• Journal of Communications and Networks
• /
• v.8 no.1
• /
• pp.93-105
• /
• 2006

The domination of the Internet by TCP-based services has spawned many efforts to provide high network utilization with low loss and delay in a simple and scalable manner. Active queue management (AQM) algorithms attempt to achieve these goals by regulating queues at bottleneck links to provide useful feedback to TCP sources. While many AQM algorithms have been proposed, most suffer from instability, require careful configuration of nonintuitive control parameters, or are not practical because of slow response to dynamic traffic changes. In this paper, we propose a new AQM algorithm, hybrid random early detection (HRED), that combines the more effective elements of recent algorithms with a random early detection (RED) core. HRED maps instantaneous queue length to a drop probability, automatically adjusting the slope and intercept of the mapping function to account for changes in traffic load and to keep queue length within the desired operating range. We demonstrate that straightforward selection of HRED parameters results in stable operation under steady load and rapid adaptation to changes in load. Simulation and implementation tests confirm this stability, and indicate that overall performances of HRED are substantially better than those of earlier AQM algorithms. Finally, HRED control parameters provide several intuitive approaches to trading between required memory, queue stability, and response time.

• ETRI Journal
• /
• v.38 no.3
• /
• pp.487-493
• /
• 2016

Two new methods are proposed for an unsupervised adaptation of a language model (LM) with a single sentence for automatic transcription tasks. At the training phase, training documents are clustered by a method known as Latent Dirichlet allocation (LDA), and then a domain-specific LM is trained for each cluster. At the test phase, an adapted LM is presented as a linear mixture of the now trained domain-specific LMs. Unlike previous adaptation methods, the proposed methods fully utilize a trained LDA model for the estimation of weight values, which are then to be assigned to the now trained domain-specific LMs; therefore, the clustering and weight-estimation algorithms of the trained LDA model are reliable. For the continuous speech recognition benchmark tests, the proposed methods outperform other unsupervised LM adaptation methods based on latent semantic analysis, non-negative matrix factorization, and LDA with n-gram counting.

• MALSORI
• /
• no.58
• /
• pp.83-99
• /
• 2006

One key factor that hinders the widespread deployment of speaker identification technologies is the requirement of long enrollment utterances to guarantee low error rate during identification. To gain user acceptance of speaker identification technologies, adaptation algorithms that can enroll speakers with short utterances are highly essential. To this end, this paper applies MLLR speaker adaptation for speaker enrollment and compares its performance against other speaker modeling techniques: GMMs and HMM. Also, to speed up the computational procedure of identification, we apply speaker clustering method which uses principal component analysis (PCA) and weighted Euclidean distance as distance measurement. Experimental results show that MLLR adapted modeling method is most effective for short enrollment utterances and that the GMMs performs better when long utterances are available.

• Proceedings of the KIEE Conference
• /
• 1991.11a
• /
• pp.357-360
• /
• 1991

In this paper, we have introduced a network and showed how this can be realised as an adaptive equalizer. The walsh equlizer is built from a set of Walsh-Block pulse functions and LMS algorithms with the optimal convergence factor(C.F.). The convergence and the adaptation speed of this algorithms depends on the proper choice of a design factor $\mu$ called the C.F.. Conventional adaptation techniques use the fixed time constant C.F. by the method of trial and error. In this paper, we propose to adaptive C.F. which are optimally tailored to adapt C.F. in real time so that their values are kept optimum for a new set of input variables.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.41 no.8
• /
• pp.914-921
• /
• 1992

In this paper, we introduce a Walsh network and an LMS algorithm, and show how these can be realized as an adaptive equalizer. The Walsh network is built from a set of Walsh and Block pulse functions. In the LMS algorithm, the convergence factor is an important design parameter because it governs stability and convergence speed, which depend on the proper choice of the convergence facotr. The conventional adaptation techniques use a fixed time constant convergence factor by the method of trial and error. In this paper, we propose an optimal method in the choice of the convergence factor. The proposed algorithm depends on the received signal and the output of the Walsh network in real time.

• The Journal of the Acoustical Society of Korea
• /
• v.22 no.3
• /
• pp.188-193
• /
• 2003

Regardless of the distribution of the adaptation data in the testing environment, model-based adaptation methods that have so far been reported in various literature incorporates the adaptation data undiscriminatingly in reducing the mismatch between the training and testing environments. When the amount of data is small and the parameter tying is extensive, adaptation based on outlier data can be detrimental to the performance of the recognizer. The distribution of the adaptation data plays a critical role on the adaptation performance. In order to maximally improve the recognition rate in the testing environment using only a small number of adaptation data, supervised weighted training is applied to the structural maximum a posterior (SMAP) algorithm. We evaluate the performance of the proposed weighted SMAP (WSMAP) and SMAP on TIDIGITS corpus. The proposed WSMAP has been found to perform better for a small amount of data. The general idea of incorporating the distribution of the adaptation data is applicable to other adaptation algorithms.

• Journal of Auto-vehicle Safety Association
• /
• v.12 no.4
• /
• pp.13-22
• /
• 2020

This paper presents an adaptive feedback based actuator fault detection and tolerant control algorithms for longitudinal functional safety of autonomous driving. In order to ensure the functional safety of autonomous vehicles, fault detection and tolerant control algorithms are needed for sensors and actuators used for autonomous driving. In this study, adaptive feedback control algorithm to compute the longitudinal acceleration for autonomous driving has been developed based on relationship function using states. The relationship function has been designed using feedback gains and error states for adaptation rule design. The coefficients in the relationship function have been estimated using recursive least square with multiple forgetting factors. The MIT rule has been adopted to design the adaptation rule for feedback gains online. The stability analysis has been conducted based on Lyapunov direct method. The longitudinal acceleration computed by adaptive control algorithm has been compared to the actual acceleration for fault detection of actuators used for longitudinal autonomous driving.

• 제어로봇시스템학회:학술대회논문집
• /
• 1995.10a
• /
• pp.18-21
• /
• 1995

In this paper we describe a new method for multimodal function optimization using genetic algorithms(GAs). We propose adaptation rules for GA parameters such as population size, crossover probability and mutation probability. In the self organizing genetic algorithm(SOGA), SOGA parameters change according to the adaptation rules. Thus, we do not have to set the parameters manually. We discuss about SOGA and those of other approaches for adapting operator probabilities in GAs. The validity of the proposed algorithm will be verified in a simulation example of system identification.

• The Transactions of the Korean Institute of Power Electronics
• /
• v.4 no.1
• /
• pp.50-56
• /
• 1999

This paper presents consideration on validity of instantaneous correlation power theory. The proposed power theory is defined and analyzed by time domain approach, thus it is easy to understand and instrument. The power is decomposed into active, fundamental reactive and harmonics components based on the autocorrelation and crosscorrelation signal techniques between voltage and current waveforms. On the compensation property, active power filter deal with three components only. Also, for real time control of active power filter, the power models with difficult concept are not cost effective. To verify the validity of the instantaneous correlation power theory, experimental work for voltage type DSP based active power filter is achieved. The power of thyristor controlled motor drives is decomposed into three orthogonal components by proposed power theory. From compensation results, validity of proposed theory is confirmed. feedback controller needs the information on some motor parameters. New recursive adaptation algorithms for rotor resistance and mutual inductance which can be applied to our nonlinear feedback controller are also presented in this paper. The recursive adaptation algorithms make the estimated values of rotor resistance and mutual inductance track their real values. Some simulation and experimental results show that the adaptation algorithms are robust against the variation of stator resistance and stator inductance.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.5
• /
• pp.29-44
• /
• 2000

Adaptation of operator probabilities is one of the most important and promising issues in evolutionary computation areas. This is because the setting of appropriate probabilities is not only very tedious and difficult but very important to the performance improvement of genetic algorithms. Many researchers have introduced their algorithms for setting or adapting operator probabilities. Experimental results in most previous works, however, have not been satisfiable. Moreover, Tuson have insisted that “the adaptation is not necessarily a good thing” in his papers［$^1$$^2$］. In this paper, we propose a self-tuning scheme for adapting operator probabilities in genetic algorithms. Our scheme was extensively tested on four function optimization problems and one combinational problem； and compared to simple genetic algorithms with constant probabilities and adaptive genetic algorithm proposed by Srinivas et al［$^3$］. Experimental results showed that our scheme was superior to the others. Our scheme compared with previous works has three advantages： less computational efforts, co-evolution without additional operations for evolution of probabilities, and no need of additional parameters.