Outline
Problems with neural networks
Support Vector Machines
Controlling complexity in statistical models
Questions about neural networks
Why do they have such a bad rap?
To what extent are neural networks brainlike?
They take a long time to train.
Anderson (1991):
Rational Analysis
Goal of categorization: feature prediction for
biological species (or similar natural kinds).
Structure of biological species: feature clusters
Some animals fly, have wings, and live in trees.
Some animals swim, have f
Outline
Controlling complexity in Bayesian neural
networks
Controlling complexity in infinite mixture
models
Discussion
Computational strengths and weaknesses
Cognitive relevance
How to choose control parameters?
Bayesian Occams razor
Demo
Smaller
Outline
Non-parametric models for categorization:
exemplars, neural networks
Controlling complexity in statistical models
Bayesian classification
Image removed due to copyright considerations.
The task: Observe x generated from c1 or c2, compute:
p ( x
Approaches to structure learning
Constraint-based learning (Pearl, Glymour, Gopnik):
Assume structure is unknown, no knowledge of
parameterization or parameters
Bayesian learning (Heckerman, Friedman/Koller):
Assume structure is unknown, arbitrary par
Outline
Theory-based Bayesian framework for
property induction
Causal structure induction
Constraint-based (bottom-up) learning
Theory-based Bayesian learning
The Bayesian approach
?
Species 1
Species 2
Species 3
Species 4
Species 5
Species 6
Specie
Species 1
Species 2
Species 3
Species 4
Species 5
Species 6
Species 7
Species 8
Species 9
Species 10
?
?
?
?
?
?
?
?
Features
New property
'
(Osherson, D. N., et. al. "Category-based Induction." Psychological Review 197 (1990): 185-200.+
Outline
Theory-based Bayesian framework for
property induction
Causal structure induction
Constraint-based (bottom-up) learning
Theory-based Bayesian learning
The origins of causal knowledge
Question: how do people reliably come to
true beliefs ab
Outline
Theory-based Bayesian causal induction
On intuitive theories: their structure,
function, and origins
Domain theory generates hypothesis
space of causal models
Ontology of types and predicates.
What is there?
Constraints on causal relations be