so much about the world form limited evidence (input).

What makes people smart?:
induction from partial or noisy information.
Top-down approach. Using Bayesian framework to tackle the induction:

Probabilistic models (ingridients):
Prior: degrees of believe over hypothesis before observing data
Likelihood: relation between the hypotheses and observable data
Inference: Learning approximates optimal statistical inference.

Advantages:
- Formal/Explicit
- Identification of "optimal" solutions
- Autonomy from implementational level:
	- Flexibility for exploring different representational formats (IMPORTANT)
	- Explanatory scope (common language for different sources of data)
	- Unifying power (common language for different cognitive domains)

Problems:
- What is known in advance (the prior)
- What are the ovjective and constraints (the loss function)
- Lack of real basis in actual mechanisms (as-if models of performance)
- How to relate function and structure
- Experamentally underconstrained

Is the Brain really Bayesian

It all boils down to: You start with having a computational problem, finding a way to solve it, but don't really care about the neural implementation. You do engineering then, not brain research.

(bottom-up, connectionist)
What makes people smart?
cognition as emergent consequence of interplay of a large number of dumb processes!
-> focus on "real" cognitive processes
-> Emergence from low-level dynamics

Ingredients of Emerginistic Models
Neuron-like processing units
Knowledge stored in the connections among units.
Cognitive activity depends on experience drive connection adjustment rules
-> Representation lacks explicit structures

Advantages:
-> should start from low level, as this is the way evolution goes
-> Focus on "real" mechanisms:
	* Integrated approach to cognition
	* Emphasis on temporal and spatial constraints of ...
Problems:
-> Oversimplification of the mechanisms
-> Why \ How exactly emergent behaviour of a complex system solves a certain problem?
-> Implicit theoretical commintments about hypothesis spaces and structure

Issues for Debate:
What is the goal of cognition?
What makes a certain approach progressive?
What type of predictions d othe two approaches give us?
Under what conditions can we say that a model describes "real" phenomena?
Should we persue unity or plurarism in the CogSciences?
What are sociological/technological factors influencing "framework-selection"?

-----
*** Matthew Chalk ***

(general comments + case studies)
Low level models make high-level assumptions.
Flexibility of high level probabilistic models makes them unconstained - low level provides the constrains.
=> they are somehow inseperable

### Best approach depends on scientific question ###

Case study: Binocular rivalry - probabilistic model (Griffiths et al. 2009) - bottom-up model (Matsuoka et al. 1984)