Task 5 Flashcards
What is meant by cognitive science ?
- understanding the processes that the brain uses to accomplish complex tasks
What are cognitive models ?
- want to explain cognitive processes and how they interact,
- Trying to understand and conceptualize cognitive processes like categorization,
What are the 5 steps of cognitive modeling?
- Translate theoretical framework to a mathematical or computer language
- Add additional assumption (ad hoc assumptions)
- Estimate parameters
- Compare predictions with competing models
- Start over again
What differs cognitive models from other models ?
- they derived from basic principles of cognition
Name two types of cognitive models:
- prototype and example model of categorization
What is meant by the prototype model of categorization ?
- Putting an object into a category based on prototypes (average object of each category)
What is meant by the exampler model of categorization ?
- Putting an object into a category based on finding the matching part from ALL past remembered examples
What are the advantages of cognitive models over conceptual model?
- since they use mathematical and computer language, they guranted produce only logically valid predictions
- capable of precise prediction
- Generalizeability = can make predictions that go beyond the original data
How can u translate a conceptual model into a cognitive model ?
- via changing the verbal statement into mathematical or computational language
What is an example of a rule based system ? (comparison to deep learning)
- SOAR modle
What is the main component of a rule based model ?
- The production rule ->
- If part -> Condition
- Then part -> Action
What is meant by condition ?
- specifies what must be true for a rule to apply
- Can be multiple factors
What is meant by action ?
- specifies a set of things to do if the production applies
- can be multiple factors
What is the limitation of multiple “actions” and “conditions” ?
- The multiple action can not be contradictory
What kind of knowledge can be presented in rules ?
- General information about the world (non changable logical rules / bilogical /physics)
- example would be language
What is the difference between logic and rule based system ?
- rules do not have to be interpreted as universall true (rather as a default)
- can better represent strategic information about what to do
- logic based system focus on deduction regarding thinking
- rule based system focus on SEARCH regarding thinking
- rule based system have less representational power but more computational and psychological power in comparison to the logic system
What is meant by representational power ?
- How much knowledge about the world can be represented?
- Or how to do things in the world
What is meant by computational power ?
- How powerful and efficient are rule-based system
- How good is there Problem solving, planning, explanations, learning, language = useful
How does rule based problem solving work ?
- They first search trough a problem space to find a path from current to goal state
- In doing so people rely on heuristics bc it is impossible to search the entire space (to complex)
What is meant by problem space ?
- is the space of possibilities that you must navigate trough
How do people search in the problem space ?
- Serial processing (one rule at a time)
- parallel processing (multiple rules at a time)
Define planing:
- Taking the best options out of the problem space and put them together
What kind of planing strategies do rule based model use ?
- Backward and forward reasoning and bidirectional search
What is meant by backward reasoning ?
- uses the logic of “which requires” -> starting from the goal and then going back
What is meant by forward reasoning ?
- Uses modus ponens from current situation to -> goal situation
What is meant by bidirectional search ?
- combines forward and backward reasoning
How do u learn according to a rule model ?
- via applying rules over and over again
How do we build rules ?
- Rules can be innate (from birth on)
- Rules can be learned via inductive generalization
- rules can be formed from other rules (via chunking and spezialization)
What is a rule based system such as the SOAR capable of ?
- can model high cognitive reasoning task
- can model human learning processes
- can copy human like behaviour in games
Why is rule based system currently so well known ? In comparion to the formal logic system?
- because it has the best psychological application -> Can account for many aspects of human behviour
- And also it has a slight neurological plausability because there is a small analogy between rules and neurons connected by synapsed
What is meant by the ACT-(R) system ?
- The model states that complex cognition arises from an interaction between declarative and procedural memory.
- Adaptive Control of Thought-Rational
What does the R stands for ?
Rational analysis theory
What is meant by rational analysis theory?
- states that each component of the cognitive system is optimized with respect to demands from the environment
What types of knowledge/memory does the ACT-R contain of ?
Declarative and procedural memory
What is meant by declarative memory ?
- Thinks we are aware of and can report on “we know “ (conscious)
- facts that are represented in chunks
What is meant by procedural memory ?
- contains procedures (“how to do stuff”)
- instructions are represented in units called production rules
How are declarative and procedural memory rleated ?
- They are independent but interrelated via connections
What type of connections relate declarative and procedural memory ?
- The production rules (retrieval request)
- production compilation.
What is meant by the production rule ?
- production rules fire in procdeural memory and may require elements from declarative memory
- A production rule is a: “If ->then” relationship
- THIS PROCESS IS CALLED RETRIEVAL REQUEST
What is meant by the production complementation ?
- It is the creation of new production rules
- they get created in procedural memory from ‘chunks’ in declarative memory
What are the key components of the ACT- R model ?
- The goal stack
- the current goal
- procedural memory
- declarative memory
- production complementation
- retrieval request
- ACT-R takes its environment into account
What is mant by the goal stack ?
- Contains all goals which are currently not put on attention but are still needed to be asnwered (subgoal)
- components:goals can be either “pushed” (Added) or “popped” (taken away) from current goal
- Goals can only be pushed or popped in reverse order (last in = first out)
What is meant by the current goal ?
- represents what ACT-R is currently trying to do
atttention is on
Explain declarative memory more in depth
- it is a collection of chuncks
- 2 to 4 elements/slots per chunck
What is meant by a chuck ?
- A chuck consist of SLOTS a
- First slot is called ISA which determines the type of the chunck (operator)
- All chunks are not isolated but interconnected and spread activation via connections
- Summary = each chunk has a level of activation and this activation tends to leak out and add to the activation of all the other chunks it is connected to.
What happens when a chunck has lots of activtaion?
- it will be easily found and retrieved from memory
- resemblance short-term memory
What happens when a chunck has not that much activtaion?
- it will be hard to find and may never be retrieved
- resemblance = long-term memory
What happens when a chunck has no activation ?
- it’ll be forgotten (at least temporarily until activation levels increase)
What is so special about the activation lvl of a chunk ?
- the activation lvl is not constant
- the more it is used the more it increases, when it isn’t used, it slowly decreases
- This allows to model a priming effect
Explain procedural memory more in depth:
- It stores procedures in form of production rules (multiple)
- which have a condition and an action
How does the procedural memory know which production rule is to choose?
- whether its condition (the IF part) matches the current goal
- whether the required chunks are in declarative memory.
What do chucks and prodcution rules have in coman ?
- They both have actvation
What happens when a production rule has low activation ?
- rule might not be used even though the conditions match the current state
What happens when a production rule has lots of activation ?
- the rule is more likely to be retrieved and used again in the future
Name all four stages of the production compilation:
- Understanding instructions and examples (instructions represent = new knowledge)
- Production compilation: we try to solve problems by applying these instructions
- Practice: through practice we solve problems with increasing speed and accuracy)
GOAL: First u have chunks in delcarative emory -> these chucks are turned into rules in procedural memory via practice (automatically response)-> rules create new chunks in declarative memory
Explain in detail the second step of producing a new production rule ?
- the declarative representation of instructions becomes transformed into a production rule
- > Specific values are replaced by varaibles -> The rule becomes more general
Explain in detail the third step of producing a new production rule:
- Throug out practice the activation of the production rule and of the chuck increases
What is a crtitic on the ACT- R model ?
- it lacks psychological plausability bc the system does perfectly recall every subgoal on the goalstack and humans do not
What example could we use to see how the ACT R works ?
- word list memory task
- Presented with a list of items (one after another) and asked to recall them after some delay
How can we compare a human with the ACTR model based on the list memory task ?
- recall latency (i.e. time taken to recall an item)
- recall accuracy
What were humans score on the list memory task ?
- Primacy and recency effects (first and last elemnt showed highest accuracy )
- Recall is slower for elements 1, 4 and 7 (due to the way items are chunked in memory)
The ACTR is a hybrid model because:
- it is connectionist: because in declarative memory graceful degradation
- And rule based: because the pocedural memory allways needs rules
How well did the model act in comparison to the human work regarding list memory ?
- very similar to human
- primacy and rencency effect was also present
- Lattency was also similiar 1,4,7 It takes longer to recall something if u first have to switch the group
- Limitation: Certain Parameters need to be set for differing types of recall
What is meant by partial matching ?
- a chuck can bes selected that only matches partially the item
- happends: when chunck is highly active and a complete match is absent
what is meant by default ?
it is a rough generalisations that admit exceptions
What is meant by a production rule ?
- rule based system
- IF -> then
What if multiple conditions matches the current goal ? (how do we deide then which production rule we should use)?
- Based on the activation lvl of the production rule !
- As allways high activation lvls are prefered
Does the ACT (R) has a specific set of production rules?
- The amount of production rule is not fixed
- The ACT (R) is able to learn new production rules
Which approach is integreated in the ACT- R system ?
- Goal directed approach
- they take a current goal and a current state, and the system acts either to achieve the goal or add a new goal that needs to be completed first.
