Variational studies and replica symmetry breaking in the generalization problem of the binary perceptron

We analyze the average performance of a general class of learning algorithms for the nondeterministic polynomial time complete problem of rule extraction by a binary perceptron. The examples are generated by a rule implemented by a teacher network of similar architecture. A variational approach is used in trying to identify the potential energy that leads to the largest generalization in the thermodynamic limit. We restrict our search to algorithms that always satisfy the binary constraints. A replica symmetric ansatz leads to a learning algorithm which presents a phase transition in violation of an information theoretical bound. Stability analysis shows that this is due to a failure of the replica symmetric ansatz and the first step of replica symmetry breaking (RSB) is studied. The variational method does not determine a unique potential but it allows construction of a class with a unique minimum within each first order valley. Members of this class improve on the performance of Gibbs algorithm but fail to reach the Bayesian limit in the low generalization phase. They even fail to reach the performance of the best binary, an optimal clipping of the barycenter of version space. We find a trade-off between a good low performance and early onset of perfect generalization. Although the RSB may be locally stable we discuss the possibility that it fails to be the correct saddle point globally. ©2000 The American Physical Society.

Quantum tunneling fragmentation model

A nonthermal quantum mechanical statistical fragmentation model based on tunneling of particles through potential barriers is studied in compact two- and three-dimensional systems. It is shown that this fragmentation dynamics gives origin to several static and dynamic scaling relations. The critical exponents are found and compared with those obtained in classical statistical models of fragmentation of general interest, in particular with thermal fragmentation involving classical processes over potential barriers. Besides its general theoretical interest, the fragmentation dynamics discussed here is complementary to classical fragmentation dynamics of interest in chemical kinetics and can be useful in the study of a number of other dynamic processes such as nuclear fragmentation. ©2000 The American Physical Society.

Optimal design of smart structures using bonded piezoelectrics for vibration control

Smart material technology has become an area of increasing interest for the development of lighter and stronger structures which are able to incorporate actuator and sensor capabilities for collocated control. In the design of actively controlled structures, the determination of the actuator locations and the controller gains, is a very important issue. For that purpose, smart material modelling, modal analysis methods, control and optimization techniques are the most important ingredients to be taken into account. The optimization problem to be solved in this context presents two interdependent aspects. The first one is related to the discrete optimal actuator location selection problem, which is solved in this paper using genetic algorithms. The second is represented by a continuous variable optimization problem, through which the control gains are determined using classical techniques. A cantilever Euler-Bernoulli beam is used to illustrate the presented methodology.

Novos resultados sobre a estabilidade e controle de sistemas nao-lineares utilizando modelos fuzzy e LMI

Relaxed conditions for the stability study of nonlinear, continuous and discrete-time systems given by fuzzy models are presented. A theoretical analysis shows that the proposed method provides better or at least the same results of the methods presented in the literature. Digital simulations exemplify this fact. These results are also used for the fuzzy regulators design. The nonlinear systems are represented by the fuzzy models proposed by Takagi and Sugeno. The stability analysis and the design of controllers are described by LMIs (Linear Matrix Inequalities), that can be solved efficiently by convex programming techniques. The specification of the decay rate, constraints on control input and output are also described by LMIs. Finally, the proposed design method is applied in the control of an inverted pendulum.

Teaching real-time with a scheduler simulator

In this paper we describe a scheduler simulator for real-time tasks, RTsim, that can be used as a tool to teach real-time scheduling algorithms. It simulates a variety of preprogrammed scheduling policies for single and multi-processor systems and simple algorithm variants introduced by its user. Using RTsim students can conduct experiments that will allow them to understand the effects of each policy given different load conditions and learn which policy is better for different workloads. We show how to use RTsim as a learning tool and the results achieved with its application on the Real-Time Systems course taught at the B.Sc. on Computer Science at Paulista State University - Unesp - at Rio Preto.

A barrier method for constrained nonlinear optimization using a modified Hopfield network

The ability of neural networks to realize some complex nonlinear function makes them attractive for system identification. This paper describes a novel barrier method using artificial neural networks to solve robust parameter estimation problems for nonlinear model with unknown-but-bounded errors and uncertainties. This problem can be represented by a typical constrained optimization problem. More specifically, a modified Hopfield network is developed and its internal parameters are computed using the valid-subspace technique. These parameters guarantee the network convergence to the equilibrium points. A solution for the robust estimation problem with unknown-but-bounded error corresponds to an equilibrium point of the network. Simulation results are presented as an illustration of the proposed approach.

An accurate low-voltage analog memory-cell with built-in multiplication

A CMOS memory-cell for dynamic storage of analog data and suitable for LVLP applications is proposed. Information is memorized as the gate-voltage of input-transistor of a gain-boosting triode-transconductor. The enhanced output-resistance improves accuracy on reading out the sampled currents. Additionally, a four-quadrant multiplication between the input to regulation-amplifier of the transconductor and the stored voltage is provided. Designing complies with a low-voltage 1.2μm N-well CMOS fabrication process. For a 1.3V-supply, CCELL=3.6pF and sampling interval is 0.25μA≤ ISAMPLE ≤ 0.75μA. The specified retention time is 1.28ms and corresponds to a charge-variation of 1% due to junction leakage @75°C. A range of MR simulations confirm circuit performance. Absolute read-out error is below O.40% while the four-quadrant multiplier nonlinearity, at full-scale is 8.2%. Maximum stand-by consumption is 3.6μW/cell.

A low-voltage programmable-gain CMOS amplifier with very-low temperature-drift

A new topology for a LVLP variable-gain CMOS amplifier is presented. Input- and load-stage are built around triode-transconductors so that voltage-gain is fully defined by a linear relationship involving only device-geometries and biases. Excellent gain-accuracy, temperature-insensitivity; and wide range of programmability, are thus achieved. Moreover, adaptative biasing improves the common-mode voltage stability upon gain-adjusting. As an example, a 0-40dB programmablegain audio-amplifier is designed. Its performance is supported by a range of simulations. For VDD=1.8V and 20dB-nominal gain, one has Av=19.97dB, f3db=770KHz and quiescent dissipation of 378μW. Over temperatures from -25°C to 125°C, the 0. ldB-bandwidth is 52KHz. Dynamic-range is optimized to 57.2dB and 42.6dB for gains of 20dB and 40dB, respectively. THD figures correspond to -60.6dB@Vout= 1Vpp and -79.7dB@Vout= 0.5 Vpp. A nearly constant bandwidth for different gains is also attained.

A low-voltage triode-MOSFET four-quadrant multiplier with optimized current-efficiency

A low-voltage, low-power four-quadrant analog multiplier with optimized current-efficiency is presented. Its core corresponds to a pseudodifferential cascode, gain-boosting triode-transconductor. According to a low-voltage 1.2μm CMOS n-well process, operand differential-amplitudes are 1.0Vpp and 0.32Vpp for a 1.3V-supply. Common-mode voltages are properly chosen to maximize current-efficiency to 58%. Total quiescent dissipation is 260μW. A range of PSPICE simulation supports theoretical analysis. Excellent linearity is observed on dc characteristic. Assuming a ±0.5% mismatch on (W/L) and VTH THD at full-scale is 0.93% and 1.42%, for output frequencies of 1MHz and 10MHz, respectively.

Projeto E análise de uma rede neural para resolver problemas de programação dinâmica

Systems based on artificial neural networks have high computational rates due to the use of a massive number of simple processing elements and the high degree of connectivity between these elements. Neural networks with feedback connections provide a computing model capable of solving a large class of optimization problems. This paper presents a novel approach for solving dynamic programming problems using artificial neural networks. More specifically, a modified Hopfield network is developed and its internal parameters are computed using the valid-subspace technique. These parameters guarantee the convergence of the network to the equilibrium points which represent solutions (not necessarily optimal) for the dynamic programming problem. Simulated examples are presented and compared with other neural networks. The results demonstrate that proposed method gives a significant improvement.

Stability of trapped Bose-Einstein condensates

In three-dimensional trapped Bose-Einstein condensate (BEC), described by the time-dependent Gross-Pitaevskii-Ginzburg equation, we study the effect of initial conditions on stability using a Gaussian variational approach and exact numerical simulations. We also discuss the validity of the criterion for stability suggested by Vakhitov and Kolokolov. The maximum initial chirp (initial focusing defocusing of cloud) that can lead a stable condensate to collapse even before the number of atoms reaches its critical limit is obtained for several specific cases. When we consider two- and three-body nonlinear terms, with negative cubic and positive quintic terms, we have the conditions for the existence of two phases in the condensate. In this case, the magnitude of the oscillations between the two phases are studied considering sufficient large initial chirps. The occurrence of collapse in a BEC with repulsive two-body interaction is also shown to be possible.

Autosolitons in trapped Bose-Einstein condensates with two- and three-body inelastic processes

The conditions for the existence of autosolitons were considered in trapped Bose-Einstein condensates with attractive atomic interactions. The expression for the parameters of the autosoliton was derived using the time-dependent variational approach for the nonconservative 3-dimensional Gross-pitaevskii equation and their stability was checked. The results were in agreement with the exact numerical calculations. It was shown that the transition from unstable to stable point solely depends on the magnitude of the parameters.

Precision measurements of the ∧c + and D0 lifetimes

A measurement technique of charm baryons lifetimes from hadro-production data was presented. The measurement verified the lifetime analysis procedure in a sample with higher statistical precision. Other effects studied include mass reflections; effects of the presence of a second charm particle; and mismeasurement of charm decays. Monte carlo simulations were used for the detailed study of systematic effects using the charm data.

Diet, phylogeny, and basal metabolic rate in phyllostomid bats

Aside from the pervasive effects of body mass, much controversy exists as to what factors account for interspecific variation in basal metabolic rates (BMR) of mammals; however, both diet and phylogeny have been strongly implicated. We examined variation in BMR within the New World bat family Phyllostomidae, which shows the largest diversity of food habits among mammalian families, including frugivorous, nectarivorous, insectivorous, carnivorous and blood-eating species. For 27 species, diet was taken from the literature and BMR was either measured on animals captured in Brazil or extracted from the literature. Conventional (nonphylogenetic) analysis of covariance (ANCOVA), with body mass as the covariate, was first used to test the effects of diet on BMR. In this analysis, which assumes that all species evolved simultaneously from a single ancestor (i.e., a star phylogeny), diet exerted a strong effect on mass-in-dependent BMR: nectarivorous bats showed higher mass-independent BMR than other bats feeding on fruits, insects or blood. In phylogenetic ANCOVAs via Monte Carlo computer simulation, which assume that species are part of a branching hierarchical phylogeny, no statistically significant effect of diet on BMR was observed. Hence, results of the nonphylogenetic analysis were misleading because the critical values for testing the effect of diet were underestimated. However, in this sample of bats, diet is perfectly confounded with phylogeny, because the four dietary categories represent four separate subclades, which greatly reduces statistical power to detect a diet (= subclade) effect. But even if diet did appear to exert an influence on BMR in this sample of bats, it would not be logically possible to separate this effect from the possibility that the dietary categories differ for some other reason (i.e., another synapomorphy of one or more of the subclades). Examples such as this highlight the importance of considering phylogenetic relationships when designing new comparative studies, as well as when analyzing existing data sets. We also discuss some possible reasons why BMR may not coadapt with diet. © by Urban & Fischer Verlag.

Transient stability indices from a hybrid approach

This paper deals with hybrid method for transient stability analysis combining time domain simulation and a direct method. Nowadays, the step-by-step simulation is the best available tool for allowing the uses of detailed models and for providing reliable results. The main limitation of this approach involves the large time of computational simulations and the absence of stability margin. On the other hand, direct methods, that demand less CPU time, did not show ample reliability and applicability yet. The best way seems to be using hybrid solutions, in which a direct method is incorporated in a time domain simulation tool. This work has studied a direct method using the transient potential and kinetic energy of the critical machine only. In this paper the critical machine is identified by a fast and efficient method, and the proposal is new for using to get stability margins from hybrid approaches. Results from systems, like 16-machine, show stability indices to dynamic security assessment. © 2001 IEEE.