• Journal of IKEEE
• /
• v.22 no.4
• /
• pp.1238-1241
• /
• 2018

In this paper, MNIST classification CNN with zero skipping is implemented. Activation of CNN results in 30% to 40% zero. Since 0 does not affect the MAC operation, skipping 0 through a branch can improve performance. However, at the convolution layer, skipping over a branch causes a performance degradation. Accordingly, in the convolution layer, an operation is skipped by giving a NOP that does not affect the operation. Fully connected layer is skipped through the branch. We have seen performance improvements of about 1.5 times that of existing CNN.

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.12 no.3
• /
• pp.30-36
• /
• 1975

Distortions are analyzed and compared for two TDM channel expansion methods- periodic sample skipping and sampling frequency reduction. Signal is assumed to be stationary random signal with zero.mean. Channel noise and interference are not considered in the analysis. For speech signal, it is shown that the periodic sample skipping method could be a better choice under practical design constraints.

• Proceedings of the IEEK Conference
• /
• 2007.07a
• /
• pp.323-324
• /
• 2007

In this paper, we present the method of skipping the quantization for fast video encoding. Based on the theoretical analyzes for the integer transform and quantization in H.264 video coder, we can derive a sufficient threshold under which each quantized coefficient becomes zero. In addition, in order to reduce the complexity of the fast video encoding, complexity is improved, leading to improvement of total encoding time saving for given threshold. The simulation results show the capability of the proposed algorithm.

• Journal of Nutrition and Health
• /
• v.42 no.3
• /
• pp.256-267
• /
• 2009

This study was conducted to investigate nutritional knowledge related to breakfast and dietary behavior of elementary school students by gender and skipping breakfast. The subjects of this study were the 6th grade elementary school students (Total 237; 119 males and 118 females) in Daejeon. The group who ate breakfast six to seven times per week was 58.6%; The group who ate breakfast zero to five times per week was 41.4%. Most of subjects' families (84.4%) were 'nuclear family' and 63.8% of mothers held jobs. There was a significant difference between family income and skipping breakfast. As for the type of breakfast, 79.0% of the male students and 81.4% of female students answered 'cooked rice' and the reasons for skipping breakfast were 'no appetite' followed by 'not enough time', 'over sleeping'. In the eating breakfast group, frequency of having breakfast with family was higher compared to the skipping breakfast group. Total scores of nutritional knowledge related to breakfast in the female students were significantly higher than those of the male students. Also significant differences were found on such questions as 'the relation between eating breakfast and body weight control' and 'importance of breakfast' by skipping breakfast. The dietary behavior scores of the group who ate breakfast were higher than those of the group who skipped breakfast. Therefore, a need exists to develop and distribute a simple breakfast menu that contains Korean-style food and can be more appetizing and appealing to Korean elementary school students. Also it is necessary to develop a systemic awareness program that emphasize the importance of breakfast and the harmfulness of skipping breakfast.

• Journal of Broadcast Engineering
• /
• v.21 no.3
• /
• pp.369-379
• /
• 2016

This paper deals with an efficient method for skipping inter-channel phase differences (IPD) in the MPEG surround of the unified speech and audio coding (USAC). Based on the psycho-acoustic sensitivity on the IPD, we estimate a threshold on IPD, below which we can not notice degradation in spatial cue. We propose an IPD skip method, in which any IPDs within the threshold are set to zero and are not transmitted. The proposed IPD skip method gives about 38% savings in terms of bit amount for IPD. Nevertheless, in the MUSHRA test, the proposed method does not show any noticeable degradation in the decoded audio quality.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.7
• /
• pp.3455-3474
• /
• 2018

As a unique form of signed-digit representation, non-adjacent form (NAF) minimizes Hamming weight by removing a stream of non-zero bits from the binary representation of positive integer. Thanks to this strong point, NAF has been used in various applications such as cryptography, packet filtering and so on. In this paper, to improve the NAF conversion speed of the $NAF_w$ algorithm, we propose a new NAF conversion algorithm, called w-bit Shifting Non-Adjacent Form($SNAF_w$), where w is width of scanning window. By skipping some unnecessary bit comparisons, the proposed algorithm improves the NAF conversion speed of the $NAF_w$ algorithm. To verify the excellence of the $SNAF_w$ algorithm, the $NAF_w$ algorithm and the $SNAF_w$ algorithm are implemented in the 8-bit microprocessor ATmega128. By measuring CPU cycle counter for the NAF conversion under various input patterns, we show that the $SNAF_2$ algorithm not only increases the NAF conversion speed by 24% on average but also reduces deviation in the NAF conversion time for each input pattern by 36%, compared to the $NAF_2$ algorithm. In addition, we show that $SNAF_w$ algorithm is always faster than $NAF_w$ algorithm, regardless of the size of w.

• Journal of Broadcast Engineering
• /
• v.17 no.5
• /
• pp.725-733
• /
• 2012

MPEG and VCEG have constituted a collaboration team called JCT-VC(Joint Collaborative Team on Video Coding) and have been developing the HEVC(High Efficiency Video Coding) standard. The next generation video coding standard HEVC shows higher compression rate compared with the H.264/AVC standard, but the encoder computational complexity of the HEVC encoder is significantly high. In order to reduce this computational complexity in the HEVC encoder, a fast prediction unit decision is proposed. The proposed fast prediction unit decision method reduces the encoder complexity by skipping the remaining prediction units if the current prediction unit does not have any non-zero quantized transform coefficient. The proposed method reduces the encoder computational complexity by 50.3% comparing with HM6.0 but it maintains the same level of coding efficiency.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.44 no.4 s.316
• /
• pp.51-59
• /
• 2007

In this paper, we propose an efficient hardware architecture for H.264/AVC CAVLC (Context-based Adaptive Variable Length Coding) decoding which used for baseline profile and extended profile. Previous CAVLC architectures are consisted of five step block and each block gets effective bits from Controller block and Accumulator. If large number of non-zero coefficients exist, process for getting effective bits has to iterates many times. In order to reduce this unnecessary process, we propose two techniques, which combine five steps into four steps and reduce process to get efficiency bit by skipping addition step. By adopting these two techniques, the required processing time was reduced about 26% compared with previous architectures. It was designed in a hardware description language and total logic gate count was 16.83k using 0.18um standard cell library.

• The Korean Journal of Food And Nutrition
• /
• v.28 no.1
• /
• pp.47-59
• /
• 2015

In this study, we investigated the dietary habits, snacks, and self-purchasing snacks (SPS) intake behaviors of 519 elementary school students (boys=239, girls=280). Obesity was significantly higher (p<0.05) in boys (24.8%) than in girls (14.7%) and the proportion of underweight subjects was higher compared to normal or other weight groups for both the boys and the girls. There were 7.5% of the subjects in the group that always skipped breakfast, and the main reason of skipping breakfast was insufficient time (51.9%). The snack intake frequency was once or twice per week for 23.1% of the subjects and three or four times per week for 25.1%. The SPS intake frequency was the subjects zero for 35.6% of the highest level of the responders, while 6.8% of the respondents took more than once SPS per day. 59.6% of the respondents consumed SPS due to hunger while 15.0% consumed SPS out of boredom. The SPS was purchased from supermarkets in 34.5% of the cases, from convenience stores in 24.1% and from snack corners in 20.0% of cases or from a store near school in 14.5% of the cases. Analysis of SPS behaviors according to obesity index showed that parent's opinion of 'permission to buy SPS as needed' had a significant effect in 64.5% over weight subjects compared to only 53.7% in underweight groups. The subjects who used more than 3/4 of their pocket money to buy SPS was higher in the overweight groups (16.4%) than in the underweight groups (7.0%) and normal weight groups (9.8%). The favorite snacks and SPS were milk and yogurt for 45.7% of the subjects, fruits for 42.7%, ice cream for 26.4%, fruit juices for 23.8%, sweet stuff for 16.4%, frozen dessert for 8.9%, and chocolate or candy for 8.1% in descending order. The intake frequency of milk, yogurt, and fruit juices was higher in the underweight groups, but the intake frequency of sweet stuff, frozen dessert, and chocolate or candy was higher in over weight groups. The intake of frozen dessert was more than four times higher in the overweight groups than in the underweight groups. In conclusion, dietary habits, snacks, and SPS intake behaviors were similar between the boys and the girls and obesity groups, but most students appeared to have a high preference for intake snacks and SPS. Therefore, education for appropriate snacks intake habits will be beneficial for improving their dietary habits and health.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2011.10a
• /
• pp.431-435
• /
• 2011

This paper presents optimized searching technique to improve the performance of H.264/AVC. The proposed CAVLC encoder uses forward and backward searching algorithm to compute the parameters. By zero-block skipping technique and pipelined scheduling, the proposed CAVLC encoder can obtain better performance. The experimental result shows that the proposed architecture needs only 66.6 cycles on average for each $16{\times}16$ macroblock encoding. The proposed architecture improves the performance by 13.8% than that of previous designs. The proposed CAVLC encoder was implemented using VerilogHDL and synthesized with Megnachip $0.18{\mu}m$ standard cell library. The synthesis result shows that the gate count is about 15.6K with 125Mhz clock frequency.