Edge computing is a promising solution for handling high-dimensional, multispectral analog data from sensors and IoT devices for applications such as autonomous drones. However, edge devices' limited storage and computing resources make it challenging to perform complex predictive modeling at the edge. Compute-in-memory (CiM) has emerged as a principal paradigm to minimize energy for deep learning-based inference at the edge. Nevertheless, integrating storage and processing complicates memory cells and/or memory peripherals, essentially trading off area efficiency for energy efficiency. This paper proposes a novel solution to improve area efficiency in deep learning inference tasks. The proposed method employs two key strategies. Firstly, a Frequency domain learning approach uses binarized Walsh-Hadamard Transforms, reducing the necessary parameters for DNN (by 87% in MobileNetV2) and enabling compute-in-SRAM, which better utilizes parallelism during inference. Secondly, a memory-immersed collaborative digitization method is described among CiM arrays to reduce the area overheads of conventional ADCs. This facilitates more CiM arrays in limited footprint designs, leading to better parallelism and reduced external memory accesses. Different networking configurations are explored, where Flash, SA, and their hybrid digitization steps can be implemented using the memory-immersed scheme. The results are demonstrated using a 65 nm CMOS test chip, exhibiting significant area and energy savings compared to a 40 nm-node 5-bit SAR ADC and 5-bit Flash ADC. By processing analog data more efficiently, it is possible to selectively retain valuable data from sensors and alleviate the challenges posed by the analog data deluge.
Various mobile applications that comprise dependent tasks are gaining widespread popularity and are increasingly complex. These applications often have low-latency requirements, resulting in a significant surge in demand for computing resources. With the emergence of mobile edge computing (MEC), it becomes the most significant issue to offload the application tasks onto small-scale devices deployed at the edge of the mobile network for obtaining a high-quality user experience. However, since the environment of MEC is dynamic, most existing works focusing on task graph offloading, which rely heavily on expert knowledge or accurate analytical models, fail to fully adapt to such environmental changes, resulting in the reduction of user experience. This paper investigates the task graph offloading in MEC, considering the time-varying computation capabilities of edge computing devices. To adapt to environmental changes, we model the task graph scheduling for computation offloading as a Markov Decision Process (MDP). Then, we design a deep reinforcement learning algorithm (SATA-DRL) to learn the task scheduling strategy from the interaction with the environment, to improve user experience. Extensive simulations validate that SATA-DRL is superior to existing strategies in terms of reducing average makespan and deadline violation.
Tensor Processing Units (TPUs) are specialized hardware accelerators for deep learning developed by Google. This paper aims to explore TPUs in cloud and edge computing focusing on its applications in AI. We provide an overview of TPUs, their general architecture, specifically their design in relation to neural networks, compilation techniques and supporting frameworks. Furthermore, we provide a comparative analysis of Cloud and Edge TPU performance against other counterpart chip architectures. Our results show that TPUs can provide significant performance improvements in both cloud and edge computing. Additionally, this paper underscores the imperative need for further research in optimization techniques for efficient deployment of AI architectures on the Edge TPU and benchmarking standards for a more robust comparative analysis in edge computing scenarios. The primary motivation behind this push for research is that efficient AI acceleration, facilitated by TPUs, can lead to substantial savings in terms of time, money, and environmental resources.
We introduce a distinctive real-time, causal, neural network-based active speaker detection system optimized for low-power edge computing. This system drives a virtual cinematography module and is deployed on a commercial device. The system uses data originating from a microphone array and a 360-degree camera. Our network requires only 127 MFLOPs per participant, for a meeting with 14 participants. Unlike previous work, we examine the error rate of our network when the computational budget is exhausted, and find that it exhibits graceful degradation, allowing the system to operate reasonably well even in this case. Departing from conventional DOA estimation approaches, our network learns to query the available acoustic data, considering the detected head locations. We train and evaluate our algorithm on a realistic meetings dataset featuring up to 14 participants in the same meeting, overlapped speech, and other challenging scenarios.
Deep neural networks (DNNs) underpin many machine learning applications. Production quality DNN models achieve high inference accuracy by training millions of DNN parameters which has a significant resource footprint. This presents a challenge for resources operating at the extreme edge of the network, such as mobile and embedded devices that have limited computational and memory resources. To address this, models are pruned to create lightweight, more suitable variants for these devices. Existing pruning methods are unable to provide similar quality models compared to their unpruned counterparts without significant time costs and overheads or are limited to offline use cases. Our work rapidly derives suitable model variants while maintaining the accuracy of the original model. The model variants can be swapped quickly when system and network conditions change to match workload demand. This paper presents DNNShifter, an end-to-end DNN training, spatial pruning, and model switching system that addresses the challenges mentioned above. At the heart of DNNShifter is a novel methodology that prunes sparse models using structured pruning. The pruned model variants generated by DNNShifter are smaller in size and thus faster than dense and sparse model predecessors, making them suitable for inference at the edge while retaining near similar accuracy as of the original dense model. DNNShifter generates a portfolio of model variants that can be swiftly interchanged depending on operational conditions. DNNShifter produces pruned model variants up to 93x faster than conventional training methods. Compared to sparse models, the pruned model variants are up to 5.14x smaller and have a 1.67x inference latency speedup, with no compromise to sparse model accuracy. In addition, DNNShifter has up to 11.9x lower overhead for switching models and up to 3.8x lower memory utilisation than existing approaches.
Extended reality-enabled Internet of Things (XRI) provides the new user experience and the sense of immersion by adding virtual elements to the real world through Internet of Things (IoT) devices and emerging 6G technologies. However, the computational-intensive XRI tasks are challenging for the energy-constrained small-size XRI devices to cope with, and moreover certain data requires centralized computing that needs to be shared among users. To this end, we propose a cache-assisted space-air-ground integrated network mobile edge computing (SAGIN-MEC) system for XRI applications, consisting of two types of edge servers mounted on an unmanned aerial vehicle (UAV) and low Earth orbit (LEO) equipped with cache and the multiple ground XRI devices. For system efficiency, the four different offloading procedures of the XRI data are considered according to the type of information, i.e., shared data and private data, as well as the offloading decision and the caching status. Specifically, the private data can be offloaded to either UAV or LEO, while the offloading decision of the shared data to the LEO can be determined by the caching status. With the aim of maximizing the energy efficiency of the overall system, we jointly optimize UAV trajectory, resource allocation and offloading decisions under latency constraints and UAV's operational limitations by using the alternating optimization (AO)-based method along with Dinkelbach algorithm and successive convex optimization (SCA). Via numerical results, the proposed algorithm is verified to have the superior performance compared to conventional partial optimizations or without cache.
Computation offloading has become a popular solution to support computationally intensive and latency-sensitive applications by transferring computing tasks to mobile edge servers (MESs) for execution, which is known as mobile/multi-access edge computing (MEC). To improve the MEC performance, it is required to design an optimal offloading strategy that includes offloading decision (i.e., whether offloading or not) and computational resource allocation of MEC. The design can be formulated as a mixed-integer nonlinear programming (MINLP) problem, which is generally NP-hard and its effective solution can be obtained by performing online inference through a well-trained deep neural network (DNN) model. However, when the system environments change dynamically, the DNN model may lose efficacy due to the drift of input parameters, thereby decreasing the generalization ability of the DNN model. To address this unique challenge, in this paper, we propose a multi-head ensemble multi-task learning (MEMTL) approach with a shared backbone and multiple prediction heads (PHs). Specifically, the shared backbone will be invariant during the PHs training and the inferred results will be ensembled, thereby significantly reducing the required training overhead and improving the inference performance. As a result, the joint optimization problem for offloading decision and resource allocation can be efficiently solved even in a time-varying wireless environment. Experimental results show that the proposed MEMTL outperforms benchmark methods in both the inference accuracy and mean square error without requiring additional training data.
The prevalence of mobility impairments due to conditions such as spinal cord injuries, strokes, and degenerative diseases is on the rise globally. Lower-limb exoskeletons have been increasingly recognized as a viable solution for enhancing mobility and rehabilitation for individuals with such impairments. However, existing exoskeleton control systems often suffer from limitations such as latency, lack of adaptability, and computational inefficiency. To address these challenges, this paper introduces a novel online adversarial learning architecture integrated with edge computing for high-level lower-limb exoskeleton control. In the proposed architecture, sensor data from the user is processed in real-time through edge computing nodes, which then interact with an online adversarial learning model. This model adapts to the user's specific needs and controls the exoskeleton with minimal latency. Experimental evaluations demonstrate significant improvements in control accuracy and adaptability, as well as enhanced quality-of-service (QoS) metrics. These findings indicate that the integration of online adversarial learning with edge computing offers a robust and efficient approach for the next generation of lower-limb exoskeleton control systems.
Low-power event-based analog front-ends (AFE) are a crucial component required to build efficient end-to-end neuromorphic processing systems for edge computing. Although several neuromorphic chips have been developed for implementing spiking neural networks (SNNs) and solving a wide range of sensory processing tasks, there are only a few general-purpose analog front-end devices that can be used to convert analog sensory signals into spikes and interfaced to neuromorphic processors. In this work, we present a novel, highly configurable analog front-end chip, denoted as SPAIC (signal-to-spike converter for analog AI computation), that offers a general-purpose dual-mode analog signal-to-spike encoding with delta modulation and pulse frequency modulation, with tunable frequency bands. The ASIC is designed in a 180 nm process. It supports and encodes a wide variety of signals spanning 4 orders of magnitude in frequency, and provides an event-based output that is compatible with existing neuromorphic processors. We validated the ASIC for its functions and present initial silicon measurement results characterizing the basic building blocks of the chip.
The edge computing paradigm helps handle the Internet of Things (IoT) generated data in proximity to its source. Challenges occur in transferring, storing, and processing this rapidly growing amount of data on resource-constrained edge devices. Symbolic Representation (SR) algorithms are promising solutions to reduce the data size by converting actual raw data into symbols. Also, they allow data analytics (e.g., anomaly detection and trend prediction) directly on symbols, benefiting large classes of edge applications. However, existing SR algorithms are centralized in design and work offline with batch data, which is infeasible for real-time cases. We propose SymED - Symbolic Edge Data representation method, i.e., an online, adaptive, and distributed approach for symbolic representation of data on edge. SymED is based on the Adaptive Brownian Bridge-based Aggregation (ABBA), where we assume low-powered IoT devices do initial data compression (senders) and the more robust edge devices do the symbolic conversion (receivers). We evaluate SymED by measuring compression performance, reconstruction accuracy through Dynamic Time Warping (DTW) distance, and computational latency. The results show that SymED is able to (i) reduce the raw data with an average compression rate of 9.5%; (ii) keep a low reconstruction error of 13.25 in the DTW space; (iii) simultaneously provide real-time adaptability for online streaming IoT data at typical latencies of 42ms per symbol, reducing the overall network traffic.
Split learning is a privacy-preserving distributed learning paradigm in which an ML model (e.g., a neural network) is split into two parts (i.e., an encoder and a decoder). The encoder shares so-called latent representation, rather than raw data, for model training. In mobile-edge computing, network functions (such as traffic forecasting) can be trained via split learning where an encoder resides in a user equipment (UE) and a decoder resides in the edge network. Based on the data processing inequality and the information bottleneck (IB) theory, we present a new framework and training mechanism to enable a dynamic balancing of the transmission resource consumption with the informativeness of the shared latent representations, which directly impacts the predictive performance. The proposed training mechanism offers an encoder-decoder neural network architecture featuring multiple modes of complexity-relevance tradeoffs, enabling tunable performance. The adaptability can accommodate varying real-time network conditions and application requirements, potentially reducing operational expenditure and enhancing network agility. As a proof of concept, we apply the training mechanism to a millimeter-wave (mmWave)-enabled throughput prediction problem. We also offer new insights and highlight some challenges related to recurrent neural networks from the perspective of the IB theory. Interestingly, we find a compression phenomenon across the temporal domain of the sequential model, in addition to the compression phase that occurs with the number of training epochs.
In the era of 5G mobile communication, there has been a significant surge in research focused on unmanned aerial vehicles (UAVs) and mobile edge computing technology. UAVs can serve as intelligent servers in edge computing environments, optimizing their flight trajectories to maximize communication system throughput. Deep reinforcement learning (DRL)-based trajectory optimization algorithms may suffer from poor training performance due to intricate terrain features and inadequate training data. To overcome this limitation, some studies have proposed leveraging federated learning (FL) to mitigate the data isolation problem and expedite convergence. Nevertheless, the efficacy of global FL models can be negatively impacted by the high heterogeneity of local data, which could potentially impede the training process and even compromise the performance of local agents. This work proposes a novel solution to address these challenges, namely personalized federated deep reinforcement learning (PF-DRL), for multi-UAV trajectory optimization. PF-DRL aims to develop individualized models for each agent to address the data scarcity issue and mitigate the negative impact of data heterogeneity. Simulation results demonstrate that the proposed algorithm achieves superior training performance with faster convergence rates, and improves service quality compared to other DRL-based approaches.
Next-generation edge intelligence is anticipated to benefit various applications via offloading techniques. However, traditional offloading architectures face several issues, including heterogeneous constraints, partial perception, uncertain generalization, and lack of tractability. In this paper, we propose a Large AI Model-Based Offloading (LAMBO) framework with over one billion parameters for solving these problems. We first use input embedding (IE) to achieve normalized feature representation with heterogeneous constraints and task prompts. Then, we introduce a novel asymmetric encoder-decoder (AED) as the decision-making model, which is an improved transformer architecture consisting of a deep encoder and a shallow decoder for global perception and decision. Next, actor-critic learning (ACL) is used to pre-train the AED for different optimization tasks under corresponding prompts, enhancing the AED's generalization in multi-task scenarios. Finally, we propose an active learning from expert feedback (ALEF) method to fine-tune the decoder of the AED for tracking changes in dynamic environments. Our simulation results validate the advantages of the proposed LAMBO framework.
Binary neural networks (BNNs) have been widely adopted to reduce the computational cost and memory storage on edge-computing devices by using one-bit representation for activations and weights. However, as neural networks become wider/deeper to improve accuracy and meet practical requirements, the computational burden remains a significant challenge even on the binary version. To address these issues, this paper proposes a novel method called Minimum Spanning Tree (MST) compression that learns to compress and accelerate BNNs. The proposed architecture leverages an observation from previous works that an output channel in a binary convolution can be computed using another output channel and XNOR operations with weights that differ from the weights of the reused channel. We first construct a fully connected graph with vertices corresponding to output channels, where the distance between two vertices is the number of different values between the weight sets used for these outputs. Then, the MST of the graph with the minimum depth is proposed to reorder output calculations, aiming to reduce computational cost and latency. Moreover, we propose a new learning algorithm to reduce the total MST distance during training. Experimental results on benchmark models demonstrate that our method achieves significant compression ratios with negligible accuracy drops, making it a promising approach for resource-constrained edge-computing devices.
Over the past decade, deep learning models have exhibited considerable advancements, reaching or even exceeding human-level performance in a range of visual perception tasks. This remarkable progress has sparked interest in applying deep networks to real-world applications, such as autonomous vehicles, mobile devices, robotics, and edge computing. However, the challenge remains that state-of-the-art models usually demand significant computational resources, leading to impractical power consumption, latency, or carbon emissions in real-world scenarios. This trade-off between effectiveness and efficiency has catalyzed the emergence of a new research focus: computationally efficient deep learning, which strives to achieve satisfactory performance while minimizing the computational cost during inference. This review offers an extensive analysis of this rapidly evolving field by examining four key areas: 1) the development of static or dynamic light-weighted backbone models for the efficient extraction of discriminative deep representations; 2) the specialized network architectures or algorithms tailored for specific computer vision tasks; 3) the techniques employed for compressing deep learning models; and 4) the strategies for deploying efficient deep networks on hardware platforms. Additionally, we provide a systematic discussion on the critical challenges faced in this domain, such as network architecture design, training schemes, practical efficiency, and more realistic model compression approaches, as well as potential future research directions.
For multiple Unmanned-Aerial-Vehicles (UAVs) assisted Mobile Edge Computing (MEC) networks, we study the problem of combined computation and communication for user equipments deployed with multi-type tasks. Specifically, we consider that the MEC network encompasses both communication and computation uncertainties, where the partial channel state information and the inaccurate estimation of task complexity are only available. We introduce a robust design accounting for these uncertainties and minimize the total weighted energy consumption by jointly optimizing UAV trajectory, task partition, as well as the computation and communication resource allocation in the multi-UAV scenario. The formulated problem is challenging to solve with the coupled optimization variables and the high uncertainties. To overcome this issue, we reformulate a multi-agent Markov decision process and propose a multi-agent proximal policy optimization with Beta distribution framework to achieve a flexible learning policy. Numerical results demonstrate the effectiveness and robustness of the proposed algorithm for the multi-UAV-assisted MEC network, which outperforms the representative benchmarks of the deep reinforcement learning and heuristic algorithms.
Swarm robotics is envisioned to automate a large number of dirty, dangerous, and dull tasks. Robots have limited energy, computation capability, and communication resources. Therefore, current swarm robotics have a small number of robots, which can only provide limited spatio-temporal information. In this paper, we propose to leverage the mobile edge computing to alleviate the computation burden. We develop an effective solution based on a mobility-aware deep reinforcement learning model at the edge server side for computing scheduling and resource. Our results show that the proposed approach can meet delay requirements and guarantee computation precision by using minimum robot energy.
The digital twin edge network (DITEN) aims to integrate mobile edge computing (MEC) and digital twin (DT) to provide real-time system configuration and flexible resource allocation for the sixth-generation network. This paper investigates an intelligent reflecting surface (IRS)-aided multi-tier hybrid computing system that can achieve mutual benefits for DT and MEC in the DITEN. For the first time, this paper presents the opportunity to realize the network-wide convergence of DT and MEC. In the considered system, specifically, over-the-air computation (AirComp) is employed to monitor the status of the DT system, while MEC is performed with the assistance of DT to provide low-latency computing services. Besides, the IRS is utilized to enhance signal transmission and mitigate interference among heterogeneous nodes. We propose a framework for designing the hybrid computing system, aiming to maximize the sum computation rate under communication and computation resources constraints. To tackle the non-convex optimization problem, alternative optimization and successive convex approximation techniques are leveraged to decouple variables and then transform the problem into a more tractable form. Simulation results verify the effectiveness of the proposed algorithm and demonstrate the IRS can significantly improve the system performance with appropriate phase shift configurations. Moreover, the results indicate that the DT assisted MEC system can precisely achieve the balance between local computing and task offloading since real-time system status can be obtained with the help of DT.
Resource-constrained perception systems such as edge computing and vision-for-robotics require vision models to be both accurate and lightweight in computation and memory usage. While knowledge distillation is a proven strategy to enhance the performance of lightweight classification models, its application to structured outputs like object detection and instance segmentation remains a complicated task, due to the variability in outputs and complex internal network modules involved in the distillation process. In this paper, we propose a simple yet surprisingly effective sequential approach to knowledge distillation that progressively transfers the knowledge of a set of teacher detectors to a given lightweight student. To distill knowledge from a highly accurate but complex teacher model, we construct a sequence of teachers to help the student gradually adapt. Our progressive strategy can be easily combined with existing detection distillation mechanisms to consistently maximize student performance in various settings. To the best of our knowledge, we are the first to successfully distill knowledge from Transformer-based teacher detectors to convolution-based students, and unprecedentedly boost the performance of ResNet-50 based RetinaNet from 36.5% to 42.0% AP and Mask R-CNN from 38.2% to 42.5% AP on the MS COCO benchmark.
In light of the increasing adoption of edge computing in areas such as intelligent furniture, robotics, and smart homes, this paper introduces HyperSNN, an innovative method for control tasks that uses spiking neural networks (SNNs) in combination with hyperdimensional computing. HyperSNN substitutes expensive 32-bit floating point multiplications with 8-bit integer additions, resulting in reduced energy consumption while enhancing robustness and potentially improving accuracy. Our model was tested on AI Gym benchmarks, including Cartpole, Acrobot, MountainCar, and Lunar Lander. HyperSNN achieves control accuracies that are on par with conventional machine learning methods but with only 1.36% to 9.96% of the energy expenditure. Furthermore, our experiments showed increased robustness when using HyperSNN. We believe that HyperSNN is especially suitable for interactive, mobile, and wearable devices, promoting energy-efficient and robust system design. Furthermore, it paves the way for the practical implementation of complex algorithms like model predictive control (MPC) in real-world industrial scenarios.