通过跨域几何一致性在VFM衍生潜伏空间中校准偏差分布
IEEE transactions on pattern analysis and machine intelligence
|February 9, 2026
概括
这项研究弥合了训练数据和真实分布之间的差距,使用来自基础模型的几何知识. 这种方法通过校准数据分布来增强联合学习和长尾识别.
科学领域:
- 计算机科学 计算机科学
- 人工智能的人工智能
- 机器学习 机器学习
背景情况:
- 深度学习模型面临的挑战是由于观察到的训练样本和真实数据分布之间的差异,由采样偏差和噪声等因素引起.
- 基础模型提供了强大的特征提取能力,但它们在解决分布差距方面的应用需要进一步调查.
研究的目的:
- 引入一个以几何知识为指导的分布校准框架,以解决数据分布差异.
- 用基础模型证明几何特征分布形状在域和数据集中的可转移性.
- 在具有挑战性的环境中验证框架的有效性,例如联合学习和长尾识别.
主要方法:
- 利用现成的视觉基础模型 (例如CLIP,DINOv2) 进行特征提取以捕捉几何分布形状.
- 开发一种保护隐私的技术,以获得全球几何形状,用于样本生成的联合学习.
- 利用从数据丰富的类别中转移的几何知识来重建长尾识别中数据稀缺类的分布.
主要成果:
- 通过基础模型提取的特征分布的几何形状显示了不同领域和数据集的显著可转移性.
- 拟议的框架通过生成新的客户样本,有效地弥合了联合学习中的本地和全球观察之间的差距.
- 在长尾识别中,几何知识转移显著改善了样本稀缺尾部类的真实分布的恢复.
- 综合实验证实,在联合学习和长尾识别设置中,跨基准的性能都得到了提升.
结论:
- 以几何知识为指导的分布校准是一种可行的策略,可以克服由数据异质性和样本不平衡引起的信息缺陷.
- 从基础模型中转移几何特征分布属性的可能性为提高深度学习模型稳定性和性能提供了一个有希望的方向.
- 该框架在解决联合学习和长尾认可方面的关键挑战方面具有实际实用性,增强了模型的概括性和公平性.
相关概念视频
Geometric Mean
4.1K
The mean is a measure of the central tendency of a data set. In some data sets, the data is inherently multiplicative, and the arithmetic mean is not useful. For example, the human population multiplies with time, and so does the credit amount of financial investment, as the interest compounds over successive time intervals.
In cases of multiplicative data, the geometric mean is used for statistical analysis. First, the product of all the elements is taken. Then, if there are n elements in the...
In cases of multiplicative data, the geometric mean is used for statistical analysis. First, the product of all the elements is taken. Then, if there are n elements in the...
4.1K
Confirmation Biases
8.3K
The confirmation bias is the tendency to focus on information that confirms our existing beliefs and ignore information that is inconsistent with our expectations. For example, if you think that your professor is not very nice, you notice all of the instances of rude behavior exhibited by the professor while ignoring the countless pleasant interactions he is involved in on a daily basis. Have you ever fallen prey to the confirmation bias, either as the source or target of such bias?
8.3K
Geometric Sequences
289
In systems where values diminish by a constant proportion at each stage, the resulting sequence follows a geometric structure. Each new value in the sequence is obtained by applying a fixed multiplier to the preceding term. This regular, proportional decline type is often used to represent processes involving gradual loss, such as energy dissipation or reduction in amplitude over time.When analyzing the total effect of such a process across unlimited iterations, the series of values is referred...
289
Crossing Over
172.3K
Unlike mitosis, meiosis aims for genetic diversity in its creation of haploid gametes. Dividing germ cells first begin this process in prophase I, where each chromosome—replicated in S phase—is now composed of two sister chromatids (identical copies) joined centrally.
The homologous pairs of sister chromosomes—one from the maternal and one from the paternal genome—then begin to align alongside each other lengthwise, matching corresponding DNA positions in a process...
The homologous pairs of sister chromosomes—one from the maternal and one from the paternal genome—then begin to align alongside each other lengthwise, matching corresponding DNA positions in a process...
172.3K
Hindsight Biases
4.3K
Hindsight bias leads you to believe that the event you just experienced was predictable, even though it really wasn’t. In other words, you knew all along that things would turn out the way they did. Can you relate this to the phrase "Hindsight is 20/20" now?
4.3K
Bias
7.4K
Bias refers to any tendency that prevents a question from being considered unprejudiced. In research, bias occurs when one outcome or answer is selected or encouraged over others in sampling or testing. Bias can occur during any research phase, including study design, data collection, analysis, and publication.
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
7.4K


