In this paper, we propose a full Bayesian model for neural networks. This model treats the model dimension (number of neurons), model parameters, regularisation parameters and noise parameters as random variables that...
In this paper, we propose a full Bayesian model for neural networks. This model treats the model dimension (number of neurons), model parameters, regularisation parameters and noise parameters as random variables that need to be estimated. We then propose a reversible jump Markov chain Monte Carlo (MCMC) method to perform the necessary computations. We find that the results are not only better than the previously reported ones, but also appear to be robust with respect to the prior specification. Moreover, we present a geometric convergence theorem for the algorithm.
The speech waveform can be modelled as a piecewise-stationary linear stochastic state space system, and its parameters can be estimated using an expectation-maximisation (EM) algorithm. One problem is the initialisati...
The speech waveform can be modelled as a piecewise-stationary linear stochastic state space system, and its parameters can be estimated using an expectation-maximisation (EM) algorithm. One problem is the initialisation of the EM algorithm. Standard initialisation schemes can lead to poor formant trajectories. But these trajectories however are important for vowel intelligibility. The aim of this paper is to investigate the suitability of subspace identification methods to initialise *** paper compares the subspace state space system identification (4SID) method with the EM algorithm. The 4SID and EM methods are similar in that they both estimate a state sequence (but using Kalman filters and Kalman smoothers respectively), and then estimate parameters (but using least-squares and maximum likelihood respectively). The similarity of 4SID and EM motivates the use of 4SID to initialise EM. Also, 4SID is non-iterative and requires no initialisation, whereas EM is iterative and requires initialisation. However 4SID is sub-optimal compared to EM in a probabilistic sense. During experiments on real speech, 4SID methods compare favourably with conventional initialisation techniques. They produce smoother formant trajectories, have greater frequency resolution, and produce higher likelihoods.
Actuator delays can substantially degrade the performance of vehicle following control. In this paper, we investigate the control design problem of vehicle following systems with actuator delay. The main requirement i...
详细信息
Actuator delays can substantially degrade the performance of vehicle following control. In this paper, we investigate the control design problem of vehicle following systems with actuator delay. The main requirement in designing such systems is to ensure the stability of the individual vehicle and that there are no slinky-effects in the platooning. Based on Lyapounov's method, we first construct a constant bound for the time delay to guarantee the individual vehicle stability. Next, we derive a condition for achieving no slinky-effects. Finally, we prove that the proposed control laws can achieve zero steady state errors.
In this paper, we study the robust stability of a linear system with uncertainties. By considering the sign of the uncertain bounds and utilizing matrix measures, an improved result of robust stability is presented. F...
详细信息
In this paper, we study the robust stability of a linear system with uncertainties. By considering the sign of the uncertain bounds and utilizing matrix measures, an improved result of robust stability is presented. Furthermore, the proposed method is extended to linear time-delay systems with uncertainties. Several examples are given to illustrate the advantages of the presented method.
作者：
Kosecka, JRobotics laboratory
Department of Electrical Engineering and Computer Science 333 Cory Hall 98 UC Berkeley Berkeley CA 94720-1772 USA
Rich sensory information, robust control strategies and proper representation of the environment are crucial for successful navigation of the mobile robot. We propose a model of the environment which is suitable for g...
详细信息
Rich sensory information, robust control strategies and proper representation of the environment are crucial for successful navigation of the mobile robot. We propose a model of the environment which is suitable for global navigation using visual sensing. At the lowest level of interaction of the robot with the environment we employ visual servoing techniques which facilitate robust local navigation by means of relative positioning. Further we demonstrate how to use these local control strategies in a global setting. The environment is represented in terms of place graph, where the nodes correspond to places and arcs have associated servoing strategies for moving from one place to another. The global navigation is expressed as a sequence of relative positioning tasks obtainable from the search of a place graph. The proposed model facilitates generation of motion plans which can be executed in a robust manner thanks to the presence of the sensory information in the feedback loop.
The spectral response, luminance, and outgassing characteristics of new, thin-film Ce-activated SrGa2S4 phosphors for application in field-emission flat-panel displays are presented. The tested phosphor materials are ...
详细信息
The spectral response, luminance, and outgassing characteristics of new, thin-film Ce-activated SrGa2S4 phosphors for application in field-emission flat-panel displays are presented. The tested phosphor materials are prepared using molecular beam epitaxy. The sample screens are tested with e-beam excitation at currents up to 50 μA within the 200-2000 V (i.e., "low-voltage") and 3-8 kV (i.e., "medium-voltage") ranges. The spectral coordinates are compared to those of commercial low-voltage phosphors. Phosphor outgassing, as a function of time, is measured with a residual gas analyzer at a fixed 50-μA beam current in the low-voltage range. We find that levels of outgassing stabilize to low values within the first hour of excitation. The desorption rates measured from thin films are compared to previously studied desorption parameters for powder phosphor layers.
Yao showed how to use a sufficiently secure cryptographic permutation to construct pseudo-random generators to de-randomize arbitrary randomized algorithms. To do this, he used the fact that the XOR of independent ran...
详细信息
We present an Analog Neural Network Synthesis System based on a circuit simulator and a silicon assembler for neural networks. The circuit simulator makes use of the fact that neural networks with multilayer perceptro...
详细信息
ISBN:
(纸本)0780341236
We present an Analog Neural Network Synthesis System based on a circuit simulator and a silicon assembler for neural networks. The circuit simulator makes use of the fact that neural networks with multilayer perceptron architecture consist of many decoupled blocks if the blocks are designed in MOS technology. We implement on-chip training on the software by incorporating the Madaline Rule III into our simulator. The assembler generates the layout by reading the standard cells from a library once the architecture of the network is given.
暂无评论