01873nas a2200253   4500000000100000000000100001008004100002260001200043653001200055653001800067653001400085653001100099653001500110100002700125700001600152700002300168700002000191245012100211856008000332300001200412490000600424520117500430022001401605       2023                            d  c06/202310aAndroid10aDeep Learning10aFramework10aImages10aTensorFlow1 aBeatriz Sainz-de-Abajo1 aSergio Laso1 aJose Garcia-Alonso1 aJavier Berrocal00aAdaptation of Applications to Compare Development Frameworks in Deep Learning for Decentralized Android Applications  uhttps://www.ijimai.org/journal/sites/default/files/2023-05/ijimai8_2_21.pdf  a224-2310 v83 aNot all frameworks used in machine learning and deep learning integrate with Android, which requires some prerequisites. The primary objective of this paper is to present the results of the analysis and a comparison of deep learning development frameworks, which can be adapted into fully decentralized Android apps from a cloud server. As a work methodology, we develop and/or modify the test applications that these frameworks offer us a priori in such a way that it allows an equitable comparison of the analysed characteristics of interest. These parameters are related to attributes that a user would consider, such as (1) percentage of success; (2) battery consumption; and (3) power consumption of the processor. After analysing numerical results, the proposed framework that best behaves in relation to the analysed characteristics for the development of an Android application is TensorFlow, which obtained the best score against Caffe2 and Snapdragon NPE in the percentage of correct answers, battery consumption, and device CPU power consumption. Data consumption was not considered because we focus on decentralized cloud storage applications in this study.   a1989-1660