S. Thrun and F. ' Smieja. A general feed-forward algorithm for gradientdescent in connectionist networks. Technical Report 483, GMD, Sankt Augustin, FRG, November 1990. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Modeling Dynamical Systems with Recurrent Neural Networks - Tsung (1994) (1 citation) (Correct)
....forward gradient algorithm for continuous networks; the main difference being that they use more complicated, biologically inspired activation functions that include gap junctions and reversal potentials. However, these algorithms have not been tested in nontrivial simulations. Thrun and Smieja [TS90] generalize the forward gradient methods to include multiple delay links. In feedforward nets, the propagation is assumed to take no extra time; i.e. they have delay 0 links. RTRL nets have delay 1 links. Elman and Jordan nets have a combination of delay 0 forward links and delay 1 backward ....
....equations are: r k ij (t Deltat) j y k fl i (t Deltat) 1 Gamma Deltat k )r k ij (t) Deltat k f 0 (x k (t) Gamma ik w kj y 0 j (t Gamma fl kj ) X l w kl r l ij (t Gamma fl kl ) # (V. 11) These results have been derived independently in [Pea90, TS90] Applying time constant learning, the network is able to learn a much better sine wave, both qualitatively (by visual inspection) and quantitatively (the maximum error on a unit is decreased from 0.1 to below 0.04) For this problem we observed final time constants between 2.0 to 2.5 for the ....
S. Thrun and F. Smieja. A general feed-forward algorithm for gradientdescent in connectionist networks. Technical Report Technical Report No. TR 483, GMD--German National Research Center for Computer Science, FRG, November 1990. 90
On Planning And Exploration In Non-Discrete Environments - Thrun, Möller (1991) (4 citations) Self-citation (Thrun) (Correct)
....that these quantities have been visible some time before, otherwise the mapping of the world is not a function any more. Thus, the world model has to be able to store context information through time. By using a recurrent network [Jor86, Elm88, Ghe89, Moz88, Pin87, Pea88, Pea90, RF87, WZ88, TS90] instead of a feed forward network as a model this problem can be overcome. We will focus on recurrent networks of the Elman type [Elm88] Fig. 2 b) which seem to be best suited for our purpose. If we speak of states in turn, we mean externally observable states rather than internal ones. At a ....
S. Thrun and F. ' Smieja. A general feed-forward algorithm for gradientdescent in connectionist networks. Technical Report 483, GMD, Sankt Augustin, FRG, November 1990.
Adaptive Look-Ahead Planning - Sebastian Thrun, Knut Möller.. (1990) (1 citation) Self-citation (Thrun) (Correct)
....compute an error gradient, which is used for adapting the internal parameters of the model network namely the weights and the biases in order to decrease the prediction error (c.f. figure 1) This is done by propagating the error back through the network using the backpropagation algorithm [16, 18]. Adaptive Look Ahead Planning The planning procedure presented in this paper is an approximation procedure, which starts with a initial plan and improves this plan stepwise by gradient descent in order to maximize the next N reinforcements. Let us assume we have such an initial N step look ahead ....
....in Action Space As mentioned above the environment is modeled by a non recurrent multilayer backpropagation network. This restriction is sufficient for our simulation results the extension of the algorithm to recurrent networks [3, 4, 5, 9, 13, 14, 15, 21] is straightforward and shown in [18]. The external input of the world model network is a state vector s(t) and an action vector a(t) Both state and action vector are the external input I(t) of the model network; for all non input units this external input is 0. The output of the network is the predicted state and the ....
S. Thrun. A general feed-forward algorithm for gradient-descent in neural networks. Technical Report In press, GMD, Sankt Augustin, FRG, 1990.
Planning with an Adaptive World Model - Thrun (1991) (5 citations) Self-citation (Thrun) (Correct)
....model is adopted from [Bar89, Jor89, Mun87] Formally, the world maps actions to subsequent states and reinforcements (Fig. 1) The world model used here is a standard non recurrent or a recurrent connectionist network which is trained by backpropagation or related gradient descent algorithms [WZ88, TS90]. Each time an action is performed on the world their resulting state and reinforcement is compared with the corresponding prediction by the model network. The difference is used for adapting the internal parameters of the model in small steps, in order to improve its accuracy. The resulting model ....
....obtain the optimized plan, the first action 1 of which is then performed on the world. Now the whole procedure is repeated. The gradients of the plan with respect to E reinf can be computed either by backpropagation through the chain of models or by a feed forward algorithm which is related to [WZ88, TS90]: Hand in hand with the activations we propagate also the gradients j is ( j activation j ( action i (s) 1) through the chain of models. Here i labels all action input units and j all units of the whole model network, 1 N ) is the time associated with the th model of the ....
S. Thrun and F. ' Smieja. A general feed-forward algorithm for gradientdescent in connectionist networks. TR 483, GMD, FRG, Nov. 1990.
Dynamic Recurrent Neural Networks: a Dynamical Analysis - Draye, Pavisic, Cheron.. (1996) (1 citation) (Correct)
No context found.
S. Thrun and F. Smieja. A general feed-forward algorithm for gradient-descent in connectionist networks. Technical Report Nr TR 483, GMD-German National Research Center for Computer Science, November 1990.
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