AdaSpring

Recent breakthroughs in deep neural networks (DNNs) have fueled a tremendously growing demand for bringing DNN-powered intelligence into mobile platforms. While the potential of deploying DNNs on resource-constrained platforms has been demonstrated by DNN compression techniques, the current practice suffers from two limitations: 1) merely stand-alone compression schemes are investigated even though each compression technique only suit for certain types of DNN layers; and 2) mostly compression techniques are optimized for DNNs’ inference accuracy, without explicitly considering other application-driven system performance (e.g., latency and energy cost) and the varying resource availability across platforms (e.g., storage and processing capability). To this end, we propose AdaDeep, a usage-driven, automated DNN compression framework for systematically exploring the desired trade-off between performance and resource constraints, from a holistic system level. Specifically, in a layer-wise manner, AdaDeep automatically selects the most suitable combination of compression techniques and the corresponding compression hyperparameters for a given DNN. Thorough evaluations on six datasets and across twelve devices demonstrate that <inline-formula><tex-math notation="LaTeX">${\sf AdaDeep}$</tex-math></inline-formula> can achieve up to <inline-formula><tex-math notation="LaTeX">$18.6\times$</tex-math></inline-formula> latency reduction, <inline-formula><tex-math notation="LaTeX">$9.8\times$</tex-math></inline-formula> energy-efficiency improvement, and <inline-formula><tex-math notation="LaTeX">$37.3\times$</tex-math></inline-formula> storage reduction in DNNs while incurring negligible accuracy loss. Furthermore, <inline-formula><tex-math notation="LaTeX">${\sf AdaDeep}$</tex-math></inline-formula> also uncovers multiple novel combinations of compression techniques.

With the proliferation of IoT and edge devices, massive amounts of heterogeneous data are generated in real time. Efficient compression?decompression mechanisms are essential to reduce storage overhead, minimize transmission latency, and optimize energy consumption. However, traditional compression algorithms lack adaptability to dynamic environments where resource constraints and workload patterns vary rapidly. This research proposes a Deep Reinforcement Learning (DRL)-based adaptive framework for real-time compression?decompression optimization in edge devices. The agent dynamically selects compression levels, algorithms, and bit allocation strategies based on current device constraints (CPU, memory, bandwidth, and energy) and application-specific quality requirements. Experimental evaluations show that the DRL-based approach achieves up to 40% reduction in latency and 30% improvement in energy efficiency, while maintaining near-lossless reconstruction accuracy compared to state-of-the-art baselines.

Recent breakthroughs in deep learning (DL) have led to the emergence of many intelligent mobile applications and services, but in the meanwhile also pose unprecedented computing challenges on resource-constrained mobile devices. This paper builds a collaborative deep inference system between a resource-constrained mobile device and a powerful edge server, aiming at joining the power of both on-device processing and computation offloading. The basic idea of this system is to partition a deep neural network (DNN) into a front-end part running on the mobile device and a back-end part running on the edge server, with the key challenge being how to locate the optimal partition point to minimize the end-to-end inference delay. Unlike existing efforts on DNN partitioning that rely heavily on a dedicated offline profiling stage to search for the optimal partition point, our system has a built-in online learning module, called Autodidactic Neurosurgeon (ANS), to automatically learn the optimal partition point on-the-fly. Therefore, ANS is able to closely follow the changes of the system environment by generating new knowledge for adaptive decision making. The core of ANS is a novel contextual bandit learning algorithm, called μLinUCB, which not only has provable theoretical learning performance guarantee but also is ultra-lightweight for easy real-world implementation. We implement our system on a video stream object detection testbed to validate the design of ANS and evaluate its performance. The experiments show that ANS significantly outperforms state-of-the-art benchmarks in terms of tracking system changes and reducing the end-to-end inference delay.

The growing demand for efficient and reliable sensor networks in applications such as environmental monitoring, healthcare, and smart cities has highlighted the need for optimizing communication protocols, power consumption, and data management. Traditional methods often focus on optimizing individual layers of the network without considering the interactions across layers. This approach limits the overall performance, especially as networks scale in size and complexity. The lack of effective cross-layer optimization strategies hinders the performance of sensor networks, leading to suboptimal energy usage, low throughput, and high latency. Moreover, sensor nodes in such networks often face constraints such as limited energy, computation power, and memory, making traditional optimization methods inadequate for achieving high efficiency across all layers of the network. This paper proposes a cross-layer design and optimization framework leveraging Artificial Intelligence (AI) and Deep Learning (DL) algorithms, specifically Recurrent Neural Networks (RNN) and Deep Belief Networks (DBN), to address these challenges. The RNN is employed to model the temporal dependencies in the sensor data, while the DBN is used for optimizing decision-making processes across multiple network layers. The proposed framework dynamically adjusts routing, data aggregation, and power control parameters based on real-time conditions, improving overall network performance. Simulation results demonstrate that the proposed approach outperforms traditional cross-layer design methods. The RNN-DBN framework achieved a 35% improvement in energy efficiency, a 25% reduction in latency, and a 40% increase in data throughput compared to existing optimization techniques. These enhancements are particularly significant in large-scale, real-time sensor networks.

Recent advances have demonstrated the efficacy of deep learning models for mobile biometrics. In this paper, we evaluate and compare different deep learning models in terms of their efficiency and accuracy when used for biometric user authentication on mobile devices. To this aim, well-known pre-trained architectures such as VGG, ResNet, DenseNet and models proposed for mobile vision applications such as MobileNetVl, MobileNetV2, and NasNet-mobile, are compared with respect to their matching performance and computational cost. Further, we propose a compact and fast custom deep learning model for efficient operations on resource-constrained mobile devices. Experimental investigations were performed as a case study on mobile ocular biometrics using a large-scale VISOB dataset. Reported results suggest the best trade-off between performance and computational cost by our custom model over existing deep learning architectures.

Test cost has become a critical issue for large industrial integrated circuits. Various test compression techniques have been adopted in the industry to reduce test cost. However, appropriate input and output channel counts must be selected to utilize the test compression technology best. This paper presents an efficient and effective method to estimate the test pattern counts under different compression configurations for the Embedded Deterministic Test (EDT) compression technique. In searching for the accurate estimation method, we build mathematical models that reveal the internal relationship among different compression configurations. The models are established based on novel theoretical analysis as well as actual experimental data. Accurate estimation of test pattern counts for a wide range of compression configurations can be obtained based on the results of only two ATPG runs. Experimental results on nine industrial circuits show that the average error rate of pattern count estimation is about 5%, with very few outliers. With the proposed method, a test compression designer can easily pick the best input and output channel configuration to fit the design needs.

This research aims at analyzing how IoT and Deep Learning can be used to improve real-time financial and traffic control in autonomous vehicle systems with special reference to the marketing optimization. The offered common framework for routing makes use of Dynamic Route Optimization using Deep Reinforcement Learning (DRL) for overcoming advanced real-time routing issues optimized. Apache Kafka is used for efficient data streaming so as to enhance the interoperation of Internet of things sensors and automobile systems where as TensorFlow is an perfect platform for deep learning model execution. The methodology also places significant emphasis on minimizing response time and achieving capability and capacity and supporting large-scale environments. Using metrics such as accuracy, latency reduction, and the reduced amount of fuel consumed, we show the efficiency of using the DRL-based approach rather than heuristic- and machine learning-based approaches. The research shows a revolution in the operation efficiency and marketing applications of autonomous systems.

© The Institution of Engineering and Technology 2014. The full search scheme employed in H.264/AVC significantly improves the coding performance, but it also introduces a very high computational complexity which limits the applications in resource-constrained mobile devices. In this study, the authors firstly present a discretisation total variation and orientation gradient-based hierarchical intra-prediction mode decision method for mobile video applications. By shrinking the candidate mode set in the rate-distortion optimisation (RDO) process, the proposed algorithm reduces the computational complexity and power consumption of the encoder. Furthermore, they extend the hierarchical algorithm to a complexity scalable version in which the coding complexity is measured on five levels by reserving various numbers of modes for RDO. Experimental results demonstrate that the proposed mode decision algorithm reduces the coding complexity significantly with negligible performance degradation and the proposed complexity scalable algorithm is effective and efficient for mobile video application.

An input-centric performance model for computational offloading of mobile applications

Power normalized cepstral robust features of deep neural networks in a cloud computing data privacy protection scheme

In recent years, dynamic visual SLAM techniques have been widely used in autonomous navigation, augmented reality, and virtual reality. However, the increasing demand for computational resources by SLAM techniques limits its application on resource-constrained mobile devices. MEC technology combined with 5G ultra-dense networks enables complex computational tasks in visual SLAM systems to be offloaded to edge computing servers, thus breaking the resource constraints of terminals and meeting real-time computing requirements. This paper firstly introduces the research results in the field of visual SLAM in detail through three categories: static SLAM, dynamic SLAM, and SLAM techniques combined with deep learning. Secondly, the three major parts of the technology comparison between mobile edge computing and mobile cloud computing, 5G ultra-dense networking technology, and MEC and UDN integration technology are introduced to sort out the basic technologies related to the application of 5G ultra-dense network to offload complex computing tasks from visual SLAM systems to edge computing servers.

While deep learning has achieved great success in building vision applications for mainstream users, there is relatively less work for the blind and visually impaired to have a personal, on-device visual assistant for their daily life. Unlike mainstream applications, vision system for the blind must be robust, reliable and safe-to-use. In this paper, we propose a fine-grained currency recognizer based on CONGAS, which significantly surpasses other popular local features by a large margin. In addition, we introduce an effective and light-weight coarse classifier that gates the fine-grained recognizer on resource-constrained mobile devices. The coarse-to-fine approach is orchestrated to provide an extensible mobile-vision architecture, that demonstrates how the benefits of coordinating deep learning and local feature based methods can help in resolving a challenging problem for the blind and visually impaired. The proposed system runs in real-time with ~150ms latency on a Pixel device, and achieved 98% precision and 97% recall on a challenging evaluation set.

Real and apparent age estimation of human face has attracted increased attention due to its numerous real-world applications. Different intelligent application scenarios can benefit from these computer-based systems that predict the ages of people correctly. Automatic apparent age system is particularly useful in medical diagnosis, facial beauty product development, movie role casting, the effect of plastic surgery, and anti-aging treatment. Predicting the real and apparent age of people has been quite difficult for both machines and humans. More recently, Deep learning with Convolutional Neural Networks (CNNs) methods have been extensively used for these classification task. It has incomparable advantages in extracting discriminative image features from human faces. However, many of the existing CNN-based methods are designed to be deeper and larger with more complex layers that makes it challenging to deploy on mobile devices with resource-constrained features. Therefore, we design a lightweight CNN model of fewer layers to estimate the real and apparent age of individuals from unconstrained real-time face images that can be deployed on mobile devices. The experimental results, when analyzed for classification accuracy on FG-NET, MORPH-II and APPA-REAL, with large-scale face images containing both real and apparent age annotations, show that our model obtains a state-of-the-art performance in both real and apparent age classification when compared to state-of-the-art methods. The new results and model size, therefore, confirm the usefulness of the model on resource-constrained mobile devices.

Due to development in cloud services, lots of data is transferred between users and servers of the cloud. This transmission of data consumes huge amounts of energy. This energy consumption occurs during the operation of network infrastructure, the conversion of electrical to optical signals to travel long distances, and signal amplification. As Green computing is the use of computing devices in an environmentally friendly way, i.e., using electrical energy efficiently as Data centers require a significant amount of electricity to operate and cool the servers, leading to carbon emissions from the burning of fossil fuels. Green computing in cloud services is about optimizing energy consumption and by incorporating deep learning algorithms, we can enhance the energy efficiency of cloud infrastructure. These algorithms can analyze real-time data from sensors, optimize resource allocation, and dynamically adjust power usage. Through intelligent workload scheduling, server consolidation, and power management, deep learning enables the reduction of energy waste and carbon emissions. The integration of deep learning in cloud services not only improves energy efficiency but also enhances performance and cost-effectiveness. Here we are using a deep learning model which can be used for workload prediction and resource provisioning. By analyzing historical workload patterns and user behavior, deep learning algorithmscan predict future resource demands and allocate resources accordingly, leading to more efficient resource utilization and energy savings

We introduce a smart multi-objective optimization framework towards the performance enhancement of the 4G LTE network using an integrated Artificial Intelligence (AI) and Whale Optimization Algorithm (WOA) setup. This approach aims to integrate tested regression-based KPI prediction along with decision-variable-based meta-heuristic optimization in a reproducible experimental environment within MATLAB R2023b. A weighted multi-objective formulation for optimizing the performance of 4 key indicators, i.e., throughput, latency, packet loss ratio, and energy efficiency, is jointly executed. By comparison against existing baseline LTE configuration, we observe a 25.3% throughput improvement, 28.9% reduction in latency, 38.7% reduction in packet loss ratio, and 23.4% improvement in energy efficiency. We validate robustness with weight sensitivity analysis (±10% and ±20%), population scalability testing (N = 20, 30, 50), decay-strategy comparison (linear vs. exponential), and Pareto-front approximation (2,960 non-dominated solutions from 3,000 samples). The above results corroborate that, after approximately 70 iterations, convergence is stable and performance is consistent over the span of 30 independent runs. Thus, the proposed AI–WOA construction offers an organized and replicable process to fine-tune LTE performance optimization, a basis for downstream implementations towards a more distributed and next-generation wireless environment.