The history of neural computing is an epic tale of pioneering research, roadblocks, and breakthroughs. This intricate discipline seeks to mimic the workings of animal nervous systems, the complexities of which have fueled advancements in artificial neural networks (ANNs). Here, we take a closer look at the development of this fascinating field, its milestones, and its real-world applications.
The Genesis: McCulloch and Pitts
At the dawn of neural computing, researchers McCulloch and Pitts dedicated themselves to investigating the behavior of animal nervous systems. They hoped to use these insights to establish the blueprint of artificial neural networks (ANNs). However, their model had a notable limitation—it lacked a learning mechanism, a fundamental part of any neural system.
Pioneering Progress: The Rosenblatt Era
In came Frank Rosenblatt, who introduced an iterative algorithm known as the Perceptron Learning Rule. This rule was instrumental in defining the weights in a single-layered network, a structure where all neurons connect directly to inputs. Despite the progress, the field took a hit when Minsky and Papert declared that single-layer perceptrons couldn’t solve specific problems, such as the exclusive OR (XOR) function. This statement caused a substantial reduction in federal funding for neural network research.
Renaissance in the 1980s: The Impact of Hopfield and Backpropagation
The early 1980s saw the field revive with the groundbreaking work of Hopfield. His Hopfield networks—an asynchronous network model—leveraged an energy function to provide potential solutions to NP-complete problems.
The mid-80s marked another milestone with the discovery of backpropagation (commonly known as ‘backprop’). This learning algorithm proved invaluable for multilayered networks and has found applications in predicting Dow Jones averages and reading printed material in optical character recognition systems.
Neural Networks in the Driver’s Seat: Control Systems
Neural networks have found significant application in control systems. For example, the ALVINN project at Carnegie Mellon University leveraged a backpropagation network to sense highway conditions and assist in steering a Navlab vehicle. This system was designed to alert drivers who may be impaired by sleep deprivation, alcohol, or other factors, helping prevent lane deviations.
The potential applications of this technology extend far beyond safety alerts. We envision a future where such systems autonomously drive vehicles, leaving us free to indulge in activities like reading newspapers or chatting on our cell phones. The promise of extra free time makes this an exciting prospect.
Related Articles
