The cascade-correlation learning: a projection pursuit learningperspective

Cascade-correlation (Cascor) is a popular supervised learning architecture that dynamically grows layers of hidden neurons of fixed nonlinear activations (e.g., sigmoids), so that the network topology (size, depth) can be efficiently determined. Similar to a cascade-correlation learning network (CCLN), a projection pursuit learning network (PPLN) also dynamically grows the hidden neurons. Unlike a CCLN where cascaded connections from the existing hidden units to the new candidate hidden unit are required to establish high-order nonlinearity in approximating the residual error, a PPLN approximates the high-order nonlinearity by using trainable parametric or semi-parametric nonlinear smooth activations based on minimum mean squared error criterion. An analysis is provided to show that the maximum correlation training criterion used in a CCLN tends to produce hidden units that saturate and thus makes it more suitable for classification tasks instead of regression tasks as evidenced in the simulation results. It is also observed that this critical weakness in CCLN can also potentially carry over to classification tasks, such as the two-spiral benchmark used in the original CCLN paper.     

Design of low-error fixed-width multiplier for DSP applications

A low-error design of the fixed-width parallel multiplier for digital signal processing (DSF) applications is proposed. Applying two n bit inputs, it generates the n bit, instead of 2n bit, product with lower relative product errors, but uses only about half the area of a standard parallel multiplier. These features make it very suitable for use in many DSP applications such as arithmetic coding, wavelet transformation, digital filtering     

Saving fuel consumption and reducing pollution emissions for industrial furnace

The furnace efficiency can be maintained or improved slightly by lowering the excess air ratio and raising the air preheat temperature in the heat furnace. The results obtained with a plant-size industrial furnace show that when the excess air oxygen concentration is reduced from 4% to 3%, the furnace efficiency is raised by 0.6%. However, raising the temperature of the pre-heated air will cause a higher furnace temperature and a faster flowing speed of the air in the furnace. Thus, the heat in the radiation zone of the furnace can be evenly transmitted to shorten the time for the fuel to reach the igniting point, and reduce the heat loss from the hot air stream. This will reduce the emission of CO₂ and NO_x. If the temperature of the pre-heated air is raised from 200 °C to 240 °C, about 2.3 × 10⁶ m³ of natural gas can be saved annually to reduce 3.1 × 10³ t of CO₂ and 14.3% of nitrogen oxide emissions. Such operational adjustments of the furnace using the recovered FG will save energy, reduce CO₂ emission and alleviate the adverse impact to environment.     

Optimization of gallium-doped ZnO thin films grown using Grey–Taguchi technique

In this paper, optimization of the process parameters considering multiple performance characteristics to prepare the transparent conducting gallium-doped zinc oxide thin films with radio frequency (RF) magnetron sputtering was investigated. Experiments based on the Grey–Taguchi technique were conducted to examine the effect of deposition parameters including RF power, process pressure, substrates temperature and process time, aiming to obtain highly transparent and conductive films. Comparing with the optimal parameter set selected from orthogonal array by Taguchi method, the optimal grey theory prediction design (GTPD) can receive an improvement of 5.75?% in electrical resistivity and 1.47?% in optical transmittance. Further refinements respectively to RF power and process pressure with fixing other parameters level in GTPD were explored. The results show the alteration on RF power and process pressure in the GTPD can receive over 31?% and 51?% of improvement in electrical resistivity, respectively, with keeping the visible range optical transmittance over?85?%.     

The intention and willingness to pay moving violation citations among Taiwan motorcyclists

This study uses three-level scenario design and Spike model constructs to investigate the risk premium Taiwan motorcycle operators are willing to pay for moving violations. The primary focus of the investigation is on four types of moving violations including speeding, running red lights, right turn on red violations, and drunk driving. In the model, these four types of violations influence the willingness to pay a risk premium. The results show that, in addition to increasing enforcement, raising the level of fines is one of the most effective methods to influence levels of compliance. Estimated results through the Spike model show that speeders will accept a risk premium of NT$740 (US$1 = NT$30), while motorcycle operators who run red lights will accept a risk premium of NT$1100, motorcycle operators who turn right on red will accept a risk premium of NT$367, and motorcyclists who drive drunk will accept a risk premium of NT$18,540. This indicates that current fines in Taiwan could be raised.     

An LDPC Decoder Chip Based on Self-Routing Network for IEEE 802.16e Applications

An LDPC decoder chip fully compliant to IEEE 802.16e applications is presented. Since the parity check matrix can be decomposed into sub-matrices which are either a zero-matrix or a cyclic shifted matrix, a phase-overlapping message passing scheme is applied to update messages immediately, leading to enhance decoding throughput. With only one shifter-based permutation structure, a self-routing switch network is proposed to merge 19 different sub-matrix sizes as defined in IEEE 802.16e and enable parallel message to be routed without congestion. Fabricated in the 90 nm 1P9M CMOS process, this chip achieves 105 Mb/s at 20 iterations while decoding the rate-5/6 2304-bit code at 150 MHz operation frequency. To meet the maximum data rate in IEEE 802.16e, this chip operates at 109 MHz frequency and dissipates 186 mW at 1.0 V supply.     

Antigenic drift between the haemagglutinin of the Hong Kong influenza strains A/Aichi/2/68 and A/Victoria/3/75

A DNA copy of the gene coding for the influenza A/Aichi/2/68 haemagglutinin protein was cloned in the plasmid pBR322 and the complete nucleotide sequence determined. Comparison of this primary structure and the deduced amino acid sequence with the haemagglutinin gene and protein of strains belonging to the same (H3) subtype and to different subtypes, of both human (H2) and avian (Hav1) origin, documents further at the molecular level the two independent modes of antigenic variation of the virus--drift and shift.     

Adaptive Control of the Reverse Link in cdma2000

This paper looks at a way to adaptively operate the cdma2000 reverse link (RL) using velocity estimation. A simple analysis of the power control process is carried out to better understand the impact of power control on the efficacy of the RL operation, both at the link and system levels. The structure and performance of the RL is presented, and different areas where the velocity estimation could be used are identified, e.g., turning off power control selectively, adjusting traffic-to-pilot ratios and controlling the transmit level of pilots for mobiles in dormant mode. The algorithm for doing the velocity estimation and its performance concludes the paper.     

Diode grids for electronic beam steering and frequency multiplication

Loading a grid with diodes offers the possibility of two-dimensional control of millimeter waves that is analogous to holography and nonlinear optics. These grids are attractive because they are suitable for monolithic integration with gallium-arsenide Schottky diodes and for high-power operation. Here we present grid designs for electronic beam-steering and harmonic generation. The beam-steering grid is a programmable reflector, where the diode bias controls the phase shift of the reflection. The variation of the phase across the grating sets the direction of the reflected beam. The reflection loss in computer simulations is 3dB at 90GHz. The harmonic-generating grid acts as a nonlinear reactive surface, where the nonlinear capacitance of the diodes produces the harmonic frequencies. Quasioptical filters select the desired harmonic. Computer simulations predict that a 65GHz-to-130GHz doubler would have an output power of 0.56W/cm² and a conversion efficiency of 35%.