Inductive Inference Flashcards
What is inductive inference?
Making conclusions that extend beyond the data, conclusions have some degrees of support by the data and conclusions can be false
What are some examples of infants making meaningful extrapolations from very little data?
Learning the meaning of words, learning the causes of behaviour and inferring unobserved variables
When learning the meaning of words, what are we able to do?
Learn from only a few examples, use the word appropriately in new situations and grasp the boundaries of categories from very little data
What is nativism and the disadvantaged of it?
People’s abstract knowledge is not learned and concepts are innate. Dis ->difficulty to explain high flexibility in learning
What is connectionism?
Human knowledge is not actually as abstract as it seems and knowledge acquisition is strengthening/weaking associations between concepts
Dis-> difficult to explain learning from little data
What is the hierarchical bayesian accounts of cognitive development?
Learenrs can draw accurate inferences from just a few data points if their inferences are constrained by more abstract theories and they can change their theories with sufficient evidence
How is the hierarchical bayesian account used for causal learning example?
Theory: “buttons make machine go” -> helps to make sense of sparse -> Data: this will also be the case in this specific situation
What was the children’s behaviour after the experimenter did Puts button A on machine. Nothing happens.
* Puts button B on machine. Nothing happens.
* Puts buttons A and B on machine. Nothing happens
* Says “Machine, please go!”. Machine starts making noise.
→ “Can you make the machine be quiet?”
75% of children gave a verbal command to machine
What are the theories of the hierarchical bayesian account like?
They are not innate, provide assumptions that help interpret data. Data can change theories and theories exist at different levels of abstraction
Why are these accounts called hierarchical bayesian?
Hierarchy of theories created by Thomas Bayes - invented the bayes rule which is a calculation of conditional probabilities
Bayes rule
𝑃 𝑇ℎ𝑒𝑜𝑟𝑦 𝐷𝑎𝑡𝑎 ∝ 𝑃 𝐷𝑎𝑡𝑎 𝑇ℎ𝑒𝑜𝑟𝑦 ⋅ 𝑃(𝑇ℎ𝑒𝑜𝑟𝑦)
How much people
believe in a theory after
seeing data
How much people
believe in a theory
before seeing data
How well the theory
explains the data that
they see
posterior probability likelihood prior probability
What are the advantaged of formal theories?
More precise than verbal theories - assumptions are clear and no room for vague concepts and definitions. Can be used to define optimal behaviour and help identify unintuitive predictions of theories
Where can inductive inference lead us?
Helps us to develop formal models that make precise predictions - use for improving our understanding of children’s development, training ai to be more human like and developing more effective teaching strategies
