*1.3. Methodology*

Publications in the scope of this work have been collected in the following ways: Firstly, by systematically checking conference proceedings from the bi-annually 'International NILM Workshop' 2020 to 2016 (nilmworkshop.org, accessed on 11 January 2021) and from the 'ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys)' 2020 to 2015 (buildsys.acm.org, accessed on 11 January 2021). The first conference is specifically dedicated to NILM, and the second in 2020 featured a dedicated NILM track. Secondly, by searching on Google Scholar and IEEE Xplore ® for keyword combinations of 'DNN', 'deep learning', 'NILM', 'non-intrusive', 'load monitoring', and 'load disaggregation'. This search has been done on several occasions and by different persons. Thirdly, we checked very thoroughly all the references in identified papers for anything not ye<sup>t</sup> on our list. While this approach might have missed a few recent publications, we are fairly sure that the survey is quite complete for the past years because of the systematic checking of references. The last iteration of our search process has been performed at the end of November 2020. We resulted with the DNN-NILM publications listed in Table 2, which reflects accordingly the body of work this survey paper is based on. The literature review, discussion, and all conclusions are deduced solely from these publications.
