Topological structure of complex predictions

Topological structure of complex predictions
复制标题

复杂预测的拓扑结构

DOI:
10.1038/s42256-023-00749-8
复制
发表时间:
2023
影响因子:
23.8
通讯作者:
Gleich, David F.
Gleich, David F.
中科院分区:
计算机科学1区
文献类型:
--
作者:
Liu, Meng;Dey, Tamal K.;Gleich, David F.

文献摘要

参考文献

相似文献

当前复杂的预测模型是将深度神经网络、图卷积网络或传感器拟合到一组训练数据的结果。这些模型的一个关键挑战是它们是高度参数化的,这使得描述和解释预测策略变得困难。我们使用拓扑数据分析将这些复杂的预测模型转换为预测景观的简化拓扑视图。结果是一个预测图,可以比tSNE和UMAP等降维方法更具体地检查模型结果。这些方法可以扩展到不同领域的大型数据集。我们提出了一个基于转换器的模型的案例研究,该模型先前被设计用于预测数千个基因组轨迹中的DNA片段的表达水平。当该模型用于研究BRCA1基因的突变时,我们的拓扑分析表明,它对突变的位置和BRCA1的外显子结构敏感,这是基于降维的工具无法发现的。此外,拓扑框架提供了多种方法来检查结果,包括比模型不确定性更准确的误差估计。进一步的研究表明,这些想法如何在基于图的学习和图像分类中产生有用的结果。
Current complex prediction models are the result of fitting deep neural networks, graph convolutional networks or transducers to a set of training data. A key challenge with these models is that they are highly parameterized, which makes describing and interpreting the prediction strategies difficult. We use topological data analysis to transform these complex prediction models into a simplified topological view of the prediction landscape. The result is a map of the predictions that enables inspection of the model results with more specificity than dimensionality-reduction methods such as tSNE and UMAP. The methods scale up to large datasets across different domains. We present a case study of a transformer-based model previously designed to predict expression levels of a piece of DNA in thousands of genomic tracks. When the model is used to study mutations in theBRCA1gene, our topological analysis shows that it is sensitive to the location of a mutation and the exon structure ofBRCA1in ways that cannot be found with tools based on dimensionality reduction. Moreover, the topological framework offers multiple ways to inspect results, including an error estimate that is more accurate than model uncertainty. Further studies show how these ideas produce useful results in graph-based learning and image classification.
DOI: 10.1017/9781009099950
发表时间: 2022-02
期刊: --
影响因子: --
作者:
T. Dey;Yusu Wang
通讯作者: T. Dey;Yusu Wang
DOI: 10.1038/nature21056
发表时间: 2017-02-02
期刊: Nature
影响因子: 64.8
作者:
Esteva A;Kuprel B;Novoa RA;Ko J;Swetter SM;Blau HM;Thrun S
通讯作者: Thrun S
DOI: 10.1073/pnas.1900654116
发表时间: 2019-10-29
影响因子: 11.1
作者:
Murdoch, W. James;Singh, Chandan;Yu, Bin
通讯作者: Yu, Bin
DOI: 10.1038/s42256-019-0087-3
发表时间: 2019-09-01
影响因子: 23.8
作者:
Bergomi, Mattia G.;Frosini, Patrizio;Quercioli, Nicola
通讯作者: Quercioli, Nicola
用新的眼光看待:意义、空间、数据、真相
DOI: --
发表时间: 2023
期刊: Design and Culture
影响因子: 0.7
作者:
J. Christiansen
通讯作者: J. Christiansen