Dendritic Neuron Model

Dendritic Neuron Model (DNM) is a novel single neuron model with plastic dendritic morphology. Inspired by biological phenomena of neurons in vivo. the neural structure of DNM is composed of four layers, namely, the synaptic layer, the dendritic layer, the membrane layer and the cell body. Each layer owns diverse excitation functions to carry out corresponding neural  functions. Theoretical and empirical evidence demonstrates the DNM can provide satisfactory performances on various practical applications, such as computer-aided medical diagnosis, business risk assessment and financial time series prediction.

Distinctive Features of the DNM

There are numbers of distinctive features distinguishing the DNM from other conventional artificial neural networks, which have been summarized below:

• The DNM takes into consideration a vast and plastic dendritic tree in the structural morphology of a single neuron.
• Through a neural pruning scheme, the DNM can eliminate superfluous synapses and dendritic branches to simplify its architecture and form an unique neuron morphology for each specific task.
• Through a logic approximation scheme, the DNM can transform the simplified structure into Logic Circuit Classifiers (LCCs), which merely consist of the comparators and logic AND, OR and NOT gates.
• The LCC executes the calculations entirely in binary without sacrificing model performance, and it has extremely fast computation speed and low computational costs.
• The LCC has strong advantages in solving the problems with high-speed data streams, compared with other machine learning algorithms requiring floating-point calculations.
• The LCC is easy to implement for large-scale parallel computing on hardware, such as field programmable gate array (FPGA) and very large scale integration (VLSI).