Exploratory Machine Learning with Unknown Unknowns

Authors: Peng Zhao, Yu-Jie Zhang, Zhi-Hua Zhou10999-11006

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Theoretical analysis and empirical study on both synthetic and real datasets validate the efficacy of our proposal. We present empirical evaluations on synthetic data to illustrate the idea and further validate the effectiveness on real datasets.
Researcher Affiliation Academia Peng Zhao, Yu-Jie Zhang, Zhi-Hua Zhou National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China {zhaop, zhangyj, zhouzh}@lamda.nju.edu.cn
Pseudocode No Algorithm details are presented in the full version (Zhao, Zhang, and Zhou 2021). The main paper does not contain pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described in this paper.
Open Datasets Yes We further evaluate on a UCI benchmark dataset Mfeat (van Breukelen et al. 1998) and Real Disp2, which is an activities recognition task (Ba nos et al. 2012). http://archive.ics.uci.edu/ml/datasets/REALDISP+Activity+Recognition+Dataset
Dataset Splits Yes The threshold for the initial rejection model is selected by cross validation to ensure 95% accuracy on high-confidence samples.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, memory amounts) used for running its experiments, only discussing the methods and datasets.
Software Dependencies No The paper describes the mathematical components and parameters used (e.g., Gaussian kernel, regularization parameters), but does not provide specific software dependencies like programming languages, libraries, or solvers with version numbers.
Experiment Setup Yes The threshold θ is choose as one achieving best accuracy on the testing data from the pool [0.1, 0.2, 0.3, 0.4]. The budget ratio is b = 20%. The threshold for the initial rejection model is selected by cross validation to ensure 95% accuracy on high-confidence samples.