• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.3
• /
• pp.347-354
• /
• 2022

To reduce forgetfulness in continuous learning, in this paper, we propose an improved LwF model based on the knowledge transfer method, and we show its effectiveness by experiment. In LwF, if the domain of the learned data is different or the complexity of the data is different, the previously learned results are inaccurate due to forgetting. In particular, when learning continues from complex data to simple data, the phenomenon tends to get worse. In this paper, to ensure that the previous learning results are sufficiently transferred to the LwF model, we apply the knowledge transfer method to LwF, and propose an algorithm for efficient use. As a result, the forgetting phenomenon was reduced by an average of 8% compared to the existing LwF results, and it was effective even when the learning task became long. In particular, when complex data was first learned, the efficiency was improved more than 30% compared to LwF.

• Korean Journal of Agricultural Science
• /
• v.8 no.1
• /
• pp.51-63
• /
• 1981

The results of a crossbreeding experiment with a total of 315 litters and 325 pigs of Berkshires, Hampshires, Durocs, Landraces, Large Whites, eight different two-breed crosses and twelve different three-breed crosses, produced at Livestock Experiment Station from 1975 through 1979, are summarized as follows. 1. Number born alive per litter was largest in the D♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ mating, followed by the D♂${\times}$(H♂${\times}$L♀) $F_1$ mating, and smallest in the Hampshires. The pigs in the 3rd-6th parities had larger litter size at birth than those in other parities. 2. Birth weight of pig was heaviest in L♂${\times}$Lw♀ mating and lightest in the Large White. The total litter weight at birth was heaviest in the D♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ mating, followed by D♂${\times}$(H♂${\times}$L♀) $F_1$ ♀ and Lw♂${\times}$L♀ matings, and was smaller in Hampshires and Birkshires. 3. Litter size at weaning was largest in the D♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ mating, followed by D♂${\times}$(H♂${\times}$L♀) $F_1$ ♀ and Lw♂${\times}$L♀ matings, and was smaller in Durocs and Hampshires. The pigs in the 3rd-6th parities had larger litter size at weaning than those in other parities. 4. The total litter weight at weaning was heaviest in the D♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ mating, followed by H♂${\times}$(Lw♂${\times}$L♀) $F_1$♀ and Lw♂${\times}$L♀ matings, and was lighter in Durocs and Hampshires. The weaning weight of pig was largest in D♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ mating and lightest in L♂${\times}$H♀ mating. 5. Survival rate at weaning was highest in L♂${\times}$Lw♀ mating, followed by D♂${\times}$(L♂${\times}$H♀) $F_1$ ♀ and D♂${\times}$(H♂${\times}$L♀) $F_1$ ♀ mating, and was lowest in Durocs. 6. The three-breed cross from D♂${\times}$(H♂${\times}$L♀) $F_1$ ♀ mating had the highest average gain and lowest feed requirement per unit gain, followed by the D♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ and H♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ matings. The Birkshires and Landraces ranked lowest among the 25 mating groups compared for both of the traits. Males had higher average daily gain than females by about 0.06kg and had lower feed requirement by about 0.14. 7. The three-breed crosses from D♂${\times}$(H♂${\times}$L♀) $F_1$ ♀, D♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ and H♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀ matings reached 90kg body weight at younger age than the other groups. The D♂${\times}$(H♂${\times}$L♀) $F_1$ ♀ group reached 90kg at younger age than the Landrace by 39 days. 8. The dressing percentage and lean meat percentage tended to be higher in H♂${\times}$(Lw♂${\times}$L♀) $F_1$ ♀, H♂${\times}$L♀ and L♂${\times}$B♀ matings compared to the other mating groups. The loin-eye area was largest in the Lw♂${\times}$L♀ mating and smallest in the B♂${\times}$L♀ mating. Males had higher dressing percentage, higher lean meat percentage and lion-eye area than females. The backfat was thinnest in purebred Hampshire and was thickest in B♂${\times}$L♀ mating. 9. The results obtained in this study suggest that the two-breed cross from Lw♂${\times}$L♀ mating, and the three-breed crosses from D♂${\times}$(Lw♂${\times}$L-♀) $F_1$ ♀ and D♂${\times}$(H♂${\times}$L♀) $F_1$♀ matings are superior crossbreds for reproductive and pork prodnction performance.

• Journal of IKEEE
• /
• v.25 no.3
• /
• pp.412-419
• /
• 2021

In this paper, we propose a performance improvement of the LwF method using efficient path selection in Continuous Learning Environment. We compare performance and structure with conventional LwF. For comparison, we experiment with performance using MNIST, EMNIST, Fashion MNIST, and CIFAR10 data with different complexity configurations. Experiments show up to 20% improvement in accuracy for each task, which mitigating the Catastrophic Forgetting phenomenon in Continuous Learning environments.

• Journal of IKEEE
• /
• v.26 no.1
• /
• pp.19-26
• /
• 2022

In this paper, we propose an algorithm that applies a different variable lambda to each loss value to solve the performance degradation caused by domain differences in LwF, and show that the retention of past knowledge is improved. The lambda value could be variably adjusted so that the current task to be learned could be well learned, by the variable lambda method of this paper. As a result of learning by this paper, the data accuracy improved by an average of 5% regardless of the scenario. And in particular, the performance of maintaining past knowledge, the goal of this paper, was improved by up to 70%, and the accuracy of past learning data increased by an average of 22% compared to the existing LwF.

• Asian-Australasian Journal of Animal Sciences
• /
• v.6 no.4
• /
• pp.515-519
• /
• 1993

Young goats were fed low quality roughage ad libitum and supplements of insect-damaged Lablab purpureus (var. Highworth) seed fed at approximately 3, 6 or 12 g/kg liveweight (LW), or sweet lupin seed (Lupinus angustifolius var. Uniharvest) fed at 12 g/kg LW. Roughage intake was not changed by 3 or 6 g/kg LW levels of Lablab or by 12 g/kg LW lupin supplement, but was reduced (p<0.05) by 35% by 12 g/kg LW Lablab supplement. Organic matter (OM) digestibility was increased by all supplements, and digestible OM intake was increased by the 6 g/kg LW Lablab and 12 g/kg LW lupin supplements. LW gain and feed conversion ratio were not changed by 3 or 6 g/kg LW Lablab or the 12 g/kg LW lupin, but were reduced (p<0.05) by 12 g/kg LW Lablab supplement. It was concluded that young goats could efficiently utilize supplements of Lablab purpureus seed fed at levels of up to 6 g/kg LW. However, when 12 g/kg of the Lablab seed was fed, poor performance suggested that the goats were adversely affected by anti-nutritional factors which were not neutralized by rumen fermentation.

• Korean Journal of Organic Agriculture
• /
• v.6 no.1
• /
• pp.51-61
• /
• 1997

This experiment was carried out to investigate the growth response of 3 grasses to seed coating with chitosan solution and the attempt was made to estimate adequate seed coating concentrations of chitosan solution in each grass for the growth to be stimulated. Three species used in this experiment were orchardgrass, tall fescue and reed canarygrass. Six different seed coating concentrations of chitosan solution were applied as 0%(control), 0.01%, 0.05%, 0.1% and 1.0%, respectively. the results were obtained as follows; 1. Dry weight of tiller(WT), leaf area(LA), dry weight of leaf(LW), dry weight of stem(SW), dry weight of shoot(SHW), biological yield(BY) and C/F ratio were significantly different between species. 2. Number of tillers per plant(NT), dry weight of tiller(WT), dry weight of leaf(LW), dry weight of root(RW), dry weight of shoot(SHW), biological yield(BY) and T/R ration were significantly different between seed coating concentrations of chitosan solution. 3. The adequate seed coating concentrations of chitosan solution for the growth stimulating effect were different between species. The highest values of yield components and dry weight of plant parts were obtained at 1% in orchardgrass and tall fescue, and 0.05% in reed canarygrass, respectively. 4. Growth stimulating effect of seed coating in each species were different. The highest values were obtained in leaf area(LA), dry weight of leaf(LW), dry weight of root(RW), dry weight of shoot(SHW) and dry weight of biological yield(BY) in orchardgrass. The values of dry weight of stem(SW) and C/F ration were highest in reed canarygrass. 5. An increase in number of tillers per plant(NT), dry weight of leaf(LW), dry weight of stem(SW) and dry weight of root(RW) according to seed coating was attributed to the increase in dry weight of shoot(SHW). Among the aboved increasing factors, the dry weight of leaf(LW) was a main factor for the increase in dry weight of shoot(SHW). 6. An increase in dry weight of leaf(LW), dry weight of stem(SW) and dry weight of root(RW) according to seed coating was attributed to the increase in biological yield(BY). Both the dry weight of leaf(LW) and dry weight of root(RW) were main factors for the increase in biological yield(BY).

• Asian Journal of Turfgrass Science
• /
• v.1 no.1
• /
• pp.37-41
• /
• 1987

Effect of Rolling factor on the growth and thatch accumulation in tall fescue was studied from the viewpoint of estimate the rolling factor to obtain the highest values of growth characters and analysis of thatch accumulation. Rolling factors were 1.82, 3.33, 4.29 and 4.85, respectively. The results are may he summarized as follows ; 1. Rolling factor was affected to obstructive on the growth of tall fescue. Thus, the rolling factor ( RF) had significant negative correlated with the dry weight of plant ( DW ), leaf weight (LW), stem weight (SW), dry weight of thatch (Th), number of tillers (NT) and C / F ratio. 2. Rolling factor of 1.82 was an adequate rolling factor for the growth. 3. The dry weight of thatch(Th) had significant positive correlated with DW, LW, and NT, but negative correlated with the dry weight of thatch per a tiller ( th / NT). 4. Thatch accumulation system can be shown in following diagram. yield compnents - DW ~ Th - Rf (LW, SW, NT, C/F) th/NT 5. The dry weight of thatch per a tiller(th / NT) was a concerning factor for the losses of thatch.

• Korean Journal of Remote Sensing
• /
• v.18 no.4
• /
• pp.187-199
• /
• 2002

Correlation between chlorophyll a in the East China Sea and spectral bands (412, 443, 490, (510), 555, (676, 765)nm) of Ocean Scanning Multi-Spectral Imager (OSMI) including the profile multi-spectral radiometer (PRR-800) was studied. The values of remote sensing reflectance (Rrs) at the bands corresponding to the field chlorophyll $\alpha$ in the East China Sea were much higher than those in clear waters off California, USA. In case of the particle absorptions related to the chlorophyll a concentration at the spectral bands (440, 670nm) were much higher in the East China Sea than the ones in the clean waters off California. The normalized water leaving radiances (nLw) at 412, 443, 490, 555 nm of OSMI and the field chlorophyll a in the East China Sea were correlated each other. According to the results, the relationship between field chlorophyll $\alpha$ and nLw 410 nm in OSMI bands was the lowest, whereas that between field chlorophyll a and nLw 555 nm in the bands was the highest. Reciprocal action between the field chlorophyll a and the band ratio of the OSMI bands (nLw410/nLw555, nLw443/nLw555, nLw490/nLw555) was also studied. Relationship between the chlorophyll $\alpha$ and the band ratio (nLw490/nLw555) was highest in the OSMI bands. Relationship between the chlorophyll $\alpha$ and the ratio (nLw490/nLw555) was higher than one in the nLw410/nLw555. The difference in the estimated chlorophyll $\alpha$ (mg/m$^3$) between OSMI and SeaWiFS (Sea Viewing Wide Field-of-View Sensor) at the special observing stations in the northern eastern sea of Jeju Island in February 25, 2002 was about less than 0.3 mg/m$^3$ within 3 hours. It is suggested that OC2 (ocean color chlorophyll 2 algorithm) be used to get much better estimation of chlorophyll $\alpha$ from OSMI than the ones from the updated algorithms as OC4.

• Asian-Australasian Journal of Animal Sciences
• /
• v.14 no.12
• /
• pp.1724-1727
• /
• 2001

A trial was conducted to evaluate the growth performance and carcass characteristics of LW ${\times}$ Mukota $F_1$ crosses when fed diets containing graded levels of maize cob meal. Sixteen LW ${\times}$ Mukota $F_1$ crossbred pigs of approximately 4.5 months of age, were randomly allocated to four diets that contained 0, 100, 200 and 300 g maize cobs/kg, which corresponded to 276.4, 360.3, 402.9 and 523.5 g NDF/kg, respectively. The pigs were fed ad libitum for 14 weeks. The diets were formulated to contain similar levels of energy (ca. 9MJ ME/kg) and protein (ca.160 g CP/kg). Average daily feed intake (ADFI), daily gain (ADG) and feed conversion ratio (FCR) were monitored for 14 weeks. At slaughter, the cold dressed weight (CDM) and backfat thickness (BFT) were determined. There was no significant linear relationship (p>0.05) between level of cob inclusion and ADFI. The ADG decreased as level of maize cobs increased (p<0.001). The pigs that were on 300 g cobs/kg had the highest FCR (p<0.05) as compared to pigs on the other three diets. No differences (p>0.05) were observed in the CDM between pigs that were fed diets that had 0 and 100 g maize cobs/kg. In addition, pigs on 100 and 200 g cobs/kg diets had the same CDM (p>0.05). The diet that contained 300 g cobs/kg gave the lowest (p< 0.001) CDM. Both BFT parameters showed a decrease (p<0.05) as level of maize cobs increased. An increase in CDM was associated with an increase in BFT (p<0.001), with the correlation coefficient between K7.5 and CDM being 0.84 (p<0.001). It can, therefore, be concluded that crossbred pigs perform well on diets containing up to 200 g/kg maize cobs.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.1
• /
• pp.15-23
• /
• 2022

Among the learning methods for Continuous Learning environments, "Learning without Forgetting" has fixed regularization strengths, which can lead to poor performance in environments where various data are received. We suggest a way to set weights variable by identifying the features of the data we want to learn. We applied weights adaptively using correlation and complexity. Scenarios with various data are used for evaluation and experiments showed accuracy increases by up to 5% in the new task and up to 11% in the previous task. In addition, it was found that the adaptive weight value obtained by the algorithm proposed in this paper, approached the optimal weight value calculated manually by repeated experiments for each experimental scenario. The correlation coefficient value is 0.739, and overall average task accuracy increased. It can be seen that the method of this paper sets an appropriate lambda value every time a new task is learned, and derives the optimal result value in various scenarios.