Computational modeling of Human Cognition

What do we refer to when talking about Human cognition

how information is acquired, represented and processed in human brain; identify processes and mechanisms involved in learning and processing 

What do we refer to when talking about computational modeling

simulating a cognitive process via computational tools and techniques to explain the observed human behavior

what are Large Learning Models and how do they differ to humans

LLM are probabilistic machines and humans are not (how do we know that we don't perform in a probabilistic way); Humans are meaning-driven 

How do children learn new words and what is being generalized using these observations

• Away from distractions, they put two things in front of the baby and ask a question of 'can you show me the...' 

  The baby establishes the relationship between a new word and an new object  

Empirical findings of the children's behaviors and how their knowledge of word meaning develop over time are taken to build a theoretical account to explain how input of information turns into knowledge/output

how are cognitive processes explained by theories, experiments and models

The theory made aims to describe the cognitive process we are interested in and should be able to be used to predict similar behavior  

Stimulus being perceived can be controlled in order to observe the effect on behavior

a model can simulate the input, implement the cognitive process and generates the output behavior. But what do the models look like

example of a model - categorization

Everyone creates their own categories of objects, but can also agree on certain categories 

common modeling frameworks

symbolic - first generation of models influenced by early AI

• systems of knowledge representation - logic based

connectionist - inspired by the architecture of human brain  

probabilistic - following the success of statistical machine learning techniques

• connects symbolic and connectionist models

neural - a resurrection of old connectionist models

• parallel processing of data representing distribution of knowledge

Symbolic modeling framework

• Explicit formalization of the representation and processing of knowledge through a symbol processing system 

• Representation of knowledge – set of symbols and theor propositional meaning 

  • e.g. juicy & red fruit --> APPLE 

• Arbitrary, we need to represent object in certain ways 

  • Apple | O1 = APPLE | Color(O1) = RED, ANIMATE(O1) = FALSE, CATEGORY(O1) = FRUIT

symbolic apprach → learning by processing & updating knowledge using general rules under certain constraints

CFGs - Context Free Grammar: symbolic formalism for representing grammatical knowledge of language

uncertainty occurs with symbols that are inbetween categories or share similarities between the

Probabilistic modeling framework

• Apply probability theory on previous exposure to data → knowledge representation using weighted units that reflect bias or confidence based on previous observations; → learning mechanism where hypotheses are formed using algorithms for weighting and combining evidence that explain the data

often worksin combination with other techniques and formalisms

• A symbolic rule-based representation 

  • Each rule is augmented with a probability value indicating its applicability e.g. past-tense formation, probability of ‘ed’ added to a word

connectionist modeling framework

Between input and output there are layers of units that connect with each other to form the representation of knowledge. we learn through time as connection weights adjust and change

distributed inputs would have the different features and each feature would have an output

Neural modeling framework

• Resurrection of the connectionist models due to higher computational power and better training techniques 

• Adding more hidden layers increases model's power in learning abstract complex structures

what does Modeling Human cognition processes do for us

allows us to study processes through simulation

evaluate plausibility of existing theories

explain observed human behaviour during a specific process

predict behavioral patterns that have not been experimentally investigated