Using Weakly Supervised Deep Learning to Classify and Segment Single‐Molecule Break‐Junction Conductance Traces

Abstract

In order to design molecular electronic devices with high performance and stability, it is crucial to understand their structure-to-property relationships. Single-molecule break junction measurements yield a large number of conductancedistance traces, which are inherently highly stochastic. Here we propose a weakly supervised deep learning algorithm to classify and segment these conductance traces, a method that is mainly based on transfer learning with the pretrain-finetune technique. By exploiting the powerful feature extraction capabilities of the pretrained VGG-16 network, our convolutional neural network model not only achieves high accuracy in the classification of the conductance traces, but also segments precisely the conductance plateau from an entire trace with very few manually labeled traces. Thus, we can produce a more reliable estimation of the junction conductance and quantify the junction stability. These findings show that our model has achieved a better accuracy-to-manpower efficiency balance, opening up the possibility of using weakly supervised deep learning approaches in the studies of single-molecule junctions.

Machine learning and analytical methods for single-molecule conductance measurements

“…The results of unsupervised learning are often evaluated using physical interpretations, and validation is often heuristic. The two main unsupervised learning methods used in singlemolecule measurements are clustering, which involves dividing data into several groups, [98][99][100][101][102][103][104][105][106][107][108][109][110][111][112] and feature extraction, which involves reducing the dimensionality of multi-dimensional data. 95,99,106,107,[113][114][115]…”

“…Another method using deep learning has been proposed, in which traces are treated as ordinary twodimensional images. 99,106,110 This image-based method can directly capture changes in conductance.…”

Trusting our machines: validating machine learning models for single-molecule transport experiments

“…Often, single-molecule transport studies, people choose This journal is © The Royal Society of Chemistry 2022 single-threshold metrics, such as accuracy, to evaluate their supervised model. 18,22,45,48,58 Choosing accuracy can be beneficial as it is intuitive, but it assumes the cost of misclassifying false positives and false negatives are equal. This assumption is not always sufficient as we will illustrate in the following example.…”

“…Often, single-molecule transport studies, people choose single-threshold metrics, such as accuracy, to evaluate their supervised model. 18,22,45,48,58…”

“…In the single-molecule transport community, there is a hope that clustering can lead to a more unbiased analysis of our data. 16,20,21,23,[44][45][46][47][48] In Table 3, we list a few, common clustering techniques though many more exists.…”