Enabling All In-Edge Deep Learning: A Literature Review

Abstract

In recent years, deep learning (DL) models have demonstrated remarkable achievements on non-trivial tasks such as speech recognition, image processing, and natural language understanding. One of the significant contributors to its success is the proliferation of end devices that acted as a catalyst to provide data for data-hungry DL models. However, computing DL training and inference is the main challenge. Usually, central cloud servers are used for the computation, but it opens up other significant challenges, such as high latency, increased communication costs, and privacy concerns. To mitigate these drawbacks, considerable efforts have been made to push the processing of DL models to edge servers (a mesh of computing devices near end devices). Moreover, the confluence point of DL and edge has given rise to edge intelligence (EI). International Electrotechnical Commission (IEC) defines EI as the concept where the data is acquired, stored, and processed utilizing edge computing with DL and advanced networking capabilities. Broadly, EI has six levels of categories based on where the training and inference of DL take place, e.g., cloud server, edge server and end devices. This survey paper focuses primarily on the fifth level of EI, called all in-edge level, where DL training and inference (deployment) are performed solely by edge servers. All in-edge is suitable when the end devices have low computing resources, e.g., Internet-of-Things, and other requirements such as latency and communication cost are important such as in mission-critical applications (e.g., health care). Besides, 5G/6G networks are envisioned to use all in-edge. Firstly, this paper presents all in-edge computing architectures, including centralized, decentralized, and distributed. Secondly, this paper presents enabling technologies, such as model parallelism, data parallelism, and split learning, which facilitates DL training and deployment at edge servers. Thirdly, model adaptation techniques based on model compression and conditional computation are described because the standard cloud-based DL deployment cannot be directly applied to all in-edge due to its limited computational resources. Fourthly, this paper discusses eleven key performance metrics to evaluate the performance of DL at all in-edge efficiently. Finally, several open research challenges in the area of all in-edge are presented. INDEX TERMS Artificial intelligence, all in-edge, deep learning, distributed systems, decentralized systems, edge intelligence I. INTRODUCTION T HE global community is increasingly becoming a datadriven environment in which end devices are generating vast quantities of data outside of the traditional data centers. International Telecommunication Union anticipates that global internet traffic per month will reach 607 Exabytes (EB) in 2025 and 5016 EB in 2030 [1]. This enormous amount of data has a positive impact on artificial intelligence (AI) applications. In particular, deep learning (DL) rely on the availability of large quantities of data for its d...

Evaluation of artificial intelligence techniques in disease diagnosis and prediction

“…The complexity of computation and architecture in DL-based model is another challenge in this area. One potential solution for reducing the complexity of computation and architecture in DL models is to use model compression techniques such as pruning, quantization, and low-rank factorization [ 124 ]. These techniques can help reduce the number of parameters and computational resources required while maintaining good performance.…”

Deep Learning in the Ubiquitous Human–Computer Interactive 6G Era: Applications, Principles and Prospects

“…Summing up the representative works from another perspective, we can find some common schemes, as shown in Table 2. IRS element selection, phase-shift control and power allocation [60] Transmitter power, PS factor of UE and IRS phase shifts matrix [82] Task offloading rate, power consumption and computational overhead [94] Task offloading and resource allocation [95] Latency, interference management and computational resource management [70] Simplification of DRL decision space Adopt a decision selection network to filter inappropriate mode alterations from the action space of QNet model [75] Adopt LSTM to simplify the original state-action space and produce an approximate state-action space [94] Adopt the concept of rolling horizon control to slow down the growth of action aggregations [95] Deep learning as an alternative to traditional analytical methods…”

“…Obtain training data by AO (alternative optimization)-based scheme) to train a FFNN [82] Obtain training data from traditional ILP (Integer Linear Programming) to train an E-CNN [99] Resolve a NP-hard MINLP (Mix-Integer Nonlinear Programming) problem using MADRL [60] Distributed deep learning Beam forming by DRL-based Federal Learning, to reduce communication overhead and avoid phase synchronization [64] Use muti-agent DRL (MADRL) to efficiently coordinate multi-AP and multi-IRS in a distributed manner, at lower information exchange costs [60] DRL trained by MADDPG to tradeoff between energy efficiency and accuracy [77] FL (distributed DRL) to cope with the dynamic context of 6G [97] MADDPG to integrate computation and communication into edge nodes of deep edge network, aiming at pervasive intelligence in edge computing [98] First, multi-objective optimization. Due to the characteristics like sensitivity to signal blockage, the antennas and IRS of THz or mmWave communication exhibit high complexity.…”

“…The forthcoming 6G era is expecting various burgeoning human–computer interactive applications, including not only immersive applications like metaverse but also IoT applications like V2X (Vehicle to Everything) applications. Facing the unparalleled challenges put forward by ultra-high-speed communications and tremendous IoT linkages, THz MIMO-NOMA (massive multiple-input-multiple output non-orthogonal multiple access) has been proven to be an indispensable technology for 6G [ 60 ]. The THz MIMO-NOMA system leverages a large-scale antenna array supported by hybrid beamforming infrastructure, which can significantly alleviate attenuation on the THz bands and decrease the hardware complexity and energy consumption.…”

“…Deep learning can still play a vital role even if traditional analytical solutions are available. On one hand, some problems are NP-hard in the analytical solution context [ 60 , 69 ]. On the other hand, in some applications, the analytical methods result in excessively high computation complexity even if they can reach the solutions in the polynomial time [ 82 , 99 ].…”