China
Africa
Explore chapters and articles related to this topic
Machine Learning Applications In Medical Image Processing
Published in Sanjay Saxena, Sudip Paul, High-Performance Medical Image Processing, 2022
Tanmay Nath, Martin A. Lindquist
The first artificial neural network (ANN) was developed in multiple stages. Its roots lie in the neurological work of Santiago Ramon Cajal who explored the structure of nervous tissues and demonstrated how neurons communicate with each other. These structures were combined with mathematical models to understand how neurons are able to make the computations needed to perform overt behaviors. Using knowledge from neuroscience, neurophysiologist Warren McCulloch and mathematician Walter Pitts developed the first model of a neural network in 1943 [7] and claimed that neurons have binary threshold activation functions. Although McCulloch-Pitts neurons were very simple and allowed only binary input and output, it gave Donald Hebb a platform to propose his revolutionary work in 1949. The Hebb’s rule states “When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A’s efficiency, as one of the cells firing B, is increased” [8]. This illustrates the fact that neural pathways are strengthened each time they are used which is conceptually similar to the way humans learn. This proposal became the fundamental operation necessary for learning and memory. Furthermore, Frank Rosenblatt combined knowledge obtained from McCulloch-Pitts neurons and the findings of Hebb to build the first perceptron in 1962 which later became instrumental in the formation of neural networks (NNs) [9].
Intelligent Sensor Systems
Published in David C. Swanson, ®, 2011
ANNs are a very popular approach to separating classes of data automatically [2]. The concept is based on a biological model for neuron cells and the way electrochemical connections are made between these fascinating cells during learning. In the brain as connections are made between neurons at junctions called synapses to create neural pathways as part of learning, chemicals are deposited which either inhibit or enhance the connection. Each neuron can have from 1000 to 10,000 synapses. The human brain is composed of approximately 20 billion neurons. The number of possible interconnections and feedback loops is obviously extraordinary. Even more fascinating, is the fact that many of these connections are preprogrammed genetically and can be “forgotten” and relearned. While philosophically and biologically interesting, we will explore these issues no further here and concentrate specifically on the “artificial neuron” as part of the most basic ANN based on the generalized delta rule (GDR) for adaptive learning. The reader should be advised that there is far more to ANNs for pattern recognition than presented here and many issues of network design and training are beyond the scope of this book [3].
Fault tolerance and ultimate physical limits of nanocomputation
Published in David Crawley, Konstantin Nikolić, Michael Forshaw, 3D Nanoelectronic Computer Architecture and Implementation, 2020
A S Sadek, K Nikolić, M Forshaw
Another fascinating aspect of neural architecture is its latent ability to adjust and reconfigure its connectivity, this being known as plasticity in neuroscience. The molecular basis of memory is thought to arise from this, due to the Hebbian adjustment of synaptic strengths via the mechanisms of long-term potentiation (LTP) and long-term depression (LTD), but recent work has suggested that it may instead arise through long-term changes in dendritic excitability [124]. Current VLSI architectures are purely static and we have already seen in [5] how the ability to reconfigure computer architectures will be an essential feature in future nanoelectronics. Does the brain use plasticity to reconfigure around defective or damaged areas just as with our designs for future nanoprocessors? This is certainly the case and, indeed, it seems to perform this feat even more efficiently than our projected technologies. For many years it was thought that brain damage that occurred in adulthood was, for the greater part, unrecoverable and that the brain’s architecture became fixed after a period of developmental modelling up until adolescence. Several lines of study have fundamentally altered this view. In patients who suffer a stroke, part of the brain becomes permanently damaged and scarred subsequent to occlusion of the blood supply from a clot. In serious cases, there is a resultant immediate loss in neural function corresponding to the area of the brain affected (for example speech, limb movement, sight etc). However, in the weeks and months that follow an episode, the brain implements active mechanisms of recovery that include adaptive plasticity [125]. If the primary neural pathways in a brain area have become permanently damaged but there are surviving synapses that usually perform only a supporting role, these will become strengthened to reform the neural network. Thus, if say a part of the speech centre was affected subsequent to a stroke and some weak pathways here survive, speech function will recover in time through a moderate degree of relearning by the patient. If there is more serious damage and no neural tissue in this area survives, recovery of function is still possible through an even more remarkable process. Brain areas that usually perform completely different functions can adapt to take on the lost function in addition to their original ones. Naturally though, this takes more time and the lost ability has to be more or less completely relearned by the patient.
Developments in the human machine interface technologies and their applications: a review
Published in Journal of Medical Engineering & Technology, 2021
Harpreet Pal Singh, Parlad Kumar
In case of any injury or problem, the brain has the ability to modify its own connections and reroute the signals along different neural pathways. This is known as neuroplasticity or brain plasticity. Brain plasticity can be considered a crucial component in the recuperation of brain functioning. It is defined as the adaptive capacities of the central nervous system to modify its own organisation and functions [34,35]. The recovery of specific brain functions is based on distinct mechanisms such as unmasking, sprouting, denervation super-sensitivity [36–39]. Brain plasticity is interpreted as the very important fundamental properties of the nervous system that allows enduring functional changes to take place. The working mechanism of brain plasticity can include neurochemical, receptor and neurostructural changes in different portions with distinctive functional characteristics [39–42]. Brain plasticity allows to adaptation of the different prosthetic and assistive devices specifically the sensory substitution devices in divergent environments aptly while performing the different tasks based on skillful activities. More specifically the use of sensory substitution devices is based on the function of brain plasticity.