Neural Network Exploration Using Optimal Experiment Design

We consider the question "How should one act when the only goal is to learn as much as possible?" Building on the theoretical results of Fedorov [1972] and MacKay [1992], we apply techniques from Optimal Experiment Design (OED) to guide the query/action selection of a neural network learner. We demonstrate that these techniques allow the learner to minimize its generalization error by exploring its domain efficiently and completely. We conclude that, while not a panacea, OED-based query/action has much to offer, especially in domains where its high computational costs can be tolerated.

Jointly Optimal MMSE design for non-redundant FIR precoding and equalization

We provide a filterbank precoding framework (FBP) for frequency selective channels using the minimum mean squared error (MMSE) criterion. The design obviates the need for introducing a guard interval between successive blocks, and hence can achieve the maximum possible bandwidth efficiency. This is especially useful in cases where the channel is of a high order. We treat both the presence and the absence of channel knowledge at the transmitter. In the former case, we obtain the jointly optimal precoder-equalizer pair of the specified order. In the latter case, we use a zero padding precoder, and obtain the MMSE equalizer. No restriction on the dimension or nature of the channel matrix is imposed. Simulation results indicate that the filterbank approach outperforms block based methods like OFDM and eigenmode precoding.

Multi-hop Cooperative Wireless Networks: Diversity Multiplexing Tradeoff and Optimal Code Design

We consider single-source single-sink (ss-ss) multi-hop relay networks, with slow-fading links and single-antenna half-duplex relay nodes. While two-hop cooperative relay networks have been studied in great detail in terms of the diversity-multiplexing tradeoff (DMT), few results are available for more general networks. In this paper, we identify two families of networks that are multi-hop generalizations of the two-hop network: K-Parallel-Path (KPP)networks and layered networks.KPP networks, can be viewed as the union of K node-disjoint parallel relaying paths, each of length greater than one. KPP networks are then generalized to KPP(I) networks, which permit interference between paths and to KPP(D) networks, which possess a direct link from source to sink. We characterize the DMT of these families of networks completely for K > 3. Layered networks are networks comprising of layers of relays with edges existing only between adjacent layers, with more than one relay in each layer. We prove that a linear DMT between the maximum diversity dmax and the maximum multiplexing gain of 1 is achievable for single-antenna fully-connected layered networks. This is shown to be equal to the optimal DMT if the number of relaying layers is less than 4.For multiple-antenna KPP and layered networks, we provide an achievable DMT, which is significantly better than known lower bounds for half duplex networks.For arbitrary multi-terminal wireless networks with multiple source-sink pairs, the maximum achievable diversity is shown to be equal to the min-cut between the corresponding source and the sink, irrespective of whether the network has half-duplex or full-duplex relays. For arbitrary ss-ss single-antenna directed acyclic networks with full-duplex relays, we prove that a linear tradeoff between maximum diversity and maximum multiplexing gain is achievable.Along the way, we derive the optimal DMT of a generalized parallel channel and derive lower bounds for the DMT of triangular channel matrices, which are useful in DMT computation of various protocols. We also give alternative and often simpler proofs of several existing results and show that codes achieving full diversity on a MIMO Rayleigh fading channel achieve full diversity on arbitrary fading channels. All protocols in this paper are explicit and use only amplify-and-forward (AF) relaying. We also construct codes with short block-lengths based on cyclic division algebras that achieve the optimal DMT for all the proposed schemes.Two key implications of the results in the paper are that the half-duplex constraint does not entail any rate loss for a large class of cooperative networks and that simple AF protocols are often sufficient to attain the optimal DMT

Optimization process of polyester polymer mortars modified with recycled GFRP waste aggregates – application of factorial experiment design-

Glass fibre-reinforced plastics (GFRP), nowadays commonly used in the construction, transportation and automobile sectors, have been considered inherently difficult to recycle due to both: cross-linked nature of thermoset resins, which cannot be remolded, and complex composition of the composite itself, which includes glass fibres, matrix and different types of inorganic fillers. Presently, most of the GFRP waste is landfilled leading to negative environmental impacts and supplementary added costs. With an increasing awareness of environmental matters and the subsequent desire to save resources, recycling would convert an expensive waste disposal into a profitable reusable material. There are several methods to recycle GFR thermostable materials: (a) incineration, with partial energy recovery due to the heat generated during organic part combustion; (b) thermal and/or chemical recycling, such as solvolysis, pyrolisis and similar thermal decomposition processes, with glass fibre recovering; and (c) mechanical recycling or size reduction, in which the material is subjected to a milling process in order to obtain a specific grain size that makes the material suitable as reinforcement in new formulations. This last method has important advantages over the previous ones: there is no atmospheric pollution by gas emission, a much simpler equipment is required as compared with ovens necessary for thermal recycling processes, and does not require the use of chemical solvents with subsequent environmental impacts. In this study the effect of incorporation of recycled GFRP waste materials, obtained by means of milling processes, on mechanical behavior of polyester polymer mortars was assessed. For this purpose, different contents of recycled GFRP waste materials, with distinct size gradings, were incorporated into polyester polymer mortars as sand aggregates and filler replacements. The effect of GFRP waste treatment with silane coupling agent was also assessed. Design of experiments and data treatment were accomplish by means of factorial design and analysis of variance ANOVA. The use of factorial experiment design, instead of the one factor at-a-time method is efficient at allowing the evaluation of the effects and possible interactions of the different material factors involved. Experimental results were promising toward the recyclability of GFRP waste materials as polymer mortar aggregates, without significant loss of mechanical properties with regard to non-modified polymer mortars.

Optimal inverter design for an induction machine using resonant frequency matching

The frequency responses of two 50 Hz and one 400 Hz induction machines have been measured experimentally over a frequency range of 1 kHz to 400 kHz. This study has shown that the stator impedances of the machines behave in a similar manner to a parallel resonant circuit, and hence have a resonant point at which the Input impedance of the machine is at a maximum. This maximum impedance point was found experimentally to be as low as 33 kHz, which is well within the switching frequency ranges of modern inverter drives. This paper investigates the possibility of exploiting the maximum impedance point of the machine, by taking it into consideration when designing an inverter, in order to minimize ripple currents due to the switching frequency. Minimization of the ripple currents would reduce torque pulsation and losses, increasing overall performance. A modified machine model was developed to take into account the resonant point, and this model was then simulated with an inverter to demonstrate the possible advantages of matching the inverter switching frequency to the resonant point. Finally, in order to experimentally verify the simulated results, a real inverter with a variable switching frequency was used to drive an induction machine. Experimental results are presented.

On Nonlinear Dynamics and an Optimal Control Design to a Longitudinal Flight

In this work, we analyzed a bifurcational behavior of a longitudinal flight nonlinear dynamics, taking as an example the F-8 aircraft Crusader. We deal with an analysis of high angles of attack in order to stabilize the oscillations; those were close to the critical angle of the aircraft, in the flight conditions, established. We proposed a linear optimal control design applied to the considered nonlinear aircraft model below angle of stall, taking into account regions of Hopf and saddled noddle bifurcations.

On an optimal control design for Rossler system

In this Letter, an optimal control strategy that directs the chaotic motion of the Rossler system to any desired fixed point is proposed. The chaos control problem is then formulated as being an infinite horizon optimal control nonlinear problem that was reduced to a solution of the associated Hamilton-Jacobi-Bellman equation. We obtained its solution among the correspondent Lyapunov functions of the considered dynamical system. (C) 2004 Elsevier B.V All rights reserved.

Bayesian L-optimal exact design of experiments for biological kinetic models

Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)

Differentiating the Multipoint Expected Improvement for Optimal Batch Design

This work deals with parallel optimization of expensive objective functions which are modelled as sample realizations of Gaussian processes. The study is formalized as a Bayesian optimization problem, or continuous multi-armed bandit problem, where a batch of q > 0 arms is pulled in parallel at each iteration. Several algorithms have been developed for choosing batches by trading off exploitation and exploration. As of today, the maximum Expected Improvement (EI) and Upper Confidence Bound (UCB) selection rules appear as the most prominent approaches for batch selection. Here, we build upon recent work on the multipoint Expected Improvement criterion, for which an analytic expansion relying on Tallis’ formula was recently established. The computational burden of this selection rule being still an issue in application, we derive a closed-form expression for the gradient of the multipoint Expected Improvement, which aims at facilitating its maximization using gradient-based ascent algorithms. Substantial computational savings are shown in application. In addition, our algorithms are tested numerically and compared to state-of-the-art UCB-based batchsequential algorithms. Combining starting designs relying on UCB with gradient-based EI local optimization finally appears as a sound option for batch design in distributed Gaussian Process optimization.