• 응용통계연구
• /
• 제24권2호
• /
• pp.425-433
• /
• 2011

Reject inference in credit scoring is a statistical approach to adjust for nonrandom sample bias due to rejected applicants. Function estimation approaches are based on the assumption that rejected applicants are not necessary to be included in the estimation, when the missing data mechanism is missing at random. On the other hand, the density estimation approach by using mixture models indicates that reject inference should include rejected applicants in the model. When mixture models are chosen for reject inference, it is often assumed that data follow a normal distribution. If data include missing values, an application of the normal mixture model to fully observed cases may cause another sample bias due to missing values. We extend reject inference by a multivariate normal mixture model to handle incomplete characteristic variables. A simulation study shows that inclusion of incomplete characteristic variables outperforms the function estimation approaches.

• Genomics & Informatics
• /
• 제17권4호
• /
• pp.37.1-37.8
• /
• 2019

Isolation-with-migration (IM) models have become popular for explaining population divergence in the presence of migrations. Bayesian methods are commonly used to estimate IM models, but they are limited to small data analysis or simple model inference. Recently three methods, IMa3, MIST, and AIM, resolved these limitations. Here, we describe the major problems addressed by these three software and compare differences among their inference methods, despite their use of the same standard likelihood function.

• 한국농공학회지
• /
• 제45권4호
• /
• pp.89-97
• /
• 2003

Using inference models developed for estimation of the parameters necessary to implement the Runoff Block of the Stormwater Management Model (SWMM), a number of alternative inference scenarios were developed to assess the influence of inference model complexity and structure on the calibration of the catchment modelling system. These inference models varied from the assumption of a spatially invariant value (catchment average) to spatially variable with each subcatchment having its own unique values. Fur-thermore, the influence of different measures of deviation between the recorded information and simulation predictions were considered. The results of these investigations indicate that the model performance is more influenced by model structure than complexity, and control parameter values are very much dependent on objective function selected as this factor was the most influential for both the initial estimates and the final results.

• 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.

• Nuclear Engineering and Technology
• /
• 제56권7호
• /
• pp.2704-2710
• /
• 2024

Nuclear archaeology research provides scientific methods to reconstruct the operating histories of fissile material production facilities to account for past fissile material production. While it has typically focused on analyzing material in permanent reactor structures, spent fuel or high-level waste also hold information about the reactor operation. In this computational study, we explore a Bayesian inference framework for reconstructing the operational history from measurements of isotope ratios from a sample of nuclear waste. We investigate two different inference models. The first model discriminates between three potential reactors of origin (Magnox, PWR, and PHWR) while simultaneously reconstructing the fuel burnup, time since irradiation, initial enrichment, and average power density. The second model reconstructs the fuel burnup and time since irradiation of two batches of waste in a mixed sample. Each of the models is applied to a set of simulated test data, and the performance is evaluated by comparing the highest posterior density regions to the corresponding parameter values of the test dataset. Both models perform well on the simulated test cases, which highlights the potential of the Bayesian inference framework and opens up avenues for further investigation.

• 지능정보연구
• /
• 제29권4호
• /
• pp.287-308
• /
• 2023

ChatGPT는 2022년 11월에 서비스를 시작한 후 급격하게 사용자 수가 늘어나며 인공지능의 역사에서 큰 전환점을 가져올 정도로 사회 곳곳에 많은 영향을 미치고 있다. 특히 ChatGPT와 같은 거대언어모델의 추론 능력은 프롬프트 엔지니어링 기법을 통해 빠른 속도로 그 성능이 발전하고 있다. 인공지능을 워크플로우에 도입하려고 하는 기업이나 활용하려고 하는 개인에게 이와 같은 추론 능력은 중요한 요소로 고려될 수 있다. 본 논문에서는 거대언어모델에서 추론을 가능하게 한 문맥내 학습에 대한 이해를 시작으로 하여 프롬프트 엔지니어링의 개념과 추론 유형 및 벤치마크 데이터에 대해 설명하고, 이를 기반으로 하여 최근 거대언어모델의 추론 성능을 급격히 향상시킨 프롬프트 엔지니어링 기법들에 대해 조사하고 발전과정과 기법들 간의 연관성에 대해 상세히 알아보고자 한다.

• 품질경영학회지
• /
• 제27권3호
• /
• pp.125-141
• /
• 1999

Bayesian inference and model selection method for software reliability growth models are studied. Software reliability growth models are used in testing stages of software development to model the error content and time intervals between software failures. In this paper, we could avoid the multiple integration by the use of Gibbs sampling, which is a kind of Markov Chain Monte Carlo method to compute the posterior distribution. Bayesian inference and model selection method for Jelinski-Moranda and Goel-Okumoto and Schick-Wolverton models in software reliability with Poisson prior information are studied. For model selection, we explored the relative error.

• 반도체디스플레이기술학회지
• /
• 제20권4호
• /
• pp.151-156
• /
• 2021

Low-end IoT devices do not have enough computation and memory resources for DNN learning and inference. Integer quantization of real-type neural network models can reduce model size, hardware computational burden, and power consumption. This paper describes the design and implementation of a web-based quantization platform for CNN deep learning accelerator chips. In the web service platform, we implemented visualization of the model through a convenient UI, analysis of each step of inference, and detailed editing of the model. Additionally, a data augmentation function and a management function of files that store models and inference intermediate results are provided. The implemented functions were verified using three YOLO models.

• 한국시뮬레이션학회논문지
• /
• 제16권2호
• /
• pp.75-81
• /
• 2007

과거의 대기행렬에 대한 추론은 여러 가지 제약 조건이 포함되어 수행되었다. 본 논문의 기반이 되는 Larson의 추론엔진 또한 포아송 도착과정을 기본 가정으로 추론을 수행하였다. 그러나, 이러한 가정 없이 실측 데이터만을 가지고 추론을 수행할 수 있다. 이는 보다 정확한 시스템의 성능척도 계산은 물론 보이지 않는 시스템 내부를 규명하는데 유용한 도구가 될 것이다. 본 논문은 이러한 추론 방법을 제안하고 타당성을 검토하여 보다 나은 시스템 분석에 적용하고자 한다. 먼저 서버의 수를 알고 있는 경우에 대한 추론 방법을 소개하고, 이를 바탕으로 서버의 수를 모르는 경우의 추론에 대해 확장한다. 우리가 제안한 추론모델을 검증하기 위하여 시뮬레이션을 수행하고 실제 시스템 성능척도 값과 추론에 의한 값을 비교 검토하였다.

• Communications for Statistical Applications and Methods
• /
• 제11권1호
• /
• pp.161-168
• /
• 2004

A single-index model is useful in fields which employ multidimensional regression models. Many methods have been developed in parametric and nonparametric approaches. In this paper, posterior inference is considered and a wavelet series is thought of as a function approximated to a true function in the single-index model. The posterior inference needs a prior distribution for each parameter estimated. A prior distribution of each coefficient of the wavelet series is proposed as a hierarchical distribution. A direction $\beta$ is assumed with a unit vector and affects estimate of the true function. Because of the constraint of the direction, a transformation, a spherical polar coordinate $\theta$, of the direction is required. Since the posterior distribution of the direction is unknown, we apply a Metropolis-Hastings algorithm to generate random samples of the direction. Through a Monte Carlo simulation we investigate estimates of the true function and the direction.