• 정보보호학회논문지
• /
• 제33권6호
• /
• pp.1111-1123
• /
• 2023

최근 생성 모델이 발전함에 따라 생성 모델을 위협하는 연구도 활발히 진행되고 있다. 본 논문은 Text-to-Image 모델에 대한 멤버십 추론 공격을 위한 새로운 제안 방법을 소개한다. 기존의 Text-to-Image 모델에 대한 멤버십 추론 공격은 쿼리 이미지의 caption으로 단일 이미지를 생성하여 멤버십을 추론하였다. 반면, 본 논문은 Textual Inversion을 통해 쿼리 이미지에 personalization된 임베딩을 사용하고, Adversarial Prompt 생성 방법으로 여러 장의 이미지를 효과적으로 생성하는 멤버십 추론 공격을 제안한다. 또한, Text-to-Image 모델 중 주목받고 있는 Stable Diffusion 모델에 대한 멤버십 추론 공격을 최초로 진행하였으며, 최대 1.00의 Accuracy를 달성한다.

• 지능정보연구
• /
• 제9권2호
• /
• pp.19-38
• /
• 2003

In this research, we propose the mechanism to develop self-evolving expert systems (SEES) based on data mining (DM), fuzzy neural networks (FNN), and relational database (RDB)-driven forward/backward inference engine. Most researchers had tried to develop a text-oriented knowledge base (KB) and inference engine (IE). However, this approach had some limitations such as 1) automatic rule extraction, 2) manipulation of ambiguousness in knowledge, 3) expandability of knowledge base, and 4) speed of inference. To overcome these limitations, knowledge engineers had tried to develop an automatic knowledge extraction mechanism. As a result, the adaptability of the expert systems was improved. Nonetheless, they didn't suggest a hybrid and generalized solution to develop self-evolving expert systems. To this purpose, we propose an automatic knowledge acquisition and composite inference mechanism based on DM, FNN, and RDB-driven inference engine. Our proposed mechanism has five advantages. First, it can extract and reduce the specific domain knowledge from incomplete database by using data mining technology. Second, our proposed mechanism can manipulate the ambiguousness in knowledge by using fuzzy membership functions. Third, it can construct the relational knowledge base and expand the knowledge base unlimitedly with RDBMS (relational database management systems) module. Fourth, our proposed hybrid data mining mechanism can reflect both association rule-based logical inference and complicate fuzzy relationships. Fifth, RDB-driven forward and backward inference time is shorter than the traditional text-oriented inference time.

• 한국산학기술학회:학술대회논문집
• /
• 한국산학기술학회 2003년도 Proceeding
• /
• pp.99-104
• /
• 2003

In this research, we proposed the mechanism to develop self evolving expert systems (SEES) based on data mining (DM), fuzzy neural networks (FNN), and relational database (RDB)-driven forward/backward inference engine. Most former researchers tried to develop a text-oriented knowledge base (KB) and inference engine (IE). However, thy have some limitations such as 1) automatic rule extraction, 2) manipulation of ambiguousness in knowledge, 3) expandability of knowledge base, and 4) speed of inference. To overcome these limitations, many of researchers had tried to develop an automatic knowledge extraction and refining mechanisms. As a result, the adaptability of the expert systems was improved. Nonetheless, they didn't suggest a hybrid and generalized solution to develop self-evolving expert systems. To this purpose, in this study, we propose an automatic knowledge acquisition and composite inference mechanism based on DM, FNN, and RDB-driven inference. Our proposed mechanism has five advantages empirically. First, it could extract and reduce the specific domain knowledge from incomplete database by using data mining algorithm. Second, our proposed mechanism could manipulate the ambiguousness in knowledge by using fuzzy membership functions. Third, it could construct the relational knowledge base and expand the knowledge base unlimitedly with RDBMS (relational database management systems). Fourth, our proposed hybrid data mining mechanism can reflect both association rule-based logical inference and complicate fuzzy logic. Fifth, RDB-driven forward and backward inference is faster than the traditional text-oriented inference.

• 대한임베디드공학회논문지
• /
• 제17권6호
• /
• pp.347-353
• /
• 2022

In this paper, we propose a method that performs a text-video retrieval model by replacing video properties using captions. In general, the exisiting embedding-based models consist of both joint embedding space construction and the CNN-based video encoding process, which requires a lot of computation in the training as well as the inference process. To overcome this problem, we introduce a video-captioning module to replace the visual property of video with captions generated by the video-captioning module. To be specific, we adopt the caption generator that converts candidate videos into captions in the inference process, thereby enabling direct comparison between the text given as a query and candidate videos without joint embedding space. Through the experiment, the proposed model successfully reduces the amount of computation and inference time by skipping the visual processing process and joint embedding space construction on two benchmark dataset, MSR-VTT and VATEX.

• Communications for Statistical Applications and Methods
• /
• 제7권3호
• /
• pp.913-926
• /
• 2000

An electronic statistics text on the web is implemented. The introduced text provide interactive instructions on the statistical estimation and inference. As a by-product, we also provide a calculation of quantiles and p-value of t-distribution and standard normal distribution. This program was written in JAVA programming language.

• 말소리와 음성과학
• /
• 제13권4호
• /
• pp.47-53
• /
• 2021

Autoregressive한 TTS 모델은 불안정성과 속도 저하라는 본질적인 문제를 안고 있다. 모델이 time step t의 데이터를 잘못 예측했을 때, 그 뒤의 데이터도 모두 잘못 예측하는 것이 불안정성 문제이다. 음성 출력 속도 저하 문제는 모델이 time step t의 데이터를 예측하려면 time step 1부터 t-1까지의 예측이 선행해야 한다는 조건에서 발생한다. 본 연구는 autoregression이 야기하는 문제의 대안으로 end-to-end non-autoregressive 가속 TTS 모델을 제안한다. 본 연구의 모델은 Tacotron 2 - WaveNet 모델과 근사한 MOS, 더 높은 안정성 및 출력 속도를 보였다. 본 연구는 제안한 모델을 토대로 non-autoregressive한 TTS 모델 개선에 시사점을 제공하고자 한다.

• 영어어문교육
• /
• 제10권3호
• /
• pp.21-40
• /
• 2004

This study Questioned what happens in L2 reading comprehension of the expository text, as measured by recall and inference-making abilities, when a L2 reader was induced to develop a content schema about the topic of a target text, but the structure of that schema departs from the structure of the target text Seventy-four. Korean university students read either the same version text twice (consistent condition) or two different version texts (inconsistent condition) with a three-day interval between the two readings. The results of a verification test indicate that, for those subjects with higher L2 reading proficiency, the inconsistent condition was more beneficial than the consistent condition for the inference-making task. On the other hand, for lower-level L2 readers, the consistent condition was more favorable for the recall task. It was concluded that inducing a structurally inconsistent schema through an L2 pre-reading would be beneficial only when the reader's L2 linguistic ability is proficient enough to produce necessary propositions from the pre-reading.

• 한국컴퓨터정보학회논문지
• /
• 제17권2호
• /
• pp.139-147
• /
• 2012

본 논문에서는 텍스트 마이닝을 통해 생물학 문헌에서 분자 수준의 사건(event) 정보를 자동으로 추출하고, 이들 사건 정보를 기반으로 새로운 생물학 지식을 자동 추론하는 텍스트 마이닝 - 추론 통합 구조의 시스템을 다룬다. 이러한 통합 구조의 지식 발견 시스템은 미리 추출되어 데이터베이스에 등록된 정보만을 입력으로 사용하는 시스템들에 비하여 최신 정보를 보다 빨리 사용할 수 있고, 미리 정의된 형식 이외의 다양한 정보를 사용할 수 있다는 장점이 있다. 반면, 텍스트 마이닝 정보 추출 결과를 그대로 사용하기 때문에 텍스트 마이닝 모듈(module)의 성능에 따라 전체 시스템의 효용성이 크게 저하될 수도 있다는 문제가 있다. 본 논문에서는 확률 기반 필터링(filtering) 방법을 제안하여, 텍스트 마이닝 결과 중 양성 오류(false positive)를 효과적으로 제거함으로써 전체 지식 발견 시스템의 정확도 및 효용성을 높이고자 한다. 본 논문에서 제안한 확률 기반 필터링 방법은 기준(baseline) 방법으로 사용된 횟수 기반 필터링 방법보다 높은 성능을 보였다.

• 정보과학회 논문지
• /
• 제44권2호
• /
• pp.179-185
• /
• 2017

문서의 요약은 요약문 내의 문장들끼리 서로 연관성 있게 이어져야 하고 하나의 짜임새 있는 글이 되어야 한다. 본 논문에서는 위의 목적을 달성하기 위해 문장 간의 유사도와 수반 관계(Entailment)를 고려하여 문서 내에서 연관성이 크고 의미, 개념적인 연결성이 높은 문장들을 추출할 수 있도록 하였다. 본 논문에서는 Recurrent Neural Network 기반의 문장 관계 추론 모델과 그래프 기반의 랭킹(Graph-based ranking) 알고리즘을 혼합하여 단일 문서 추출요약 작업에 적용한 새로운 알고리즘인 TextRank-NLI를 제안한다. 새로운 알고리즘의 성능을 평가하기 위해 기존의 문서요약 알고리즘인 TextRank와 동일한 데이터 셋을 사용하여 성능을 비교 분석하였으며 기존의 알고리즘보다 약 2.3% 더 나은 성능을 보이는 것을 확인하였다.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2003년도 추계 학술대회 학술발표 논문집
• /
• pp.49-52
• /
• 2003

In this research, we propose a mechanism to develop an inference engine and expert systems based on relational database and SQL (structured query language). Generally, former researchers had tried to develop an expert systems based on text-oriented knowledge base and backward/forward (chaining) inference engine. In these researches, however, the speed of inference was remained as a tackling point in the development of agile expert systems. Especially, the forward inference needs more times than backward inference. In addition, the size of knowledge base, complicate knowledge expression method, expansibility of knowledge base, and hierarchies among rules are the critical limitations to develop an expert systems. To overcome the limitations in speed of inference and expansibility of knowledge base, we proposed a relational database-oriented knowledge base and forward inference engine. Therefore, our proposed mechanism could manipulate the huge size of knowledge base efficiently, and inference with the large scaled knowledge base in a short time. To this purpose, we designed and developed an SQL-based forward inference engine using relational database. In the implementation process, we also developed a prototype expert system and presented a real-world validation data set collected from medical diagnosis field.