김진훈 교수

A new variable step‐size sign subband adaptive filter (VSS‐SSAF), not requiring any a priori information, is proposed by minimising the ℓ 1 ‐norm of the subband a posteriori error under a box‐constraint on the step‐size. In addition, an efficient numerical procedure for updating its VSS is introduced to solve the non‐differentiable convex problem. The proposed algorithm provides comparable or better convergence performance, when compared with the recent VSS‐SSAF (minimising the mean‐squares devi

A closed‐form design of a two‐dimensional (2D) finite impulse response (FIR) filter with circular, rectangular, fan or quadrant‐fan shapes in the passband region is presented. The computationally efficient 2D FIR filter can be implemented using frequency transformation and sampling‐kernel‐based interpolation instead of an optimisation algorithm. When compared with the existing method, the proposed method reduces 90% of multiplications for filtering. Several design examples are demonstrated to ve

For continuous state space applications, a novel method of Q-learning is proposed, where the method incorporates a region-based reward assignment being used to solve a structural credit assignment problem and a convex clustering approach to find a region with the same reward attribution property. Our learning method can estimate a current Q-value of an arbitrarily given state by using effect functions, and has the ability to learn its actions similar to that of Q-learning. Thus, our method enabl

Neurons in the brain are complex machines with distinct functional compartments that interact nonlinearly. In contrast, neurons in artificial neural networks abstract away this complexity, typically down to a scalar activation function of a weighted sum of inputs. Here we emulate more biologically realistic neurons by learning canonical activation functions with two input arguments, analogous to basal and apical dendrites. We use a network-in-network architecture where each neuron is modeled as

In multi-channel active noise control (ANC) systems, online secondary-path modeling (OSPM) using the auxiliary noise signal is often applied. However, the additive noise signal may contribute to the residual output noise. In this paper, the conventional noise power scheduling, utilized for single-channel ANC with OSPM, is further extended to multi-channel ANC. Simulation results demonstrate that the proposed approach yields better ANC performance, compared with conventional multi-channel ANC met

Abstract In this study, Cu-based bulk metallic glass coatings were deposited by atmospheric plasma spraying with different hydrogen flow rates. The crystallization and oxidation of the coatings is assessed along with corrosion resistance. As thermal energy in the plasma jet increases, the melting fraction and oxidation of particles in the coating increases as does porosity. All of these factors have an effect on the corrosion resistance of Cu-based bulk metallic glass coatings and their relative

A monolayer of L1 <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> FePt nanoparticles was fabricated on a polymer film by introducing a Au seed layer prior to the Fe-Pt deposition on the polyimide film. The particle size distribution was tightened from 5.1plusmn1.8 nm to 3.8plusmn1.1 nm by introducing the Au seeds to induce a preferential nucleation of the deposited metal on the preexisting Au particles. Deposition and annealing was repeated