D. L. Wang, & B. Yuwono, Incremental learning of complex temporal patterns, IEEE Transactions on Neural Networks, Vol. 70, No 6, pp. 1465-1481, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....shown to be highly sensitive to context [21] and the requirement of encoded context for both learning and storage of context as well as sequence prediction in a hippocampal system [22] The neural network model should also be able to learn sequences in an incremental fashion. Wang and Yuwono [23] argue that this form of learning is desirable as it allows the model to acquire new knowledge on the basis of an existing memory, in a way similar to human learning experience. Incremental learning becomes also relevant if all training data is not available immediately and learning is an ongoing ....

.... after learning a new event, the likelihood of which is dependent on the similarity between old and new events [26] Although the recall may be disrupted, it is better than chance level and reconsolidating the memory of that particular older event is easier than learning it for the first time [23]. As such, retroactive interference is an indication that events are not independently stored inside the brain, but are somehow associated in memory. 1.2 CALM as a neural network for sequence processing The modeling of categorization and identification of sequential information therefore ....

[Article contains additional citation context not shown here]

D. L. Wang, & B. Yuwono, Incremental learning of complex temporal patterns, IEEE Transactions on Neural Networks, Vol. 70, No 6, pp. 1465-1481, 1996.

....trajectories were generated by the ROBOT ICS toolbox of Matlab [31] for a PUMA 560 robot with 6 DOF. These trajectories were previously used to evaluate recurrent [10] and associative memory neural models [32] in temporal sequence based control of robotic arms. As pointed out by Wang Yuwono [33], learning of multiple sequences can be carried out with simultaneous or sequential input presentations, and the latter was chosen in our case. By convention, the robot movements are executed within a cube of dimension 1m 1m 1m. The origin of a coordinate frame for the robot end e ector is ....

....associations for later recall. This temporal association paradigm is widely used in many neural models. The vast majority of these models are based on either multilayer perceptrons (MLP) with some temporal version of backpropagation training [38] or the Hop eld model of associative memory [21] [33]. Also, BAM type [39] 32] and ART type [40] 41] model use the chaining hypothesis to recall temporal sequences. The model proposed in this paper also follows this paradigm; however, in contrast to those models based on MLP and BAM, it learns temporal associations in a self organized manner and ....

[Article contains additional citation context not shown here]

D.-L. Wang and B. Yuwono. Incremental learning of complex temporal patterns. IEEE Transactions on Neural Networks, 7(6):1465-1481, 1996.

....made as long as contextual information is provided to resolve potential ambiguities. The majority of the models for sequence processing are based on either multilayer perceptrons (MLP) with some temporal version of backpropagation training [7] or the Hop eld model of associative memory [6] 28] [41]. Also, BAM type [29] 42] and ART type [43] 44] models use the simple associative chaining hypothesis, and have been applied to a variety of complex tasks in natural language processing, time series analysis, and motor control. The model proposed in this paper also follows this paradigm. ....

....to memory use, it has produced very good results. The automatic determination of the length L is a major problem in models that adopt such a mechanism [5] 19] but recent works on temporal sequence processing have addressed this issue from the perspective of self organizing algorithms (see [41], for example) Choice of Network Parameters : The proposed network has a relatively high number of parameters (nine) They are, however, fairly easy to select and independent of each other in the sense that a change in one of them does not a ect the others. If one always chooses amax = ....

[Article contains additional citation context not shown here]

D.-L. Wang and B. Yuwono, \Incremental learning of complex temporal patterns," IEEE Transactions on Neural Networks, vol. 7, no. 6, pp. 1465-1481, 1996.

....interference (learning a later event interferes with the recall of earlier information) when an animal learns similar events, but that catastrophic interference does not occur. There is some degree of interference but it does not cause catastrophic problems like what we see in artificial networks [43]. CHAPTER 2. LITERATURE SURVEY: A MOTIVATION 15 found that the capacity of the network was severely limited by the sparseness restrictions on the hidden layer. More sophisticated attempts at reducing this catastrophic interference, by French, continued to focus on how hidden layer activation ....

D. Wang and B. Yuwono. Incremental learning of complex temporal patterns. Technical Report OSU-CISRC-6/95-TR30, The Ohio State University, Columbus, OH, 1995.

....interference (learning a later event interferes with the recall of earlier information) when an animal learns similar events, but that catastrophic interference does not occur. There is some degree of interference but it does not cause catastrophic problems like what we see in artificial networks [47]. CHAPTER 2. LITERATURE SURVEY: A MOTIVATION 15 ated with a specific hidden node, using an unsupervised learning, winner take all strategy in addition to the standard supervised back propagation step. This attempt is to make the hidden layer output have only one strongly activated node per input ....

D. Wang and B. Yuwono. Incremental learning of complex temporal patterns. Technical Report OSU-CISRC-6/95-TR30, The Ohio State University, Columbus, OH, 1995. BIBLIOGRAPHY 76

....main controller. However, the proposed system demands additional work to be formalized and implemented in a real world robot. As the majority of the existing unsupervised models for sequence processing does not directly address the issue of multiple temporal sequence learning (see Wang and Yuwono [16] as an exception) the major aim of this work is to develop an unsupervised neural network algorithm to learn and recall temporal patterns and to apply it to robot trajectory tracking. The remaining of the paper is organized as follows. In Section 2, we present the neural algorithm. In Section 3, ....

Wang, D. L. & Yuwono, B. (1996). Incremental learning of complex temporal patterns. IEEE Transactions on Neural Networks, Vol. 7, No. 6, 1465-1481.

....handle repeated patterns. Further tests are being conducted in order to evaluate how the number of items (sampling rate) in a sequence influences the network robustness to noise and the value of the similarity radius. Also, further comparison with other models such as the network proposed by Wang Yuwono (1996) and Srinivasa Ahuja (1999) and the implementation of the model in a real PUMA 560 robot system are being pursued. Acknowledgements: The authors thank FAPESP for its financial support (Project # 98 12699 7) ....

Wang, D. L. & Yuwono, B. (1996). Incremental learning of complex temporal patterns. IEEE Trans. on Neural Networks, 7(6):1465-1481.

No context found.

D. L. Wang, & B. Yuwono, Incremental learning of complex temporal patterns, IEEE Transactions on Neural Networks, Vol. 70, No 6, pp. 1465-1481, 1996.

No context found.

D. L. Wang and B. Yuwono, \Incremental learning of complex temporal patterns," IEEE Trans. on Neural Networks 7, 1465-1481 (1996).

No context found.

D. L. Wang and B. Yuwono, "Incremental learning of complex temporal patterns ", IEEE Trans. on Neural Networks, Vol. 7, No. 6, pp. 1465-1481, 1996.

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