Neural Networks

Neural network Neural network, a highly interconnected network of information processing elements imitating human brain connectivity and functionality. Neural networks are a type of multiprocessor computer system with simple processing elements, high level interconnections, simple scalar messages, and adaptive interactions between elements. · You can not develop an algorithm solution. · You can get many examples of necessary actions.

A third-generation neural network, a pointed neural network, aims to fill the gap between neuroscience and machine learning using a biologically realistic neuron model for computation. The spike neural network (SNN) is fundamentally different from the neural network known to the machine learning community. SNN works with spikes. Spikes are not continuous values, but are individual events occurring at some point. The appearance of spikes is determined by differential equations representing various biological processes, the most important of which is the membrane potential of neurons. Essentially, when a neuron reaches a certain potential, it jumps up and the potential of the neuron is reset. The most common model is the Leaky integrated ignition (LIF) model. In addition, SNN is usually loosely coupled and uses a special network topology.

The neural network is a model of machine learning and exists for at least 50 years. The basic unit of a neural network is a node roughly based on biological neurons in the mammalian brain. Connections between neurons also mimic the way biological brains develop these connections over time ("training"). From the mid 1980's to the early 1990's, many important architectural advances were made in neural networks. However, the time and data necessary to achieve good results will slow adoption and reduce interest. In the early 21st century, the computing capacity improved dramatically and the industry witnessed the "Cambrian explosion" of computing technology, which was previously impossible. As an important competitor in this field, the decade of explosive computing growth brought deep learning, and won many important machine learning competitions.

A recent interesting recursive neural network architecture is a neural Turing machine. This network is a combination of circular neural network architecture and memory. These neural networks are Turing complete and shown to be able to learn classification algorithms and other computational tasks. Boltzmann Neural Network - One of the earliest and fully connected neural networks is the Boltzmann neural network, also known as the Boltzmann machine. These networks are the first to learn internal expressions and solve very difficult complex problems. One explanation of the Boltzmann machine is that it is a Monte Carlo version of Hopfield recurrent neural network. Nonetheless, neural networks are difficult to train, but when constrained they can prove to be more effective than traditional neural networks. The most common limitation of the Boltzmann machine is to prohibit direct connections between hidden neurons.