Lec 1-5 Flashcards
What is neuroanatomy key points
What different things can we do to help us
Various staining and labelling used to reveal groups of anatomically similar neurons. Connections made between individual neruson and the aim is to build a wiring diagram.
- neuroanatomy, neurophysiology, psychophysics, imaging techniques, cognitive neurophysiology, computational approach.
Describe neuroanatomy, neurophysiology and psychophysics
Neuroanatomy- various different staining- similar neurons, wiring diagram
Neurophysiology single cell recordings- cell activity recorded by placing tip of microelectrode close to cell. RGCs respond to appropriate stimulation with action potentials and series of them= spike potential. More rapid spiking more vigorous the response. Cell stimulated and then recorded by microelectrodess determine special action of it.
-experimental ablation-destroying animals brain assess it
-stimulating or inhibiting neural activity- electrical and chemical stimulation of brain to investigate parts of brain on certain behaviours. Eg microoelectrodes can be used current to neuron and experimenter uses psychophysics to see what affects this has.
Psychophysics- non invasive uses behavioural responses to assess how visual stimuli being processed. Responses quantifiable.
Imaging techniques, cognitive neurophysiology and computational approach
Imaging techniques- neurophysiology, pet, mri, fair, meg, optical aiming used to show which regions of brain activated by different kinds of stimuli. Optical imaging- pictures of eyes optics.
Cognitive neurophysiology- closed head injury- individuals w brain damage case studies can provide important clues to way brain is structured.
Computational- aims to understand vision by building computer models of visual system. Performance of model compared w psychophysical or neurophysiological data. Not confused w machine vision, both artificial vision system but machine vision no intention of mimicking mammalian vision and cold be several designs for the same goal.
Some of the weird questions and answers
Computational approach- dont help under and how robots see this is machine vision.
If an investigator wished to find the slowest speed detected by human visual system whats old be appropriate investigative technique.
Visual psychophysics as detection tasks are simple with it.
Explain the pond analogy in regards to
The distal stimulus
Proximal stimulus
Perception
Computational theory
And what happens to light intensities when hits our receptors and what is processed.
Detection, the waves arrive, localisation is when something dropped in, recognition dont really know. Pond analogy
-distal stimulus- physical thing out there in the world- the object dropped in water.
-Proximal stimulus- impression that distal stimulus makes on our sense organs so the pattern of the ripples you see in the waves.
-Perception- our conscious visual experience, our brain analyses the proximal stimulus to work out the distal stimulus- work out smth has been dropped.
-Computational theory= generating a formal understanding of the relationship between the distal and proximal stimulus. We need to understand mapping.
light intensities hit our receptors which transmits a signal, visual cognition leads to our perception. Colour, motion is all being processed, the retinal image also requires interpretation.
Explain bottom up and top down. With analogies.
Also talk about ambiguous figures with an analogy
Vision is an active process to make sense of the proximal stimulus:
Bottom up- data driven or descriptive, concerned w description of the proximal image so starts with the retinal image and tries to make sense of it so extracts information from the proximal stimulus and uses it.
Top down- knowledge based or concept driven, using our prior knowledge
(jigsaw analogy- if you look at the box to see where its going then this is top down as knowledge based, if you put similar things together and use shapes and colour this is a bottom up strategy where you are using the description of the proximal image so that’s bottom up)
Bottom up processes build representations that provide cues allowing us to generate a hypothesis that accounts for the data. The final percept is the most successful hypothesis. Some pics the bottom up cues can be the same w 2 different intepretions each is valid so the visual system can flip between the two and we can get the data to fit to either so we can flip back and fourth between them very easily= ambiguous figures when more than one equally good hypothesis fits the data eg necker cube.
Describe hidden figures and accidental assumptions and an analogy
And size constancy
Eg of geometric illusion cues leading to wrong illusion
Sensation and perception
Sometimes when we see lots of things it’s because the system doesnt know theres anything meaningful to search for= hidden figure which is when the cues are reduced so the hypothesis is hard to generate. Once we see it its easy to process as visual system is equipped with info on how to make sense of it and uses top down knowledge-> hypothesis. Top down processes stay w us for life= power of memory.
Penroses impossible triangle- assumptions can lead to mistakes. Accidental alignment so where we see a triangle that doesn’t exist in the real world, only on the back of our eye.
size constancy- as an object moves further away from an observer its retinal image gets smaller yet the object appears further away rather than smaller this is due to depth cues eg texture gradients, perspective, motion parallax= visual scene is depth not flat. the visual system uses these cues to scale up the image size with supposed distance to produce perceived size.
Geometric illusions where cues lead to the wrong illusion eg ponzo illusion.
Sensation and perception- study of sensation= encoding basic parameters eg brightness contrast and motion. Perception is to do with how the visual world acc appears to the observer.
Weird qs-
Which a. b. c. d. e.
of the following is not an example of bottom-up processing?
The responses of retinal receptors.
Data driven processing.
The perception of a regular face when looking at a hollow face stimulus from behind. The Marr and Hildreth edge detector.
Doing a jig-saw while ignoring the picture on the box lid.
Response of retinal receptors is
Data driven processing is
Marr and hildreth edge detector is as it refers to computational model that detects edges in visual input staring from basic visual info and building up to representation of edges. Starts with raw data the image and processes it to detect edges. So bottom up.
Doing jigsaw ignoring= not top down so bottom up using shapes to try and figure it out and use data.
Perception of regular face when looking at hollow face stimulus from behind- hollow or concave face you often see it as a normal convex face as your brain often expects faces to be convex so this uses top down processes. So we see inside out face but it looks like regular face as your brain normally knows faces are regular so it changes what you see to match this explanation. Brains expectations making you see something different. Top down
Top down processing is the motivating force behind the work of David marr?
What is the Dalmatian dog figure and what is the rat man figure and why in the question ‘in research on perception what is an ambiguous figure’ is it not exemplified by these thigns
David marrs work involved understanding computational principles which led to visual perception.
Designed 3 layer framework comp approach etc.
Ratman- ambigious figure= rat or man
Dalmation= dog or shapes abstract ones= amb figure
Though both examples of amb figures and exemplified by perception the q asks for a definition of ambigious figures so read the question carefully. These don’t answer the q eventhough they’re right.
In size constancy when do objects appear to be the same size
Objects appear to be the same size when they are physically the same size
Its not about retinal image sizes it’s about distance or angle you view them from. Process of elimination see if any mean the same thing and eliminate them.
The perceived size of a retinal afterimage is larger when viewing a distance surface than a near surace bc of
what is it not affected by
When does the acc size of retinal afterimage change
Perceived size of retinal afterimage larger distance due to size constancy.
Not affected by distance of the surface against which it is viewed.
Acc size of retinal afterimage changes only when the physical size of the adapting stimulus is changed.
Explain shepherds table and the exam question
Visual trick that makes 2 identical tables look different in size.
Longer one appears bigger though both tables re same size.
We learn that the visual system is not able to read off the content of the retinal image.
(Two table tops diff sizes on retina but look same= no as not diff sizes on retina)
(Tables perceived to be diff distances from us when not= no appear to look the same)
Length or legs are removed-no tis the drawing and the entire table
Visible light,
What are EM signals and what are they measured in
What does a variation in wavelengths lead to
What is saturation
What is the amplitude of it
Why aren’t we sensitive to a narrow range of the EM spectrum.
What does em consist of, why sensitive to them, why sensitive to visible part
visible light is part of the EM.
Em signals are oscillations measured in wavelengths. Variation in wavelengths leads to a variation in hue. Saturation= purity of colour.
Amplitude of it= intensity and brightness of colour.
Em spectrum consists of a range of signals that travel at a wide variety of speeds. Sensitive to them as they reflect back from physical surfaces. And visible part of it is visible as that is where signals have the greatest amplitude.
Why are we sensitive to a narrow range of the EM spectrum- we can’t be sensitive to all light, filtered out so we are left with narrow bandwidth. Light not coloured- subjective experience.
Rating Martians invaded Venus using x ray guns
Wiring of the retina
Describe how receptor cells hyperpolarise
Cones rods
Explain purkinje shift
Degrees of the visual angle 18 what do we have
Retina- back to front. Travels through eyes optics before forming an inverted image on retina. And then everted.
Receptor cells in retina react to light and hyperpolarise in response to amount of light they receive.
Cones- colour vision best bright light, 3 types. Predominate the central fovea 0.5mm. No blue cones in fovea and blue cones are only 10 percent of cone population. Cones most sensitive to 550nm yellow
Rods= low light scotopic conditions more abundant in retinal periphery absent in fovea. Rods more sensitive to blue or green 500nm.
Cones= yellow 550. Rods blue green 500nm. Day yellow objects appear lighter and green objects appear lighter at night= purkinje shift with shift in peak of spectral sensitivity with light levels.
18 degrees of visual angle= optic disc= lack of receptors blind spot. Each eyes is on the nasal side of retina so dont overlap no loss in vision.
Neural circuitry in the retina- graded potentials
Where do rgcs receive their inputs from one or lots recept
Transduction
as light increases, graded potentials decrease but at bipolar cells voltage output increases and the retinal ganglion cells are the output stage which sends the signals to brain via optic nerve.
They work via action potentials good for long distance. Retinal ganglion cells receive inputs from several receptors via interneurons (horizontal bipolar and a machine cells) this is known as convergence= optic disc can remain fairly small.
Transduction- we start off with light as an em wave which is transferred to a signal which is a graded voltage then we get image formation on the retina and light absorption by photo pigment and then electrical current in receptor and then neural activity in RGCs and then we have to relay neural impulses to the brain. Graded not as good over long distances.
Rods cones-
Sensitivity to light wavelengths
Metamers
Cones per retina and rods per retina
as they have diff sensitivity to light so wide range. When light increases above a certain point rods are bleached and are no longer active but cones can work in these conditions. Cones are selective to diff wavelengths= thousands of colours. We can discriminate colour.
Metamers= physically diff but look identical eg pure yellow and a mix of red and green. As cone types increase less metamers as you can distinguish more colours as if they are physically different they wont look identical so less of these.
Colour deficiency. Rods- do not support colour vision.
More rods nasally than temporally. If we had more temporally we would be able to see our nose.
Cones per retina= 8 million. Rods per retina= 120 million.
Convergence in the retina and why
Explain neural adaptation due to fatigue
Excitatory vs inhibitory response
Visual neurons vs collection of cells
RGCs how do they transmit information and what do they consist of
What is the retinal output and how does the image get to that
in retina- optic nerve only has around 1 million fibres but 100 times as many receptors. Rods have high sensitivity low spatial resolution. Higher convergence means its more sensitive to light so receives sm but cant really distinguish between them as well so has a lower spatial resolution.
If you stare at a white spot in black background-> blank paper= black spot due to neural adaptation due to fatigue of these same cells being stimulated over and over and then when they’re not stimulated the firing rates of opposite neurons increases. Smaller size as perception scales down image size eventhough retinal image size same
some response are excitatory others.
Visual neurons have single rf and one response at any one time but a whole collection of cells will have rfs in many different positions producing a spatially distributed pattern of responses.
Image goes through transduction and transformation. The retinal output= neural image.
Retinal ganglion cells transmit info as action potentials higher spike higher stimulus or intensity.
Retinal ganglion cells consist of two concentric subregions called on and off regions sometimes called excitatory and inhibitory regions positive and negative.
RGCs steady state response and increase daycares
We find receptive fields throughout vision, they are the region on the retina that when stimulated causes a change a cells response and it either increases or decreases.
Retinal ganglion cells in the retina in the absence of stimulation or under uniform stimulation release spontaneous discharge or background firing rate so certain types cause an increase and others a decrease.
A retinal ganglion cell’s steady state response will increase if light is introduced into its excitatory sub region. And it will also increase if darkness is introduced into its inhibitory subregion as the contribution from the inhibitory region is being reduced. When light is introduced into the cells inhibitory subregion the response decreases bc of its inhibitory influence. A cells response will also decrease is darkness is introduced into its excitatory subregion as the excitatory influence is being reduced.
In general a rgcs response is determined by the algebraic summation of excitatory and inhibitory influences within its receptive field. Bc this summation is additive the cells are said to be linear. To a first approximation the weights of the excitatory and inhibitory regions are equal and this means that if they receive an equal stimulation they will cancel each other out and the response of the cell will be equal to spontaneous discharge.
Rf of a RGC what did they do and what did they find
stimulated retina w uniform light to see how they change their responses and they found dimmer increased it but no matter how much light shone into eye the response didnt change they still got continuous spontaneous discharge.
Then removed the light so put it in darkness. Came ot realise that each cell has a small circular region that when stimulated will change the response.
This region was split into two areas- on centre cell (if light shone into centre then cells response went up but if shone into surrounding region went down) and off centre. So they mapped out rfs of each cell and found that if light is shone outside them regions it made no change to cells response at all.
More info
Mexican hat
Roughly equal number of on and off centre cells
Some rgcs receive inputs from several receptors
If a bright light is shone into centre of on centre cell= sudden increase in the cells firing rate followed by a gentle decline to a stable level which is higher than the initial response level. This stabilised response level is called the steady-state response. Stimulating excitatory region with light= positive contribution. Stimulating inhibitory region with light= negative contribution to cells output. So a retinal ganglion cells steady state response will increase if light is introduced into its excitatory sub region. A cells response will also decrease if darkness is introduced into a cell’s excitatory subregion as the excitatory influence is being reduced.
Rgcs response= summation of excitatory and inhibitory influences within its rf. bc this summation is additive the cells are said to be linear. To first approximation the excitatory and inhibitory regions are equal so if they receive equal stimulation they cancel each other out and the response of the cell is equal to spontaneous discharge.
The shape of the receptive field looks like a Mexican hat-
narrow but tall centre and the rim is broader but more shallow: total volume of surround eventhough spread over larger area matches the value of the centre which is smaller but taller. This is why there is no net response to diffuse illumination as its cancelled by the inhibitory surround so left w spontaneous discharge. Retinal field size increases with retinal eccentricity. Periphery convergence decreases so we have inputs from many more cons and rods so this is why receptive field size increases with eccentricity.
Diagram the spike trains- plots cells responses as spike trains for 3 different stimuli. Small spot is initially strong response and then levels out and diminishes when stimulus is removed going back to spontaneous discharge. Steady state response after big stimuli presented so response still higher than spontaneous discharge.
Cells response:
light spot in excitatory region- increases + +
dark spot in excitatory- decreases. - +
light spot in inhibitory- decreases + -
dark spot in inhibitory- increases - -
response goes up and down relative to the background firing rate
diagrams with the circles
the spatial arrangement of neighbouring positive and negative regions in a cells rf gives it some interesting properties. RGCs will respond differently from its background firing rate only if there is a d change in luminance eg an edge somewhere within its receptive field.
This diagram shows the steady state response or fitting rates of spatial arrays of neighbouring on centre (middle) and off centre (bottom) rgcs to a light dark luminance border (top) sometimes called a step edge. The responses are shown as a function of the position of the centre of each cells rf so as we move from left to right in the middle and bottom panels of the image we are looking at the response of cells whose rfs are positioned increasingly to the right. The horizontal lines show spontaneous discharge so the firing rate in response to uniform luminance sometimes called background firing rate. The lobes above and below the Laine indicates responses tht are higher and lower than spontaneous discharge.
WHY: this pattern of responses come about consider on centre cells
for those that are positioned so that their whole rf is stimulated by the light region (to the far left) both the on and off regions receive similar input so their contributions will cancel each other out and the cells response will be spontaneous discharge.
Again for cells whose rfs are placed entirely in the dark region so those positioned to the right so neither the on or off regions receive no input so neither on or off is stimulated so neither contribute to cells response and again result is spontaneous discharge
Now for a cell whose rf falls halfway across the light or dark border the on and off regions receive the sa,e amount of light so again same
when a cells rf is just to the left of the border means excitatory centre fully stimulated but inhibitory surround receive less stimulation so contribution from centre is no longer cancelled out by the contribution from the surround and the cells response is greater than spontaneous discharge.
Now if cell is to right of luminance border net decreased less than spontaneous discharge as more darkness in excitatory region and more light in inhibitory region.
Why does this make sense and convolution kernels
Step edge
This makes sense as edges provide important info about objects in the world so we must only transmit info about the edges rather than info to the side of the edges to reduce amount of data being transmitted- data compression so reduces the amount of work that has to be done.
We say the pattern of response in the figure is what we get when we convolve the input pattern or the stimulus with a receptive field. In image processing and machine vision rfs are sometimes called convolution kernels as they help us to convolve the stimulus.
Step edge is a step in the luminance profile. Entire retina covered in rfs and there is also an overlap of fields.
Convolution again
What does convolution do to the image
What is the convolution kernel
What is the output
This process is convolution= transforms the retinal image to neural image.
There is a convolution kernel at each location we are calculating a response of rf. this calculates responses and firing. Mathematical process occurs to take rfs and apply each case along various locations of images. Converts inputs to outputs. Rf= convolution kernel.
And output is the distribution of responses across space as a result of convolution.
Convolve a stimulus with this rf convolution kernel to produce this output. The response is sent from the eye to the brain and convolution transforms retinal image into neural image and the transformation achieves data compression giving us relevant data of where light is present giving us good info about the step edge. Concentrates it around feature of interest rather than lots of info about spontaneous discharge.
