• IEMEK Journal of Embedded Systems and Applications
• /
• v.15 no.5
• /
• pp.235-242
• /
• 2020

This paper analyzes the effect of reduced output width of the truncated logarithmic multiplication and application to inferences using convolutional neural networks (CNNs). For small hardware overhead, output width is reduced in the truncated Mitchell multiplier, so that fractional bits in multiplication output are minimized in error-resilient applications. This analysis shows that when reducing output width in the truncated Mitchell multiplier, even though worst-case relative error increases, average relative error can be kept small. When adopting 8 fractional bits in multiplication output in the evaluations, there is no significant performance degradation in target CNNs compared to existing exact and original Mitchell multipliers.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.11 no.2
• /
• pp.59-65
• /
• 2001

3GPP proposed a variant CBC-MAC based on the block cipher KASUMI to provide the data integrity over a radio access link. We have studied deeply the Knudsen and Mitchell\`s attack. In this paper we proposed a definite performing algorithm of the Knudsen and Mitchell\`s alack and compute the success probability and complexity of that algorithm. Moreover We also analyze a security of 3GPP-MAC comparing with the original CBC-MAC.

• ETRI Journal
• /
• v.43 no.4
• /
• pp.684-693
• /
• 2021

This paper presents a low-cost two-stage approximate multiplier for bfloat16 (brain floating-point) data processing. For cost-efficient approximate multiplication, the first stage implements Mitchell's algorithm that performs the approximate multiplication using only two adders. The second stage adopts the exact multiplication to compensate for the error from the first stage by multiplying error terms and adding its truncated result to the final output. In our design, the low-cost multiplications in both stages can reduce hardware costs significantly and provide low relative errors by compensating for the error from the first stage. We apply our approximate multiplier to the convolutional neural network (CNN) inferences, which shows small accuracy drops with well-known pre-trained models for the ImageNet database. Therefore, our design allows low-cost CNN inference systems with high test accuracy.

• Language and Information
• /
• v.10 no.1
• /
• pp.21-46
• /
• 2006

This study proposes a Word Sense Disambiguation (WSD) algorithm, based on concept learning with special emphasis on statistically meaningful lowest frequency words. Previous works on WSD typically make use of frequency of collocation and its probability. Such probability based WSD approaches tend to ignore the lowest frequency words which could be meaningful in the context. In this paper, we show an algorithm to extract and make use of the meaningful lowest frequency words in WSD. Learning method is adopted from the Find-Specific algorithm of Mitchell (1997), according to which the search proceeds from the specific predefined hypothetical spaces to the general ones. In our model, this algorithm is used to find contexts with the most specific classifiers and then moves to the more general ones. We build up small seed data and apply those data to the relatively large test data. Following the algorithm in Yarowsky (1995), the classified test data are exhaustively included in the seed data, thus expanding the seed data. However, this might result in lots of noise in the seed data. Thus we introduce the 'maximum a posterior hypothesis' based on the Bayes' assumption to validate the noise status of the new seed data. We use the Naive Bayes Classifier and prove that the application of Find-Specific algorithm enhances the correctness of WSD.

• The Korean Journal of Applied Statistics
• /
• v.7 no.1
• /
• pp.83-102
• /
• 1994

Some optimal designs and data analysis methods based on a Gaussian spatial linear model for computer simulation experiments are considered. For designs of computer experiments, Latin-hypercube designs and some optimal designs are combined. A two-stage computational (2-points exchange and Newton-type) algorithm for finding the optimal Latin-hypercube design is presented. The spatial prediction model which was discussed by Sacks, Welch, Mitchell and Wynn(1989) for computer experiments, is used for analysis of the simulated data. Moreover, a method of contructing sequential (optimal) Latin-hypercube designs is considered. An application of this approach to the quality improvement and optimization of the integrated circuit design via the main-effects plot and the sequential experimental strategy is presented.

• Journal of Environmental Science International
• /
• v.10 no.6
• /
• pp.431-436
• /
• 2001

Variations in phytoplankton concentrations result from changes of the ocean color caused by phytoplankton pigments. Thus, ocean spectral reflectance for low chlorophyll waters are blue and high chlorophyll waters tend to have green reflectance. In the Korea region, clear waters and the open sea in the Kuroshio regions of the East China Sea have low chlorophyll. As one moves even closer In the northwestern part of the East China Sea, the situation becomes much more optically complicated, with contributions not only from higher concentration of phytoplankton, but also from sediments and dissolved materials from terrestrial and sea bottom sources. The color often approaches yellow-brown in the turbidity waters (Case Ⅱ waters). To verify satellite ocean color retrievals, or to develop new algorithms for complex case Ⅱ regions requires ship-based studies. In this study, we compared the chlorophyll retrievals from NASA's SeaWiFS sensor with chlorophyll values determined with standard fluorometric methods during two cruises on Korean NFRDI ships. For the SeaWiFS data, we used the standard NASA SeaWiFS algorithm to estimate the chlorophyll_a distribution around the Korean waters using Orbview/ SeaWiFS satellite data acquired by our HPRT station at NFRDl. We studied In find out the relationship between the measured chlorophyll_a from the ship and the estimated chlorophyll_a from the SeaWiFs satellite data around the northern part of the East China Sea, in February, and May, 2000. The relationship between the measured chlorophyll_a and the SeaWiFS chlorophyll_a shows following the equations (1) In the northern part of the East China Sea. Chlorophyll_a =0.121Ln(X) + 0.504, R²= 0.73 (1) We also determined total suspended sediment mass (55) and compared it with SeaWiFS spectral band ratio. A suspended solid algorithm was composed of in-.situ data and the ratio (L/sub WN/(490 ㎚)L/sub WN/(555 ㎚) of the SeaWiFS wavelength bands. The relationship between the measured suspended solid and the SeaWiFS band ratio shows following the equation (2) in the northern part of the East China Sea. SS = -0.703 Ln(X) + 2.237, R²= 0.62 (2) In the near future, NFRDI will develop algorithms for quantifying the ocean color properties around the Korean waters, with the data from regular ocean observations using its own research vessels and from three satellites, KOMPSAT/OSMl, Terra/MODIS and Orbview/SeaWiFS.