Chapter 3 - The Turn to the Brain Flashcards
How did cognitive scientists in the early years of the field address their inability to research the brain on a fine level? In what way did they regard it?
They treated the brain as a functional system, attempting to understand its functions and result of its inner complexities (the software to its biological and physical hardware). The brain and its functional systems were reverse engineered, in the sense that its functions and qualities were taken and attempted, and cognitive scientists attempted to find where they manifest in a system.
How did Gorman and Sejnowski teach a system to detect the difference in rock and mine SONAR signals?
They used spectral analyzers to transform the echoes of SONAR signals into frequency values, and ran those values through the system, with an input frequency being matched with an expected output of either mine or rock. They continued this and used backpropagation to teach the system to recognize the difference in frequencies. Input units in the system respond to different frequencies, and through calibration of the weight of the system created connections that can reliably distinguish the SONAR signals of mines and rocks.
How did Peterson et al. design their experiment to address the shortcomings of a PET scan and show the different areas involved in lexical processing?
They took PET scans of the brain during different similarly focused activities, and used the difference in the scans to determine what parts of the brain are specifically utilized for the task. Taking a base case of attention towards the center of a blank TV, they took scans during increasingly complex but relevant cases to document the functional regions used in each.
How do parallel neural networks start, and how do they develop into their expected form/connection distribution?
Parallel neural networks start with a random weight assignment, and quickly recalibrate their weight as a result of inputs being sent in and a target output being meant to be reached. Natural algorithms changes the weight of the network until it reaches its final connection distribution.
How do units compare with each other?
They are identical structurally. The difference between them is created through their interaction with units from other layers. Think of the weight of the system and the subsequent organization of connections between units.
How was Ungerleider’s and Mishkin’s study of visual processes revolutionary in cognitive science?
It used bottoms up research methods, with the data gained from exploring the visual organs of the brain and experimenting on them being used to form theories on the overall function of the system as a whole.
What approach to research was commonly used in cognitive science before the 1980s? What caused a shift away from it?
The top down approach to research and the formulation of theories, which was used because there weren’t many tools available for the research of the brain on a microscopic level and over time. Imaging and obversational tools emerged in the 1980s that allowed for the research of the brain in a way that facilitated the fields move towards bottoms up research
What are limitations of the PET scan?
The PET scan can highlight areas with brain activity that does not involve the processes being examined, background activity.
The PET scan can highlight areas with brain activity that does not involve the processes being examined, background activity.
It does not address the limits the brain and its structures, or the hardware, might place on the functioning of cognition and its qualities, the software as observed by cognitive scientists. Forgets the biological changes over time, with gradual but gracefull degradation in age. The temporal reality of cognition, with many of the simplified observed phenomena not reflecting real conditions, as a lot of the details are abstracted away. Reaction time in most contexts is inaccurate.
What did Korbinian Brodmann do to serve king, country, and cognitive science?
He divided the cerebral cortex into 52 separate and functional areas.
What did Peterson et al. prove in their experiment on lexical processing?
Word recognition and processing is parallel process. It suggested that visual information has a direct and separate paths into areas involved in articulately coding and motor programming
What does removing the primary visual cortex from one hemisphere and the inferior temporal area from the other cause? Who experimented on and discovered this?
his does not affect visual capacity too much, as the information from the remaining primary visual cortex is sent to the other hemisphere via the corpus collosum to the remaining inferior temporal area. The biological observation was discovered via experiment by Ungerleider and Mishkin
What does removing the primary visual cortex from one hemisphere and the posterior parietal cortex from the other cause? Who experimented on and discovered this?
This cripples visual capacity. Removing the primary visual cortex has negative effects on the functioning of the visual system, and removing the posterior parietal cortex from the other hemisphere results in severe deficiencies. The biological observation was discovered via experiment by Ungerleider and Mishkin
What is a layer in parallel processing?
A layer is a layer of unconnected units that connects via its units to other layers. Its component units receive information from the previous layer, and sends them forward to the next one. Weighing is important in terms of what the units the units of a layer are connected to.
What is a learning algorithm that continously loops until the input results in a target output and error has been diminished to near zero?
Backpropagation
What is a parallel processing system? Who discovered and pioneered it?
Parallel processing is the splitting of a neural network into units organized into connected layers. Each unit can have activation values ranging between [-1, 1], which is determined by a units connection to other units in the previous layer. Rummelhart, McCelland, and the PDP research group. (Imagine Rommel with a beating heart in McDonalds sealand reading a pdf)
What is a PET scan and how does it work?
A PET scan (positron emission tomography) is a functional scan of the brain, which records areas of activity in it.
It works by injecting (a safe amount of) radioactive oxygen-15 isotopes in water into the brain, and tracking their distribution by the emission of radiation.
What is backpropagation?
A learning algorithm that continously loops until the input results in a target output and error has been diminished to near zero.
What is learning in parallel processing systems?
Learning is essentially the modification of weight in a neural system such that when an input is placed into it, the real output will more closely approach the expected one.
This is an algorithmic shift??
What is the idea of multiple realizability?
It is when the same function can be performed by more than one distinct structures. An example is the heart, which differs between animals.
What is the idea of weight in parallel processesing? What properties does it allow for, and how does it change?
Weight is simply the connectivity of units across plates in a parallel neurological system.
It determines the path signals have to take, and because it can be changed, it allows for parallel systems to adapt.
When inputs are put through a system, it compares the real output with the expected output, and it modifies itself algorithmically to be able to more accurately output the expected values/functions. This process is called learning.
What is the structure of a parallel neural network?
A parallel neural network is composed of multiple layers of units, with the first layer called the input layer receiving information from outside the system and the final layer called the output layer, much the same sending out.
The layers in between are called the hidden layers, which propagate and process information in a manner pertaining to the weight of the system. This occurs within the system itself
What justifies the idea of multiple realizability?
Humans and animals have different brains and internal structures, but share similar cognition, with us both being able to feel pain, the cold, see, etc.
What makes reverse engineering and to a certain extent studying the more complex parts of cognition less valuable?
The biological reality of multiply realizablity, which means that many of the functions in cognition can be performed by various structures of one is damaged.
Different structures can perform the same task, if one is damaged the other can replace it.
