Performance Analysis and Characterization of Training Deep Learning Models on Mobile Device

Abstract

Training deep learning models on mobile devices recently becomes possible, because of increasing computation power on mobile hardware and the advantages of enabling high user experiences. Most of the existing work on machine learning at mobile devices is focused on the inference of deep learning models, but not training. The performance characterization of training deep learning models on mobile devices is largely unexplored, although understanding the performance characterization is critical for designing and implementing deep learning models on mobile devices.In this paper, we perform a variety of experiments on a representative mobile device (the NVIDIA TX2) to study the performance of training deep learning models. We introduce a benchmark suite and a tool to study performance of training deep learning models on mobile devices, from the perspectives of memory consumption, hardware utilization, and power consumption. The tool can correlate performance results with fine-grained operations in deep learning models, providing capabilities to capture performance variance and problems at a fine granularity. We reveal interesting performance problems and opportunities, including under-utilization of heterogeneous hardware, large energy consumption of the memory, and high predictability of workload characterization. Based on the performance analysis, we suggest interesting research directions.

Detection of Fraud Transactions Using Recurrent Neural Network during COVID-19

“…Deep learning (DL) [3] belongs to broader family of Machine Learning. These techniques are consisting of algorithms those are inspired by operations of human brains.…”

“…Deep neural networks (DNN) are often considered as an improvement over traditional artificial neural network (ANN) [11] in the sense that it incorporates multiple layers into its architecture. DNN can learn hierarchical feature representation from the data itself by discovering higher level feature extraction from lower level features [3]. Any deep models are thought of as multi-layer architecture that accepts input vector and maps them into corresponding output labels.…”

“…The objective of this paper is to establish an effective and accurate fraudulent financial mobile money transaction detection model with high efficiency and low error rate. It utilises Deep Learning (DL) [3] techniques for implementing this model. These techniques are beneficial since it automatically captures hierarchical features present in the financial dataset.…”

Video-Based Parking Occupancy Detection for Smart Control System

“…Our approach displayed strong indications that it could be effectively generalized to other parking lots. In this paper, a method for efficient street-side parking occupancy detection on embedded devices is proposed, which is easily deployed on any street using existing roadside equipment and real-time processing on a Jetson TX2 [33][34][35]. The method uses existing surveillance cameras and embedded devices to produce an efficient, high-speed, and lightweight detection model.…”

Associating Unemployment with Panic Attack Using Deep Learning during COVID-19

“…Input variables are mapped into target classes using classification techniques which are supervised learning algorithms. For classification purposes, Deep Learning (DL) [4] based framework is implemented in this paper. DL technique is popular because of its self-adaptive structure that processes data with minimal processing.…”