The error rates of linear classifiers that utilize various criterion functions are investigated for the case of two normal distributions with different variances and a priori probabilities. It is found that the classi...
详细信息
The error rates of linear classifiers that utilize various criterion functions are investigated for the case of two normal distributions with different variances and a priori probabilities. It is found that the classifier based on the least mean squares (LMS) criterion often performs considerably worse than the Bayes rate. The perceptron criterion (with suitable safety margin) and the linearized sigmoid generally lead to lower error rates than the LMS criterion, with the sigmoid usually the better of the two. Also investigated are the exceptions to the general trends: only if one class is known to have much larger a priori probability or variance than the other should one expect the LMS or perceptron criteria to be slightly preferable as far as error rate is concerned. The analysis is related to the performance of the back-propagation (BP) classifier, giving some understanding of the success of BP. A neural-net classifier, the adaptive-clustering classifier, suggested by this analysis is compared with BP (modified by using a conjugate-gradient optimization technique) for two problems. It is found that BP usually takes significantly longer to train than the adaptive-clustering technique.< >
We present a new training-out algorithm for neural networks that permits good performance on nonideal hardware with limited analog neuron and weight accuracy. optical neural networks are emphasized with the error sour...
详细信息
We present a new training-out algorithm for neural networks that permits good performance on nonideal hardware with limited analog neuron and weight accuracy. optical neural networks are emphasized with the error sources including nonuniform beam illumination and nonlinear device characteristics. We compensate for processor nonidealities during gated learning (off-line training);thus our algorithm does not require real-time neural networks with adaptive weights. This permits use of high-accuracy nonadaptive weights and reduced hardware complexity. The specific neural network we consider is the Ho-Kashyap associative processor because it provides the largest storage capacity. Simulation results and optical laboratory data are provided. The storage measure we use is the ratio M/N of the number of vectors stored (M) to the dimensionality of the vectors stored (N). We show a storage capacity of M/N = 1.5 on our optical laboratory system with excellent recall accuracy, > 95%. The theoretical maximum storage is M/N = 2 (as N approaches infinity), and thus the storage and performance we demonstrate are impressive considering the processor nonidealities we present. Our techniques can be applied to other neural network algorithms and other nonideal processing hardware.
A high-accuracy optical multiplier that uses an optical correlator is described. A symbolic substitution adder that uses the modified signed-digit number representation is used as the basic module. Emphasis is placed ...
详细信息
A high-accuracy optical multiplier that uses an optical correlator is described. A symbolic substitution adder that uses the modified signed-digit number representation is used as the basic module. Emphasis is placed on the multiplication of many long words in parallel with minimum latency. The encoding method we employ in the adders permits the use of a new optical algorithm and architecture to generate partial products in symbolic form in parallel. Our multiplication algorithm and architecture are shown to be preferable to other optical techniques and to be competitive with digital technology;they are also shown to be particularly attractive for matrix-vector multiplication applications.
We consider a computer-generated hologram for the one-dimensional collimation in x of the output from a linear laser-diode array in y. Our concern is to produce one-dimensional pencil beams from each laser diode with ...
详细信息
We consider a computer-generated hologram for the one-dimensional collimation in x of the output from a linear laser-diode array in y. Our concern is to produce one-dimensional pencil beams from each laser diode with small cross talk between the output from the separate laser diodes. Such outputs can be used in matrix-vector, neural net, and interconnection applications. The efficiency and the design of the computer-generated hologram are detailed, and initial optical laboratory results with an electron-beam recorded computer-generated hologram are presented.
A high-accuracy fixed-point optical adder that operates in parallel on many long words and that uses a pipelined correlator architecture is described. A symbolic substitution algorithm with the modified signed-digit n...
详细信息
A high-accuracy fixed-point optical adder that operates in parallel on many long words and that uses a pipelined correlator architecture is described. A symbolic substitution algorithm with the modified signed-digit number representation is used to perform fixed-point additions with limited carries. A new set of substitution rules and encodings is developed to combine the recognition and substitution steps into one correlation operation. This reduces hardware requirements, improves throughput by reducing the space-bandwidth product needed, and reduces latency (the delay between when data enter the processor and when the final output is available) by a factor of 2. This algorithm and our new modified signed-digit encodings and substitution rules improve the performance of other correlator and noncorrelator optical numeric computing architectures.
The performance of distortion-invariant correlation filters in the presence of background clutter is addressed. Background images are modeled as Markov noise processes, and a synthesis procedure for the optimal filter...
详细信息
The performance of distortion-invariant correlation filters in the presence of background clutter is addressed. Background images are modeled as Markov noise processes, and a synthesis procedure for the optimal filter is described. It is shown that spatially filtering the training set images eliminates the need for the inversion of large noise covariance matrices, thus leading to a computationally efficient filter realization. The effect of errors (in the estimation of clutter correlation coefficient) on filter performance is theoretically analyzed, and a bound on the relative degradation of the SNR due to such errors is presented.
Using multi-sensor image information aids in the interpretation of a represented scene. By segmenting the scene into meaningful parts, we can analyze each individual component and create a description of the scene. Ra...
详细信息
Using multi-sensor image information aids in the interpretation of a represented scene. By segmenting the scene into meaningful parts, we can analyze each individual component and create a description of the scene. Range data is useful for segmenting an image into objects based on the distance relationships between pixels. In this paper, we demonstrate a hybrid optical/digital technique for segmenting range data into object and non-object regions. The object regions found are used as masks to be ANDed with other sensor imagery for identification of the regions.
An efficient hybrid optical/digital technique to generate a string code feature space description of the boundary or contour of an object is described. The string code produces a symbolic description of the angle and ...
详细信息
An efficient hybrid optical/digital technique to generate a string code feature space description of the boundary or contour of an object is described. The string code produces a symbolic description of the angle and length of the arcs describing an object. It is in-plane distortion-invariant and allows both global and local feature analysis.
The first optical laboratory system results employing a direct LU decomposition solution of a system of linear algebraic equations are presented for a finite element problem solution. This also represents the first la...
详细信息
The first optical laboratory system results employing a direct LU decomposition solution of a system of linear algebraic equations are presented for a finite element problem solution. This also represents the first laboratory demonstration of the use of sign-magnitude negative number representation as well as new bit partitioning techniques to increase the accuracy of an optical encoded processor beyond the number of bit channels available.
Several optical pattern recognition techniques suitable for multiple-degree-of-freedom object recognition are described. These include a correlator, an associate processor, and three feature extractors using moments, ...
详细信息
Several optical pattern recognition techniques suitable for multiple-degree-of-freedom object recognition are described. These include a correlator, an associate processor, and three feature extractors using moments, chord distributions, and a space-variant feature space. The processing architectures, the algorithm details, and initial results obtained for each approach are presented.< >
暂无评论