无模型测试时间适应用于分布外检测
IEEE transactions on pattern analysis and machine intelligence
|October 6, 2025
概括
分布外 (OOD) 检测对于ML可靠性至关重要. 一个新的框架AdaODD,使用测试数据调整模型,显著减少假阳性并提高OOD检测准确性.
科学领域:
- 机器学习 机器学习
- 人工智能的人工智能
- 计算机视觉 计算机视觉
背景情况:
- 对可靠的机器学习 (ML) 模型来说,分布外 (OOD) 检测至关重要.
- 当前的OOD检测方法通常使用固定的标准,因数据分布的变化而限制其有效性.
- 理论工作表明,OOD检测仅在分布数据上是不可靠的.
研究的目的:
- 提出一个新的框架,AdaODD (非参数测试时间调整用于分布外检测),以提高OD检测.
- 提高ML模型在测试期间适应不断变化的数据分布的适应性.
- 减少OOD检测中的错误阳性率,特别是在分布重叠的情况下.
主要方法:
- 开发了AdaODD,这是OOD检测的非参数测试时间适应框架.
- 利用在线测试样本,在测试阶段持续调整模型.
- 将OOD实例的检测集成到决策过程中.
主要成果:
- 与最先进的OOD检测方法相比,AdaODD显示出显著的性能改进.
- 在CIFAR-10基准上实现了虚假阳性率 (FPR95) 的23.23%降低.
- 在ImageNet-1k基准测试中显示了38%的FPR95减少.
结论:
- 通过适应测试时间数据分布,AdaODD有效地增强了OOD检测.
- 该框架将OOD实例纳入框架的能力提高了稳定性和准确性.
- 理论验证支持AdaODD方法对可靠的OOD检测的有效性.
相关概念视频
Quantifying and Rejecting Outliers: The Grubbs Test
3.5K
Sometimes, a data set can have a recorded numerical observation that greatly deviates from the rest of the data. Assuming that the data is normally distributed, a statistical method called the Grubbs test can be used to determine whether the observation is truly an outlier. To perform a two-tailed Grubbs test, first, calculate the absolute difference between the outlier and the mean. Then, calculate the ratio between this difference and the standard deviation of the sample. This...
3.5K
Detection of Gross Error: The Q Test
6.9K
When one or more data points appear far from the rest of the data, there is a need to determine whether they are outliers and whether they should be eliminated from the data set to ensure an accurate representation of the measured value. In many cases, outliers arise from gross errors (or human errors) and do not accurately reflect the underlying phenomenon. In some cases, however, these apparent outliers reflect true phenomenological differences. In these cases, we can use statistical methods...
6.9K
Survival Tree
385
Survival trees are a non-parametric method used in survival analysis to model the relationship between a set of covariates and the time until an event of interest occurs, often referred to as the "time-to-event" or "survival time." This method is particularly useful when dealing with censored data, where the event has not occurred for some individuals by the end of the study period, or when the exact time of the event is unknown.
Building a Survival Tree
Constructing a...
Building a Survival Tree
Constructing a...
385
Distribution Reliability and Automation
497
Distribution reliability in electrical power systems is critical for ensuring an uninterrupted power supply to consumers at minimal cost. According to IEEE Standard Terms, reliability is the probability that a device will function without failure over a specified time period or amount of usage. For electric power distribution, this translates to maintaining continuous power supply and addressing customer concerns over power outages. Several indices, as defined by IEEE Standard 1366-2012, are...
497
The Anderson-Darling Test
1.1K
The Anderson-Darling test is a statistical method used to determine whether a data sample is likely drawn from a specific theoretical distribution. Unlike parametric tests, it does not require assumptions about specific parameters of the distribution. Instead, it compares the sample's empirical cumulative distribution function (ECDF) with the cumulative distribution function (CDF) of the hypothesized distribution. Critical values for the test are specific to the chosen distribution rather...
1.1K
Steps in Outbreak Investigation
488
In the ever-evolving field of public health, statistical analysis serves as a cornerstone for understanding and managing disease outbreaks. By leveraging various statistical tools, health professionals can predict potential outbreaks, analyze ongoing situations, and devise effective responses to mitigate impact. For that to happen, there are a few possible stages of the analysis:
488
