D. Tank and J. Hopfield. Neural computation by concentrating information in time. In Proc. Natl. Acad. Sci. USA, volume 84, pages 1896--1900.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....important property of neurons in the visual cortex (see also [4] The temporal range of pattern selectivity in the present model is clearly limited by the width of the STDP learning window. This range could potentially be increased by using recurrent excitation to provide contextual information [2, 12, 13]. Our current efforts are therefore focused on exploring the effects of redistribution of synaptic efficacies in recurrent spiking networks. Acknowledgments This research is being supported by a National Defense Science and Engineering Graduate Fellowship to APS, a Sloan Research Fellowship to ....

D. Tank and J. Hopfield. Neural computation by concentrating information in time. In Proc. Natl. Acad. Sci. USA, volume 84, pages 1896--1900.

....infeasibility. In this approach, an unconstrained problem is generally formulated as a weighted sum of the objective and the constraints. Examples of methods in this class include simulated annealing (SA) 16] genetic algorithms (GA) 13] tabu search [9] gradient descent [1] Hopfield networks [14], and penalty methods. These methods generally have di#culties in finding feasible solutions when their weights are not chosen properly. A class of penalty methods adjust the penalties (or weights) dynamically according to the amount of constraint violation in order to force all the constraints ....

J. J. Hopfield and D. W. Tank. Neural computation by concentrating information in time. In Proc. National Academy of Sciences, volume 84, pages 1896--1900, Washington, D.C., 1987. National Academy of Sciences.

....optimization problems, some frequently used and closely related terms are described and compared next: Local search, global search and global optimization. Local search methods rely on information from local probes to generate candidate trial points and advance their search trajectories [7, 128, 103]. Obviously, local search methods may be trapped and confined in a small local region in their search space. Global search methods, on the other hand, have techniques for escaping from the attraction of local minima or constrained local minima in their search space [77, 15, 168, 165, 94, 191, 173, ....

J. J. Hopfield and D. W. Tank. Neural computation by concentrating information in time. In Proc. National Academy of Sciences, volume 84, pages 1896--1900.

....triggered by frequency modulated tones. An important part of the architecture of the model is an array of delay lines (see Fig. 3) An array of delay lines has been argued to be neurally plausible (Hopfield Tank, 1989) and it has been used as a basis for temporal pattern recognition (see Tank Hopfield, 1987; Waibel et al. 1989) Latencies of neuronal responses to auditory stimuli have been found at various levels of the auditory pathway, and the range of delays increases greatly in higher auditory structures (Popper Fay, 1992) For instance, electrophysiological recordings in the cat auditory ....

Tank, D.W., & Hopfield, J.J. (1987). Neural computation by concentrating information in time.

....inputted. In addition, this model can recognize the learned patterns correctly even if they are temporally extended or contracted. 1 Introduction Neural networks for pattern recognition usually deal with a spatiotemporal pattern by expanding it into a spatial pattern using time delay filters [Tank and Hopfield, 1987] or multilayer delay units [Waibel, 1989] This method, however, has some drawbacks such that the temporal length of recognizable patterns is limited to the maximum delay time and that temporal extension and contraction is difficult to handle. Another conventional way to recognize spatiotemporal ....

D. W. Tank and J. J. Hopfield, Neural computation by concentrating information in time, Proc. Natl. Acad. Sci. USA 84, 1896--1900, 1987.

....special cases particularly useful is that they can be computed by an incremental update procedure, whereas forms such as the gaussian require evaluating the convolution of the kernel with the input sequence at each time step. Such convolutions, while occasionally used (Bodenhausen Waibel, 1991; Tank Hopfield, 1987; Unnikrishnan, Hopfield, Tank, 1991) are not terribly practical because of the computational and storage requirements. Radford Neal (personal communication, 1992) has suggested a class of kernels that are polynomial functions over a fixed interval of time beginning at a fixed point in the ....

....amounts to back propagation one step in time is not as powerful as full blown back propagation through time or RTRL. Static memory models can be a reasonable approach if there is adequate domain knowledge to constrain the type of information that should be preserved in the memory. For example, Tank and Hopfield (1987) argue for a memory that has high resolution for recent events and decreasing resolution for more distant events. The argument is based on a statistical model of temporal distortions in the input. If there is local temporal uncertainty in the occurrence of an input event, then the uncertainty ....

Tank, D. W., & Hopfield, J. J. (1987). Neural computation by concentrating information in time.

....scene. The next block is the sequence recognition block: the input of this block is a sequence of superquadric parameters from previous stage and the output is the classification of the input sequence. This block has been implemented by an Hopfield neural network with tapped delay input lines [Tank Hopfield, 1986] in order to recognise the sequence of attractors as previously described. The output of this block is the class of the input superquadrics sequence; this information is sent to the symbolic level in order to allow for symbolic inferences. The next block of the system is the hypotheses generation ....

Tank, D.W., & J.J. Hopfield, LL. (1986). Neural Computation by Concentrating Information in Time. Proc. Nat.

....3 space. Although these patterns might be considered temporally displaced versions of the same basic pattern, the vectors are very different. Page 4 There are many ways in which this can be accomplished, and a number of interesting proposals have appeared in the literature (e.g. Jordan, 1986; Tank Hopfield, 1987; Stornetta, Hogg, Huberman, 1987; Watrous Shastri, 1987; Waibel, Hanazawa, Hinton, Shikano, Lang, 1987; Pineda, 1988; Williams Zipser, 1988) One of the most promising was suggested by Jordan (1986) Jordan described a network (shown in Figure 1) containing recurrent connections which ....

Tank, D.W., and Hopfield, J.J. (1987). Neural computation by concentrating information in time.

No context found.

D. Tank and J. Hopfield. Neural computation by concentrating information in time. In Proc. Natl. Acad. Sci. USA, volume 84, pages 1896--1900.

No context found.

D. W. Tank and J. J. Hopfield. Neural computation by concentrating information in time. Proc. Natl. Acad. Sci. USA, 84:1896--1900.

No context found.

D. W. Tank and J. J. Hopfield. Neural computation by concentrating information in time. Proc. Natl. Acad. Sci. USA, 84:1896--1900.

No context found.

Tank, D. and Hopfield, J. (1987). Neural Computation by Concentrating Information in Time. Proc. National Academy of Sciences USA, 84, pp. 1896-1900, April 1987.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC