Neural Computation of Arithmetic Functions
نویسنده
چکیده
The basic processing unit of a neural network i s a linear threshold element. I t has been known that neural networks can be much more powerful than traditional logic circuits, assuming that each threshold element can be built at a cost comparable to that o f AND, OR, Nor logic elements. Whereas any logic circuit o f polynomial size (in n) that computes the product of two n-bit numbers requires unbounded delay, such computations can be done in a neural network with “constant” delay. We improve some known results by showing that the product o f two n-bit numbers and sorting of n n-bit numbers can be computed by a polynomial-size neural network using only 4 and 5 unit delays, respectively. Moreover, the weights of each threshold element in our neural networks require O(1og n)-bit (instead of n-bit) accuracy.
منابع مشابه
Discontinuities in Recurrent Neural Networks
This article studies the computational power of various discontinuous real computational models that are based on the classical analog recurrent neural network (ARNN). This ARNN consists of finite number of neurons; each neuron computes a polynomial net function and a sigmoid-like continuous activation function. We introduce arithmetic networks as ARNN augmented with a few simple discontinuous ...
متن کاملDecidable and Undecidable Arithmetic Functions in Actin Filament Networks
The plasmodium of Physarum polycephalum, called slime mould, is very sensitive to its environment and reacts to stimuli by its appropriate motions. The sensitive stage as well as the motor stage of these reactions are explained by actin filament networks. This paper is devoted to actin filament networks as a computation medium. The point is that actin filaments are sensitive to outer cellular s...
متن کاملComputation of Standard Interval Functions in Multiple-Precision Interval Arithmetic
We present quadratic convergent algorithms for the computation of standard interval functions like √ , ln, arctan in a multiple-precision interval arithmetic. These algorithms depend on elliptic integrals of the first and second kind and the method of arithmetic and geometric means. Представлены квадратично сходящиеся алгоритмы для вычисления стан-дартных интервальных функций, таких как √ , ln,...
متن کاملA fast and low-power distance computation unit dedicated to neural networks, based on redundant arithmetic
This paper presents the design of a fast and low power consumption distance computation unit : ∑ i(Ai −Bi) . It is dedicated to the digital RBF neural network implementation. The proposed architecture is composed of two parts. The first computes the distance (Ai − Bi), and the second performs the sum of these distances. It is based on an efficient squarer in redundant arithmetic. Thank to this ...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2004