본문 바로가기

쓰기

 
발표자 김누리 
발표일자 2021-10-05 
저자 Carlini, Nicholas, Ulfar Erlingsson, and Nicolas Papernot 
학회명 arXiv preprint arXiv:1910.13427 (2019). 
논문지  
We develop techniques to quantify the degree to which a given (training or testing) example is an outlier in the underlying distribution. We evaluate five methods to score examples in a dataset by how well-represented the examples are, for different plausible definitions of "well-represented", and apply these to four common datasets: MNIST, Fashion-MNIST, CIFAR-10, and ImageNet. Despite being independent approaches, we find all five are highly correlated, suggesting that the notion of being well-represented can be quantified. Among other uses, we find these methods can be combined to identify (a) prototypical examples (that match human expectations); (b) memorized training examples; and, (c) uncommon submodes of the dataset. Further, we show how we can utilize our metrics to determine an improved ordering for curriculum learning, and impact adversarial robustness. We release all metric values on training and test sets we studied.

    2021

      Deep Leakage from Gradients
      2021.11.09
      발표자: 홍만수     발표일자: 2021-11-09     저자: Ligeng Zhu, Zhijian Liu, Song Han     학회명: NeurIPS 2019    
      Model-Contrastive Federated Learning
      2021.10.12
      발표자: 강용훈     발표일자: 2021-10-12     저자: Qinbin Li, Binsheng He, Dawn Song     학회명: CVPR 2021    
      Semantic Robustness of Models of Source Code
      2021.10.12
      발표자: 김사무엘     발표일자: 2021-10-12     저자: Goutham Ramakrishnan, Jordan Henkel, Zi Wang, Aws Albarghouthi, Somesh Jha, Thomas Reps