Complex Object Recognition Flashcards
How does information within the cortex flow?
it enters through layer IV, is processed and integrated in layers II and III, and is sent out via layers V and VI
What does each centre in the thalamus have?
a different flow of information through the laminae
What do inhibitory neurons tend to connect with?
other neurons within the same layer
What are long range cortical connections mostly?
completely excitatory that can also synapse onto a local inhibitory neuron
How does the brain integrate sensory information?
pathways between cortical areas enable hierarchical, parallel, and distributed processing
What is the first cortical area to receive visual information?
V1
What are the 2 visual streams from V1 to higher visual areas?
- ventral stream: V1 → V2 → V4 → Inferior temporal cortex (IT)
- dorsal stream: V1 → V2 → middle temporal area (MT) → posterior parietal cortex (PPC)
What does the IT cortex contain?
TE and TEO
What does the PPC contain?
LIP and 7a
What is the ventral visual stream involved in?
recognition and identification of objects, faces and colours
What are the different regions in the ventral stream responsible for?
- fusiform face area (FFA) specialises in face recognition
- parahippocampal place area (PPA) is involved in recognising places and scenes
- lateral occipital complex (LOC) is important for object recognition and distinguishing complex shapes
What is meant by processing along the ventral visual stream?
each step involves progressively more complex analysis of the visual input; it moves from basic features (like lines and edges in V1) to complex shapes and object recognition in the IT cortex
What sensory modalities does the ventral visual stream interact with?
language and memory to provide a comprehensive understanding of objects and their significance
What does a lesion in the ventral visual stream cause?
visual agnosia
What are symptoms of visual agnosia?
- patient does not recognise her mother visually but knows her voice
- lower vision is preserved but higher vision (e.g. edge detection) is lost
- patient unable to indicate the size, shape and orientation of an object but has normal hand pre-shaping and rotation when reaching to grasp
What is the dorsal visual stream involved in?
vision for action i.e. spatial awareness, motion detection and visual-motor integration
What are the different regions in the dorsal visual stream responsible for?
- MT/V5 is important for analysing motion and speed of moving objects
- intraparietal sulcus (IPS) is involved in visually guided movements, such as reaching and eye movements (saccades)
- superior parietal lobule processes information related to hand-eye coordination and the perception of the body in space
What does the dorsal visual stream allow?
the brain to translate visual information into precise motor commands, enabling us to interact effectively with our environment
What is the hierarchical planning of the dorsal visual stream?
basic motion detection to complex spatial analysis and motor planning
What does PPC lesion cause?
optic ataxia or visuomotor impairment
What are characteristics of PPC lesion?
- difficulty with visually guided reaching
- intact visual fields, stereoscopic (3D) vision, oculomotor control, proprioception, motor abilities and cerebellar function
- patients can perform the task if they receive proprioceptive or auditory cues (unlike cerebellar ataxia)
What is brain plasticity?
the ability to reorganise itself in response to learning, injury, or changes in sensory input
What can brain plasticity lead to?
variability in where certain functions are localised
Give an example of brain plasticity
in individuals who are blind, areas of the brain typically used for visual processing may be repurposed for processing auditory or tactile information
What is object recognition invariant against?
transformations e.g. rotation, translation and scaling
What are grandmother cells?
neurons that respond selectively and specifically to a particular stimulus, such as a specific face (e.g. one’s grandmother), a particular object, or a unique concept
How are face-selective IT neurons arranged?
clustered in patches
What is present in the medial temporal lobe?
neurons with very complex selectivities
Where is the MTL?
downstream from the IT cortex
What does the MT transition from and to what?
6-layered neocortex to 3 or 4-layered allocortex
What increases from the bottom to the top of the ventral stream?
selectivity for real world objects
What is retinal circuitry important for and why?
object recognition since it produces the on and off-centre ganglion cells which are critical for phase invariance in V1 complex cells
What can ANNs based on the retina and cortex do?
perform object recognition
What is required to create a grandmother cell in an ANN?
a system to be designed where a single neuron or a very small set of neurons responds exclusively to a highly specific input, such as a particular face or object
What should a network that recognises a grandmother do?
- output 1 when the input is a black and white 2D grandmother array
- output 0 when the input is another array
What is required for an ANN to compute a good approximation to any function?
enough layers of neurons and enough neurons in each layer
What are deep networks?
ANNs with many layers inspired by the cortex
What is deep learning?
a type of machine learning that uses deep networks to learn from large amounts of data; it mimics how the human brain processes information by recognising patterns, making decisions, and solving complex problems
What is deep learning used by?
- banks to read cheques
- Google and Microsoft for speech recognition
- Google to improve translate
- Waymo for object recognition in self-driving cars
What is the main difference between ANNs and biological neural networks?
how they learn; ANNs can compute anything, but for a specific task, a particular ANN must be trained and structured
How are ANNs trained?
by backpropagation (does not resemble leaning by biological neural networks)