• Journal of the Korea Society of Computer and Information
• /
• v.29 no.7
• /
• pp.11-20
• /
• 2024

In this paper, we propose the BAG (Bit-width Aware Generator) and the Intermediate Layer Knowledge Distillation using Channel-wise Attention to reduce the knowledge gap between a quantized network, a full-precision network, and a generator in GDFQ (Generative Data-Free Quantization). Since the generator in GDFQ is only trained by the feedback from the full-precision network, the gap resulting in decreased capability due to low bit-width of the quantized network has no effect on training the generator. To alleviate this problem, BAG is quantized with same bit-width of the quantized network, and it can generate synthetic images, which are effectively used for training the quantized network. Typically, the knowledge gap between the quantized network and the full-precision network is also important. To resolve this, we compute channel-wise attention of outputs of convolutional layers, and minimize the loss function as the distance of them. As the result, the quantized network can learn which channels to focus on more from mimicking the full-precision network. To prove the efficiency of proposed methods, we quantize the network trained on CIFAR-100 with 3 bit-width weights and activations, and train it and the generator with our method. As the result, we achieve 56.14% Top-1 Accuracy and increase 3.4% higher accuracy compared to our baseline AdaDFQ.

• Journal of the Korean Chemical Society
• /
• v.19 no.5
• /
• pp.304-316
• /
• 1975

A formal theory to explain the Stark effect in molecular exciton states is developed using the second-quantization formalism. In this theory not only the Stark effect but also the Davydov effect are explicitly taken into consideration since the observed spectral splitting in the UV spectra for molecular crystals with two or more molecules per unit cell may be the result of combination of the above two effects. Especially for molecular crystals containing two molecules in a unit cell the splitting is shown to be hyperbolically dependent upon the strength of an externally applied, uniform electric field, from which informations regarding the excited state dipole moments of a single molecule may be obtained.

• The Transactions of the Korean Institute of Electrical Engineers C
• /
• v.51 no.9
• /
• pp.397-407
• /
• 2002

A new simulator which predicts the quantum effect in NMOSFET structure is developed. Using the self-consistent method by numerical method, this simulator accurately predicts the carrier distribution due to improved calculation precision of potential in the inversion layer. However, previous simulator uses analytical potential distribution or analytic function based fitting parameter Using the developed simulator, threshold voltage increment and gate capacitance reduction due to the quantum effect are analyzed in NMOS. Especially, as oxide thickness and channel doping dependence of quantum effect is analyzed, and the property analysis for the next generation device is carried out.

• 전자공학회논문지 IE
• /
• v.47 no.3
• /
• pp.35-40
• /
• 2010

In this paper, we propose an algorithm to improve the precision of initial carrier-frequency offset estimation for multiband-OFDM (MB-OFDM) UWB system which is considering the quantization-noise effect. In the general OFDM system, the two adjacent and repeated preamble symbols are used for the initial carrier-frequency synchronization while the performance of the frequency-offset estimation is bounded by quantization effect generated from analog-to-digital conversion at the receiver. This paper proposes a method in which one-symbol interval between two adjacent preamble symbols for the initial frequency synchronization is extended to multiple-symbol interval between non-adjacent symbols in an extent that phase ambiguity does not occur. In this paper, we also present '6' as optimal multiple symbol interval for the MB-OFDM system with 30 preamble symbols on 3-band hopping and with 4-bit A/D conversion at the receiver. Under the channel environments for the MB-OFDM UWB system, the simulation results show that the proposed estimation algorithm can achieve the initial estimation in offset precision less than 5 ppm.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.9 no.11
• /
• pp.1221-1226
• /
• 2014

In an effort to improve the efficiency of the speech recognition, we investigate the effect of the number of phoneme clusters. For this purpose, codebooks of varied number of phoneme clusters are prepared by modified k-means clustering algorithm. The subsequent processing is fuzzy vector quantization (FVQ) and hidden Markov model (HMM) for speech recognition test. The result shows that there are two distinct regimes. For large number of phoneme clusters, the recognition performance is roughly independent of it. For small number of phoneme clusters, however, the recognition error rate increases nonlinearly as it is decreased. From numerical calculation, it is found that this nonlinear regime might be modeled by a power law function. The result also shows that about 166 phoneme clusters would be the optimal number for recognition of 300 isolated words. This amounts to roughly 3 variations per phoneme.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.4
• /
• pp.772-780
• /
• 1994

Vector quantization with two-channel conjugate codebook has been researched as an efficient coding technique that can reduce the computational complexity and codebook storage. This paper proposes FSVQ using two-channel conjugate codebook in order to reduce the number of state codebooks. Input vector in the two-channel conjugate FSVQ is coded with state codebook of a seperated state according to each codebook. In addition, LOT is adopted to obtain to obtain a high coding gain and to reduce blocking effect which appears in the block coding. As a result, although FSVQ can achieve higher data compression ratio than general vector quantization, it has a disadvantage of having a very large number of state codebooks. However FSVQ with two-channel conjugate codebooks can employ a significantly reduced number of state codebooks, even though it has a small loss in the PSNR compared with the conventional FSVQ using one codebook. Moreover FSVQ in the LOT domain can reduce blocking effect and high coding gain compared with FSVQ in the spatial domain.

• Journal of Convergence for Information Technology
• /
• v.11 no.9
• /
• pp.124-129
• /
• 2021

A time-interpolation technique has been applied to the conventional FLASH analog-to-digital converter (ADC) to increase a number of quantization level, thus it reduces not only a power dissipation, but also minimize an active chip area. In this work, we demonstrated 5-bit ADC which has 31 quantization levels consisting of 16 conventional voltage-mode comparators and 15 time-mode comparators. As a result, we have achieved about 48.4% voltage-mode comparator reductions. The ADC is fabricated in a 14nm fin Field-effect transistor (FinFET) process with an active die area of 0.0024 mm2 while consuming 0.82 mW through a 0.8 V supply. At 400-MS/s conversion rate, the ADC performs 28.03 dB SNDR (4.36 ENOB) at 21MHz input frequency.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.41 no.8
• /
• pp.17-24
• /
• 2004

Quantum effects in the poly-gate are analyzed in two dimensions using the density-gradient method, and their impact on the short-channel effect of double-gate MOSFETs is investigated. The 2-D effects of quantum mechanical depletion at the gate to sidewall oxide is identified as the cause of large charge-dipole formation at the corner of the gate. The bias dependence of the charge dipole shows that the magnitude of the dipole peak-value increases in the subthreshold region and there is a large difference in carrier and potential distribution compared to the classical solution. Using evanescent-nude analysis, it is found that the quantum effect in the poly-gate substantially increases the short-channel effect and it is more significant than the quantum effect in the Si film. The penetration of potential contours into the poly-gate due to the dipole formation at the drain side of the gate corner is identified as the reason for the substantial increase in short-channel effects.

• Phonetics and Speech Sciences
• /
• v.1 no.2
• /
• pp.3-8
• /
• 2009

In an effort to improve speech recognition, we investigated the effect of the number of clusters. In usual LBG clustering, the number of codebook clusters is doubled on each bifurcation and hence cannot be chosen arbitrarily in a natural way. To have the number of clusters at our control, we combined adaptive resonance theory (ART2) with LBG and perform the clustering in two stages. The codebook thus formed was used in subsequent processing of fuzzy vector quantization (FVQ) and HMM for speech recognition tests. Compared to conventional LBG, our method was shown to reduce the best recognition error rate by 0${\sim$}0.9% depending on the vocabulary size. The result also showed that between 400 and 800 would be the optimal number of clusters in the limit of small and large vocabulary speech recognitions of isolated words, respectively.

• 한국정보디스플레이학회:학술대회논문집
• /
• 2005.07b
• /
• pp.863-864
• /
• 2005

In recent years, we use overdriving scheme to improve the response time of the liquid crystal. Since conventional overdriving scheme uses memory to perform ideal processing, it is desired to reduce system cost by decreasing the data stored in these SDRAMs. As a general compression method, quantization, sub-sampling and Block Truncation Coding ( BTC ) are used, which process data in block base and cause block effect. So we proposed new data compression method by color space conversion. Because this method compresses luminance and chrominance signal by different ratio, it can efficiently reduce error of block effect in decompression image.