• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.675-679
• /
• 2020

최근 이미지 분류의 성능 향상을 위해 깊은 레이어와 넓은 채널을 가지는 모델들이 제안되어져 왔다. 높은 분류 정확도를 보이는 모델을 제안하는 것은 과한 컴퓨팅 파워와 계산시간을 요구한다. 본 논문에서는 이미지 분류 기법에서 사용되는 딥 뉴럴 네트워크 모델에 있어, 프루닝 방법을 통해 상대적으로 불필요한 가중치를 제거함과 동시에 분류 정확도 하락을 최소로 하는 동적 필터 프루닝 방법을 제시한다. 원샷 프루닝 기법, 정적 필터 프루닝 기법과 다르게 제거된 가중치에 대해서 소생 기회를 제공함으로써 더 좋은 성능을 보인다. 또한, 재학습이 필요하지 않기 때문에 빠른 계산 속도와 적은 컴퓨팅 파워를 보장한다. ResNet20 에서 CIFAR10 데이터셋에 대하여 실험한 결과 약 50%의 압축률에도 88.74%의 분류 정확도를 보였다.

• Journal of KIISE:Software and Applications
• /
• v.28 no.12
• /
• pp.875-884
• /
• 2001

Although the self organizing map (SOM) is widely utilized in such fields of data visualization and topology preserving mapping, since it should have the topology fixed before trained, it has some shortcomings that it is difficult to apply it to practical problems, and classification capability is quite low despite better clustering performance. To overcome these points this paper proposes the dynamic topology preserving self-organizing map(DTSOM) that dynamically splits the output nodes on the map and trains them, and attempts to improve the classification capability by combining multiple DTSOMs K-Winner method has been applied to combine DTSOMs which produces K outputs with winner node selection method. This produces even better performance than the conventional combining methods such as majority voting weighting, BKS Bayesian, Borda, Condorect and reliability sum. DTSOM remedies the shortcoming of determining the topology in advance, and the classification rate increases significantly by combing multiple maps trained with different features. Experimental results with handwritten digit recognition indicate that the proposed method works out to problems of conventional SOM effectively so to improve the classification rate to 98.1%.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.6
• /
• pp.1625-1635
• /
• 1997

An extension of the weighted buddy system, called the extended-weighted buddy system, for dynamic memory allocation in an object-oriented computer is presented. The extended-weighted buddy system allows block sizes of $2^k,\;3*2^k,\;5*2^k,\;7*2^k$, whereas the original weighted buddy system allowed block sizes of $2^k\;and\;3*2^k$. This extension is achieved at only the cost of additional 3 bits per block for memory management unit. Simulation results are presented which compare our method with the weighted buddy system. These results indicate that, for uniform request distributions, our system has less internal memory fragmentation than the weighted buddy system(approximately 60%). And, for exponential request distributions, it has less internal memory fragmentation than the weighted buddy method (approximately 50%). The external fragmentation is greater for this system than the weighted buddy system. But, our system has less total memory fragmentation for exponential request distributions, and two systems take a similar total memory fragmentation for uniform request distributions, so we can substitutes the extended-weighted buddy system for weighted buddy system.

• Proceedings of the Korean Information Science Society Conference
• /
• 1999.10b
• /
• pp.380-382
• /
• 1999

파노라마 영상은 카메라의 동작을 고려하여 시공간적으로 영상을 정렬한 영상으로서 효과적으로 비디오 데이터를 표현할 수 있는 방법이다. 그러나 기존의 파노라마 영상 구성 기법들은 영상을 정렬할 때 오버랩(overlap)되는 화소값들을 단순히 미디언 필터링하기 때문에 시간의 경과에 따른 밝기값의 동적인 변화를 수용하지 못한다. 따라서 본 논문에서는 이런 문제점을 해결하기 위해 시간적으로 증가하는 가중치를 사용하여 영상을 통합함으로써 최근의 영상 정보를 충분히 반영하는 파노라마 영상 구성 방법을 제안한다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.8-8
• /
• 2018

전통적인 빈도해석은 정상성 가정을 기초로 단일 확률분포를 강우 및 홍수량 자료에 적용하는 과정을 통해 확률수문량을 추정하는 것을 목적으로 하고 있다. 그러나 전지구적인 기상학적 변동성 및 기후변화로 기인하는 극치수문량의 발생 빈도 및 양적 크기의 변화는 확률통계학적 관점에서 서로 다른 분포특성을 가지게 된다. 대표적인 기상변동성인 엘니뇨가 발생하는 경우 지역에 따라 홍수 및 가뭄이 발생 발생하게 되며, 이러한 극치수문량은 일반적으로 나타나는 홍수 및 가뭄의 분포특성과는 상이한 경우가 많다. 즉, 2개 이상의 확률분포 특성이 혼재된 혼합분포의 특성을 가지는 경우가 나타내게 되며 이를 고려한 빈도해석 기법의 개발 및 적용이 필요하다. 혼합분포를 활용한 빈도해석에서 가장 중요한 사항 중에 하나는 개별 분포에 적용되는 가중치를 추정하는 것으로서 통계학적 관점에서 자료의 특성에 근거하여 내재되어 있는 은닉상태(latent process)를 추정하는 과정과 유사하다. 이와 더불어 앞서 언급된 기상학적 변동성을 빈도해석에 반영하기 위한 비정상성 해석기법의 개발 및 적용도 필요하다. 본 연구에서는 혼합분포를 활용한 비정상성빈도해석모형을 개발하는데 목적이 있으며 개별매개변수의 동적거동 뿐만 아니라 가중치에 대한 시간적인 종속성도 고려할 수 있는 모형으로 동적모형으로 다양한 실험적 해석이 가능하다. 본 연구에서는 개발된 모형을 기반으로 엘니뇨와 같은 기상변동성에 따른 강우 및 홍수빈도해석 측면에서 은닉상태에 변화, 이로 인한 확률분포의 특성 및 설계수문량의 동적변동성을 평가하고자 한다.

• The KIPS Transactions:PartB
• /
• v.11B no.4
• /
• pp.471-476
• /
• 2004

Recently, provides information which is most necessary to the user the research against the web information recommendation system for the Internet shopping mall is actively being advanced. the back which it will drive in the object. In that Dynamic Web Recommendation Method Using SOM (Self-Organizing Feature Maps) has the advantages of speedy execution and simplicity but has the weak points such as the lack of explanation on models and fired weight values for each node of the output layer on the established model. The method proposed in this study solves the lack of explanation using the Bayesian reasoning method. It does not give fixed weight values for each node of the output layer. Instead, the distribution includes weight using Hybrid SOM. This study designs and implements Dynamic Web Recommendation Method Using Hybrid SOM. The result of the existing Web Information recommendation methods has proved that this study's method is an excellent solution.

• Journal of Korea Multimedia Society
• /
• v.11 no.4
• /
• pp.471-480
• /
• 2008

The deaf can not recognize useful sound informations such as alarm, doorbell, siren, car horn, and phone ring etc., because they have the hearing impairment. To solve this problems, portable hearing assistive devices which have suitable environment sound recognition methods are needed. In this paper, the DTW algorithm for sound recognition system with new active slope weighting constraint method was proposed. The environment sound recognition methods consist of three processes. First process is extraction of start point and end point using frequency and amplitude of sound. Second process is extraction of features and third process is classification of features for given segments. As a result of the experiment, the recognition rate of the proposed method is over 90%. And, the recognition rate of the proposed method increased about 20% than the conventional algorithm. Therefore if there are developed portable assistive devices which use developed method to recognize environment sound for hearing-impaired persons, they could be more convenient in life.

• Journal of the Korea Computer Industry Society
• /
• v.5 no.9
• /
• pp.921-928
• /
• 2004

This paper proposed dynamic adaptive model based on discrete probability distribution function and user profile in web based HyperMedia educational systems. This modelsrepresents application domain to weighted direction graph of dynamic adaptive objects andmodeling user actions using dynamically approach method structured on discrete probability function. Proposed probabilitic analysis can use that presenting potential attribute to useractions that are tracing search actions of user in WebMedia structure. This approach methodscan allocate dynamically appropriate profiles to user.

• Journal of the Korea Computer Industry Society
• /
• v.2 no.11
• /
• pp.1365-1372
• /
• 2001

We have suggested three learning rules and implemented an experimental system that automatically restructures hypertext networks according to their user browsing behavior. The system applies link weights to the hypertext networks and updates these weights according to the learning rules. The learning rules are based on how often a particular hyperlink is being traversed and operate on strictly local information of link traversals. Changes in network structure are fed back to users by dynamic link ordering according to descending link weight. This approach is used collaborative filtering and navigation support techniques. The system has been investigation to be able to structure random hypertext networks into valid representations of their user browsing preferences in experiments.

• Proceedings of the Korean Institute of Communication Sciences Conference
• /
• 2004.07a
• /
• pp.174-174
• /
• 2004