Problem 3 Flashcards
According to cognitive scientists, how does the mind work ?
People have mental procedures that operate on mental representations to produce thought and action
BUT: different mental pres. foster different mental procedures
ex.: calculations with 1,2,3 are different than with I,II,III
Which methods did scientists use to search for the existence of the human mind ?
- Experimentation with human participants
- Forming + testing computational models intended to be analogous to mental operations
- -> to create theoretical framework
Central hypothesis of cognitive science
Thinking can be best understood in terms of representational structures in the mind + computational procedures that operate on those structures
–> mind = computer program
Computational-representational understanding of mind
CRUM
Refers to the approach to understanding the mind based on the central hypothesis
How is a computer an analogy to thinking or using the mind according to CRUM ?
Running a program results from applying algorithms to data structures
–> just like thinking results from applying computational procedures to mental procedures
e.g.: mental representations = data structures; computational structures = algorithms
Cognitive theory
Consists of a set of representational structures + a set of processes that operate on these structures
Computational model
Interprets these structures + processes by analogy with computer programs that consists of data structures + algorithms
–> makes the cognitive theory more precise this way
Software program
Refers to the location where the computational model is implemented in a specific programming language
–> this is done to test the model
ex.: programming language like Java
Hardware platform
Refers to the location where the software program runs on
ex.: macintosh, IBM
Name the 5 complex criteria for evaluating theories of mental representation that claim to explain thought.
- Representational power
- Computational power
- Psychological plausibility
- Neurological plausibility
- Practical applicability
Representational power
How much info a particular kind of representation can express
RESULT: Logical relations are difficult to represent in connectionist NWs
Computational power
How mental representations account for 3 important kinds of high-level thinking
- Problem solving
- -> planning, decision, explanation - Learning
- Language
–> being able to learn from experience
Psychological plausibility
Requires accounting for the
a) qualitative capacities of humans
b) quantitative results of psychological experiments concerning these capacities
–> explaining the ways humans accomplish tasks
Neurological plausibility
Using neurological techniques to assess high-level cognition + when and where in the brain certain cognitive tasks are performed
e.g.: EEG
RESULT: Real brain NWs are much more complicated, thus connectionist models are only approx. to the behavior of real neurons
Practical applicability
Cognitive science should be able to increase the understanding of how students learn
–> + suggest how to teach them better
In which way is the stud of mind interdisciplinary ?
It requires the insights + the diversity of methodologies that have been gained by philosophers, psychologists etc
Connectionist research
Emphasizes the importance of connections among simple neuron like structures
–> rebound of the computational model, inspired by the brains neuronal structures
e.g.: units = neurons; degree of activation = frequency with which neurons fire
Local representations
Refers to a model where neuron-like structures are given an identifiable interpretation in terms of specifiable concepts or propositions
e.g.: one neuron stands for one thing
Distributed representations
Refers to a model where NWs learn how to represent concepts or propositions in more complex ways
–> to distribute meaning over complexes of neuron-like structures
THUS: it is trained to accurately respond to stimuli so they can acquire concepts that apply to the stimuli
e.g.: numerous neurons converge to stand for one thing
Parallel constraint satisfaction
Managing a cognitive task in a way where ones processing simultaneously satisfies numerous constraints
–> both models can be used to perform this
ex.: managing to create the perfect timetable for each student in a school
=> negative vs positive internal constraints; + external constraint
Feedforward network
Distributed representations
Refers to a NW where info flows upward through it
Recurrent network
Distributed representations
Refers to a NW where the activation from the output unit feeds back into the input units
Synchrony
Refers to a technique that links units that represent associated elements
Vector
Refer to lists of numbers that can be understood as the firing rates of groups of neurons
What was the broad goal of cognitive science ?
Characterizing the nature of human knowledge
–> forms + contents; how knowledge is
a) used
b) processed
c) acquired
Cognitive science
Refers to the scientific study of the human mind
–> highly interdisciplinary field, combining ideas + methods from
a) psychology
b) linguistics
c) philosophy
The analogy to a computer is useful at all 3 stages of the development of cognitive theories.
Name the 3 stages + why.
- Discovery
- Modification
- Evaluation
–> a) check whether program gets same right answers and makes the same mistakes
b) helps to show that postulated representations and processes are realizable plausible
What will the degree of activation of the units tell us in local representation ?
By this we can judge about the applicability of the concept
–> the truth of proposition
External constraint
Everything that one didn’t program into the program could be an external constraint
–> everything one didn’t take into consideration
Delta rule
Refers to a gradient descent learning rule for updating the weights of the inputs to artificial neurons in a single layer neural NW
–> determines how to update a neural NW during a backpropagation step
Back propagation
Can be used to train the NW by adjusting the weights that connect the different units
–> involves finding a function (Gradient descent) that best maps/allocates a set of inputs to their correct outputs
Gradient descent
Refers to an algorithm that is used to find the set of weights that minimizes the error
Process of learning
Minimize its cost function by trying to come as close as possible to the local minimum that is used in the cost function
Brain-computer interface
BCI
Uses brain activity to control external devices, by amplifying algorithms
–> enables severely disabled patients to interact with the environment
Connectionist NWs
Refer to model that can stimulate learning via hebbian learning + back propagation
–> useful for understanding psychological processes
What do connectionist NWs consist of ?
- Units
- -> have a degree of activations, thus only fire when there is a certain amount of frequency w/ which neurons fire - Links
- -> can be
a) excitatory vs inhibitory
b) one way, thus activation flows from one unit to next
c) symmetric, activation flows back + forth
Positive constraints
Occur when the link between 2 units is excitatory
Negative constraints
Occur when the link between 2 units is inhibitory
External constraint
Involves linking units to special units
–> the special unit refers to one which may affect the unit in a way one did not consider “outsider effect”
Negative internal constraints
Are represented by inhibitory connections
–> come from incompatibly relaxation, which occurs when 2 actions/goals cannot be satisfied together
Positive internal constraints
Are represented by excitatory connections
–> come from facilitation relaxation, which occurs if an action facilitates a goal and they both go together
Activation functions
Refer to mathematical equations that determine the output of a neural NW based on the a given input
–> crucial component of deep learning
Graceful degradation
Refers to the gradual deterioration in performance of a distributed system, which is also o a characteristic of the human cognitive system