Although a number of algorithms exist for the generation of the fundamental frequency contour in automatic text-to-speech conversion systems, the absence of a general theory of intonation still prevents the correct de...
详细信息
Although a number of algorithms exist for the generation of the fundamental frequency contour in automatic text-to-speech conversion systems, the absence of a general theory of intonation still prevents the correct derivation of this important feature in unrestricted text applications. A parallel distributed approach is presented in which two neuralnetworks were designed to learn the F0 values for each phoneme and the F0 fluctuations within each phoneme for words that correspond to a small training set. The neuralnetworks used for this task have demonstrated the ability to generalize their properties on new text, and their level of success depends on the composition and size of the training corpus.
The biological neuralnetworks (BNNs) of the gravity-sensing endorgans of the inner ear, the vestibular maculas, appear to be organized in ways more complex than the parallel distributedprocessing (PDP) systems curre...
详细信息
The biological neuralnetworks (BNNs) of the gravity-sensing endorgans of the inner ear, the vestibular maculas, appear to be organized in ways more complex than the parallel distributedprocessing (PDP) systems currently under study in the artificial neuralnetwork (ANN) field. The organization and synaptic interactions of the processing elements (PEs) have been documented by computer-assisted 3-D reconstructions of terminal and receptive fields from electron micrographs of a series of 570 sections and by the mapping of more than 300 cells. The macular BNN has the following properties: groups of PEs interact to analyze information redundantly;there is an element of randomness in wiring to increase adaptability;PE activity is not fixed but is continually modified by feedforward/feedback loops, which also increases adaptability;and BNN variation in complexity from site to site permits simultaneous analysis of incoming signals for different purposes as a function of place. Data were reduced to symbolic representations and compared with features of ANN PEs. The chief conclusions are that terminal and receptive fields of macular BNNs are not modular, since no two are identical in detail, and that the network changes in threshold properties and in organizational complexity from site to site.
A parallel distributed model is developed for the text-to-speech synthesis problem. This model includes a procedure which marks those incoming letters which are soundless, but which modify the pronunciation of their n...
详细信息
A parallel distributed model is developed for the text-to-speech synthesis problem. This model includes a procedure which marks those incoming letters which are soundless, but which modify the pronunciation of their neighbors. neural-network simulation results are presented, and it is shown that the larger the network the faster the learning rate, but also the greater the computational requirements.
Decomposing a large network into a system of smaller subnets is explored. Each subnet is an independent network that performs some task. These subnets will be linked together in such a way as to collectively solve som...
详细信息
Decomposing a large network into a system of smaller subnets is explored. Each subnet is an independent network that performs some task. These subnets will be linked together in such a way as to collectively solve some larger task. Only the signals from the output units of a subnet are allowed to connect to the inputs of other subnets. That is, no connections will be made from within one subnet to another. Training is applied to each subnet independently. Advantages include distributedprocessing, reduced number of interconnections, and resource sharing. The consequences of pursuing such a system are investigated.
Visual tracking performance is described, and from the nature of experimentally recorded errors an argument is advanced for the location of this uncertainty in the nervous system. It is shown that independent identica...
详细信息
Visual tracking performance is described, and from the nature of experimentally recorded errors an argument is advanced for the location of this uncertainty in the nervous system. It is shown that independent identically distributed motor command noise, n2, which enters either A. Pellionisz's (1988) or W. Daunicht's (1988) eye kinematics expressions, cannot account for the observed error covariance Σe. Therefore, significant variability is introduced earlier, at the time of visual transduction itself, early visual processing stages, or the sensorimotor transformation. Velocity is not sensed directly, but rather requires observation of a change over time. Thus, early estimates of velocity may be poor because the denominator is so small. The authors show velocity measurements 60 ms into the eye response, an average of 180 ± 18 ms after target motion initiation. The orientation of the error ellipse with major axis in the direction of motion suggests that, although speed is poorly estimated at first, direction is better resolved.
It is hypothesized that the various kinds of attentional mechanisms seen in cognitive processing are all related to the modularization that results from the interaction of activity patterns generated by different neur...
详细信息
It is hypothesized that the various kinds of attentional mechanisms seen in cognitive processing are all related to the modularization that results from the interaction of activity patterns generated by different neural populations. The author considers four basic types of neural populations that differ in their distribution of axon terminations and the effect of their neurotransmitters. Nodes with terminations restricted to a single region and circularly distributed around the node center are called short-range (ShR), while nodes with axons going between regions are called long-range columnar (LRC). The terminations of LRC units form a column in their target region which is several times larger than the circle formed by the terminations of an ShR node. The ShR correspond to interneurons, while the LRC correspond to the relay neurons found in the brain. Both the ShR and the LRC units consist of an excitatory and an inhibitory population. The author discusses how a neuralnetwork with these populations can do allocation of attention to sensory, memory, and cognitive channels.
Summary form only given. A neural autoassociator is proposed for transitive closure derivation. A number of neuralnetwork based heuristics are identified which lend themselves to easy solutions in transitive closure ...
详细信息
Summary form only given. A neural autoassociator is proposed for transitive closure derivation. A number of neuralnetwork based heuristics are identified which lend themselves to easy solutions in transitive closure derivations. Several parallel distributedprocessing algorithms are developed in a stepwise manner based on the heuristics and a few theorems are proved which support the correctness and time efficiency of different algorithms in different cases. It is shown that, in performance a closure derivation, the neuralnetwork model self-organizes its connectivity matrix into a transitive closure using the heuristics opportunistically at run time with little overhead. In addition to practical applications, the proposed approach identifies a new category of heuristics called neuralnetwork based heuristics. It is suggested and demonstrated that this category of heuristics may gain high efficiency, which has never been possible by sequential or traditional parallel processing.< >
Summary form only given, as follows. The great majority of computational cognitive models have adhered to the physical symbol system hypothesis (PSSH) of Newell. An approach that seems to be incompatible with the PSSH...
详细信息
Summary form only given, as follows. The great majority of computational cognitive models have adhered to the physical symbol system hypothesis (PSSH) of Newell. An approach that seems to be incompatible with the PSSH is that of parallel distributedprocessing (PDP), or connectionism. It is a controversial issue as to whether PSSH or PDP is a better characterization of the mind. At the root of this controversy are two questions: what sort of computer is the brain, and what sort of programs run on that computer? A theory is presented which bridges the apparent gap between PSSH and PDP approaches.< >
Summary form only given, as follows. A parallel distributedprocessing architecture called an entropy machine (EM) is proposed. This machine, which is based on an artificial neuralnetwork composed of massive neurons ...
详细信息
Summary form only given, as follows. A parallel distributedprocessing architecture called an entropy machine (EM) is proposed. This machine, which is based on an artificial neuralnetwork composed of massive neurons and interconnections, is used for solving a variety of NP-complete optimization problems. The EM performs the parallel distributed gradient descent method or gradient ascent method to search for minima or maxima.< >
Although a number of algorithms exist for the generation of the fundamental frequency contour in automatic text-to-speech conversion systems, the absence of a general theory of intonation still prevents the correct de...
详细信息
Although a number of algorithms exist for the generation of the fundamental frequency contour in automatic text-to-speech conversion systems, the absence of a general theory of intonation still prevents the correct derivation of this important feature in unrestricted text applications. A parallel distributed approach is presented in which two neuralnetworks were designed to learn the F0 values for each phoneme and the F0 fluctuations within each phoneme for words that correspond to a small training set. The neuralnetworks used for this task have demonstrated the ability to generalize their properties on new text, and their level of success depends on the composition and size of the training corpus.< >
暂无评论