Jordan M.I. (1989). Serial Order: A parallel, distributed processing approach. In J. L. Elman & D. E. Rumelhart (Eds.), Advances in Connectionist Theory. Hillsdale, NJ: Erlbaum.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....cases feed forward neural networks are used. These networks are effective in producing reactive behavior but cannot deal with time (i.e. they always react in the same way to the same sensory state and therefore cannot integrate information over time) In other cases recurrent neural networks (Jordan, 1989; Elman, 1990) have been used. By including recurrent connections these networks can produce responses that take into account both the current and the previous experienced sensory states. These networks have been successfully applied to a navigation task in which a robot had to periodically return ....

Jordan M.I. (1989). Serial Order: A parallel, distributed processing approach. In J. L. Elman & D. E. Rumelhart (Eds.), Advances in Connectionist Theory. Hillsdale, NJ: Erlbaum.

....variation and event shift functions describing the phrase for each hierarchical level. However, these functions often might not be continuous since an event shift function is undefined during musical periods without note events. Time sequence learning algorithms, including connectionist approaches(Jordan, 1989) and statistical clustering analysis, look promising for learning and representing such functions. 5. Current Work Conclusion It is possible to disassemble musical rhythm into four separate components: a metric hierarchy, two forms of timing functions that operate in cooperation with the ....

Jordan, M. (1989). Serial order: A parallel, distributed processing approach. In Elman, J. L. & Rumelhart D. E. (Ed.), Advances in Connectionist Theory: Speech. Hillsdale: Erlbaum.

....network, and the connectivity pattern and weights between the world and the sensors, and between the sensors and the internal state units and output units. The final recurrent structure of the full network here resembles that of an Elman or Jordan style recurrent network see Elman, 1988, and Jordan, 1986. Here we have drawn the 8 Contents of World W(x,y) Units S[i] Input Hidden Units I[i] Output Units A[i] action c. b. action Output Units A[i] Hidden Units I[i] each with some bias) Output Units A[i] action a. Figure 2. Network implementations of behavioral components. a. A simple ....

Jordan, M.I. (1986). Serial order: A parallel, distributed processing approach. Technical Report ICS-8604.

....the what cat and what coo tasks have desired output vectors of two and eight elements respectively. When trained on these tasks, both experts were given eight output units, and the desired output vector contained six don t care conditions on time steps when the what cat task was to be performed (Jordan, 1986). 30 0.01) On these runs, the expert with very small receptive fields won the competition for the where cat task whereas the expert with very large receptive fields won the competition for the where coo task. For the what cat and what coo tasks, we only examined the case where one expert s ....

Jordan, M. I. (1986) Serial order: A parallel, distributed processing approach. Technical Report 8604, University of California, San Diego.

....Distributed General Problem Solvers, Andrzej Bieszczad 46 Systems and Computer Engineering, Carleton University another output would be associated with it. In a time delay network a similar process applies to all features simultaneously. 4. 5 Recurrent network The Jordan network (Jordan, [28]) is an example of a recurrent architecture (some call it partially recurrent) In addition to the feedforward links present in the backpropagation network, there is a feedback loop here. In the Jordan network, the output signals are fed back to the hidden units through some delay in, so called, ....

Jordan, M. I. (1989), Serial order: A parallel, distributed processing approach, in Advances in Connectionist Theory: Speech.

.... shows that domain theories can be supplied incrementally (as opposed to providing the domain theory at the start of the learning task) Our work on ratle is similar to our earlier work with the fskbann system (Maclin Shavlik, 1993) Fskbann uses a type of recurrent neural network introduced by Jordan (1989) and Elman (1990) that maintains information from previous activations using the recurrent network links. Fskbann extends kbann to deal with state units, but it does not create new state units. Similarly, Omlin and Giles (1992) insert prior knowledge about a finite state automaton into a ....

Jordan, M. (1989). Serial order: A parallel, distributed processing approach. In Elman, J.

....1990, p 180) To this end, a number of researchers have suggested recurrent architectures and new learning algorithms for them (Mozer, 1989; Pearlmutter, 1989; Williams and Zipser, 1989) Two simple recurrent architectures are due to Elman and Jordan and are shown in Figure 2. 2 (Elman, 1990; Jordan, 1989). These are called recurrent architectures because processing loops are created by the backward connections. In Jordan s version, the previous states of the output layer are made available to an additional bank of input units which he calls the state or plan layer. This gives the network a memory ....

Jordan, M. I. (1989). Serial order: A parallel, distributed processing approach. In Elman, J. L. and Rumelhart, D. E., editors, Advances in Connectionist Theory: Speech. Lawrence Erlbaum Associates, Hillsdale, NJ.

....of past history. Figure 3 shows four architectures that use context units (Hertz et al. 1991) Examples of recurrent networks for temporal pattern processing are (i) Elman s (Elman, 1990) sequential recurrent network of Fig. 3. a, that specifies a context from hidden units, and (ii) Jordan network (Jordan, 1989)) that uses feedback from output units to the input layer (Fig. 3.b) Figure 3. Partially recurrent networks with context units. The feedforward arrows represent fully connected adjacent layers while the feedback arrows represent connections only from the ith unit in the source layer to the ith ....

Jordan, M. (1989). Serial order: A parallel, distributed processing approach. In Elman, J. and Rumelhart, D., editors, Advances in Connectionist Theory: Speech.

....feedforward network. Boltzmann machines have been developed based on stochastic units and have been used for tasks such as pattern completion (Hinton Sejnowski, 1986) Time series problems have been attacked with recurrent networks such as the Elman network (Elman, 1990) the Jordan network (Jordan, 1989), and real time recurrent learning (Williams Zipser, 1989) to mention a few. There are neural networks that perform principal component analysis, prototyping, encoding and clustering. Examples of neural network implementations include linear vector quantization (Kohonen, 1989) adaptive ....

Jordan, M.I. (1989) Serial order: A parallel, distributed processing approach in: J. Elman & D. Rumelhart (Eds.) Advances in connectionist theory: Speech Hillsdale: Erlbaum.

....as described by the recurrence relations in Equations 17 and 18. As in the local optimization case, the equations for computing the gradient 10 Alternatively, Figure 9 can be thought of as a special case of Figure 10 in which the backpropagated error signals stop at the state units (cf. Jordan, 1986). involve the multiplication of the performance error y Gamma y by a series of transpose Jacobian matrices, several of which are unknown a priori. Our approach to estimating the unknown factors is once again to learn forward models of the underlying mappings and to propagate signals backward ....

Jordan, M. I. (1986). Serial order: A parallel, distributed processing approach.

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