• Title/Summary/Keyword: Long-tail classification

Search Result 7, Processing Time 0.026 seconds

Triplet Class-Wise Difficulty-Based Loss for Long Tail Classification

  • Yaw Darkwah Jnr.;Dae-Ki Kang
    • International Journal of Internet, Broadcasting and Communication
    • /
    • v.15 no.3
    • /
    • pp.66-72
    • /
    • 2023
  • Little attention appears to have been paid to the relevance of learning a good representation function in solving long tail tasks. Therefore, we propose a new loss function to ensure a good representation is learnt while learning to classify. We call this loss function Triplet Class-Wise Difficulty-Based (TriCDB-CE) Loss. It is a combination of the Triplet Loss and Class-wise Difficulty-Based Cross-Entropy (CDB-CE) Loss. We prove its effectiveness empirically by performing experiments on three benchmark datasets. We find improvement in accuracy after comparing with some baseline methods. For instance, in the CIFAR-10-LT, 7 percentage points (pp) increase relative to the CDB-CE Loss was recorded. There is more room for improvement on Places-LT.

Review of Application Models According to the Classification of Asymptotic Tail Distribution (근사 꼬리분포의 유형별 적용 모형 고찰)

  • Choi, Sung-Woon
    • Proceedings of the Safety Management and Science Conference
    • /
    • 2010.11a
    • /
    • pp.35-39
    • /
    • 2010
  • The research classifies three types of asymptotic tail distributions such as long(heavy, thick) tailed distribution, medium tailed distribution and short(light, thin) tailed distribution. The extreme value distributions(EVD) classified in this paper can be used in SPC(Statistical Process Control) control chart and reliability engineering.

  • PDF

Image Classification using Class-Balanced Loss (Class-Balanced Loss를 이용한 이미지 분류)

  • Jihee Park;Wonjun Hwang
    • Proceedings of the Korean Society of Broadcast Engineers Conference
    • /
    • 2022.11a
    • /
    • pp.164-166
    • /
    • 2022
  • Long-tail problem은 class 별로 sample의 개수에 차이가 있어 성능에 안 좋은 영향을 미치는 것을 말한다. 본 논문에서는 cost-sensitive learning 중 Class-Balanced Loss를 이용해 성능을 개선하여 Long-tail problem을 해결하려고 한다. 먼저, balanced data set과 imbalanced data set의 성능 차이를 살펴보도록 할 것이다. 그 후, Class-Balanced Loss를 3가지 버전으로 이용해 그 성능을 측정하고 분석해 볼 것이다.

  • PDF

Review of Classification Models for Reliability Distributions from the Perspective of Practical Implementation (실무적 적용 관점에서 신뢰성 분포의 유형화 모형의 고찰)

  • Choi, Sung-Woon
    • Journal of the Korea Safety Management & Science
    • /
    • v.13 no.1
    • /
    • pp.195-202
    • /
    • 2011
  • The study interprets each of three classification models based on Bath-Tub Failure Rate (BTFR), Extreme Value Distribution (EVD) and Conjugate Bayesian Distribution (CBD). The classification model based on BTFR is analyzed by three failure patterns of decreasing, constant, or increasing which utilize systematic management strategies for reliability of time. Distribution model based on BTFR is identified using individual factors for each of three corresponding cases. First, in case of using shape parameter, the distribution based on BTFR is analyzed with a factor of component or part number. In case of using scale parameter, the distribution model based on BTFR is analyzed with a factor of time precision. Meanwhile, in case of using location parameter, the distribution model based on BTFR is analyzed with a factor of guarantee time. The classification model based on EVD is assorted into long-tailed distribution, medium-tailed distribution, and short-tailed distribution by the length of right-tail in distribution, and depended on asymptotic reliability property which signifies skewness and kurtosis of distribution curve. Furthermore, the classification model based on CBD is relied upon conjugate distribution relations between prior function, likelihood function and posterior function for dimension reduction and easy tractability under the occasion of Bayesian posterior updating.

Genetic Composition of Korean Native Chicken Populations - National Scale Molecular Genetic Evaluation Based on Microsatellite Markers (초위성체 표지로 본 한국 재래닭 집단의 분자유전학적 구성)

  • Lee, Poong-Yeon;Yeon, Seong-Heum;Kim, Jae-Hwan;Ko, Yeoung-Gyu;Son, Jun-Kyu;Lee, Hee-Hoon;Cho, Chang-Yeon
    • Korean Journal of Poultry Science
    • /
    • v.38 no.2
    • /
    • pp.81-87
    • /
    • 2011
  • The study was conducted to select and optimize microsatellite (MS) markers for evaluate Korean Native Chicken (KNC) breeds in order to provide standard for the classification and breed definition of the indigenous breeds. The study also aimed to characterize and classify each KNC populations for inventory and management of avian genetic resources. A total of 462 chickens from 11 populations of chicken breeds including eight KNC breeds and three commercial chicken breeds were analyzed with 19 MS markers. KNC breeds, especially Long-Tail Chicken breeds, formed separate cluster from those commercial chicken breeds. Genetic distances between KNC populations (0.11~0.18) were relatively shorter. Genetic uniformity of KNC (except KNCR breed) (0.86~0.88) were higher than that of commercial breeds (except Cornish) (0.95~0.97). On the other hand, genetic uniformity of KNC Long Tail (KNCLT) were relatively higher (0.91~0.97). The result can be used to evaluate and manage animal genetic resources at national scale.

Study on the Characteristics of Feather Developing Pattern and Morphology in Early- and Late-Feathering Korean Native Chickens (한국재래닭에 있어 조우성과 만우성 깃털의 발생 양상 및 형태적 특성 고찰)

  • Bang, Min Hee;Cho, Eun Jung;Cho, Chang Yeon;Sohn, Sea Hwan
    • Korean Journal of Poultry Science
    • /
    • v.45 no.3
    • /
    • pp.155-165
    • /
    • 2018
  • Chicken feathers could be classified into early-feathering (EF) and late-feathering (LF) depending on the development and patterns of the wing and tail feathers. Currently, feather-sexing is a widely used chick sexing method in the industry. This study was carried out to suggest the method of classifying of EF and LF chicks to establish auto-sexing Korean native chicken (KNC) strains. The development and morphology of wing feathers and tail feathers in 856 KNCs from hatching to 55-days old were analyzed to classify EF and LF chicks. We also performed PCR analysis using K-specific gene primers to confirm the agreement between the phenotypes and genotypes of EF and LF chickens. In the results, the EF chicks had long primaries and coverts, and there was a significant difference in length between primaries and coverts. The LF chicks had shorter primaries and coverts than the EF chicks, and showed little difference in the length between primaries and coverts. LF chicks could be classified into four groups: LF-Less, LF-Scant, LF-Equal and LF-Reverse according to their wing feather patterns. EF chicks had 1.5 times longer primaries than LF chicks until they were 15-days old, but the lengths were almost the same at 50-days old. The tail feathers of the EF chicks were apparent at 5-days old, but those of the LF chicks were short and indefinite at that time. When EF and LF chicks were classified by the length of primaries being more or less than 9 mm, the classification accuracies for EF and LF chicks were 96.2% and 85.4%, respectively, compared to the PCR results. In conclusion, juvenile EF and LF KNC showed distinct differences in feather development and morphology, and could be easily distinguished at one day-old.

This Study of the Arms Used in the Three Kingdoms (삼국시대(三國時代) 병기체제(兵器體制)의 연구(硏究))

  • Kim, sung-tae
    • Korean Journal of Heritage: History & Science
    • /
    • v.34
    • /
    • pp.20-58
    • /
    • 2001
  • In order to unravel the characteristics of arms used in the 'Three Kingdoms,' Kokuryo, Silla and Paikje. the classification and the developing procedures of the arms should be first discussed. At first, the basic arms of the soldiers of Three Kingdoms were iron swords, iron spearheads, and bows. During that period, swords attached a ring pommel were commonly used. But after 5A. D. a sword with a decoration pommel appeared. Infantry generally used iron spearheads. From the late 4A. D. the long spearheads were broadly used in cavalry battles. In the late 6A. D. infantry mainly used long spearheads, and this resulted in the foundation of long spearheads units. There were two kinds of bows: Short Bow whose arch is small and Long Bow whose arch is long. It is known that the Short Bow was widely used in Kokuryo and Paikje up to 5A. D. In the early era, infantry used Long Bow, yet it was vastly used after 6A. D. when a castle's strategical value was great and defending a castle was. significant. Above mentioned, as basic combat weapons, iron spearhead and bow were fundamental. In particular, the spearhead was the essential weapon to a soldier. Yet, arrow gun and hook-shape cutters were important weapons. Especially, after 6A.D., when a castle became strategically pivotal in military, the arrow gun became the important weapon. This resulted in the foundation of arrow gun units. Hook-shape cutters were used to snatch horsemen or to climb up to fall the castle. Yet, the cutter was not the Three Kingodoms' basic weapon. In addition, the three stages of arms development in the Three Kingdoms are formation stage, development stage, and settlement stage. The formation stage was the period when premitive military unit appeared in the Three Kingdoms. It ranged from 1B. C. to the mid 3A. D. At that time according to regions. there were two weapon systems operating: North area including Kokuryo and the northern part of Paikje and South area including Silla, Kaya and the southern part of Paikje. ln North area a sword with a ring attached at the end of the holder, iron spear with neck and mid-size flat holder and iron arrowhead with an extension to fix, were used. In this period, during a war calvary units were mostly used and their weapon systems seemed possibly to succeed in that of Kochosun. In the development stage, when LoLang's influence on surroundings became weak, Koguryo, Paikjae and Silla had directly contacted each other. In the late 3A.D. to the early 6A.D., Silla achieved a drastic improvement in weapon system. This was the period when Kokuryo played a leading role in arms race. Kokuryo's arms manufacturing techniques passed onto Silla, Kaya and Paikje. In combat strategy a joint operation between infantry and calvary prevailed even if their military tactics were different. In a calvary battle heavily armed horsemen played import roles at this period. The horsemen and even horses were heavily guarded with iron armors. After all, the appearance of fully armed horsemen implies the very need of powerful destructive forces in weapon system. At that time, basic weapons were a big sword with a ring attached at the end of the holder, swallow's tail-shape spear with neck, and iron spearhead with neck and an extension. The settlement stage began at the mid 6A.D., when it was the revolutionary period in the arms development history. Of course, actual proofs and picture documents were not sufficient enough to penetrate full scale of the weapon system. But, according to historical circumstances and historic records, it is very certain that this period was the peak in arms development. In this period special military units, such as infantry-calvary companies, Archery units and Long spear units, that executed particular duties with special weapons, were founded. This became the characteristics of the settlement stage.