A FPGA-based neural accelerator for small IoT devices

The Internet of Things (IoT) has had a considerable influence on our daily lives by enabling enhanced connection of devices, systems, and services that extends beyond machine-to-machine interactions and encompasses a wide range of protocols, domains, and applications. However, despite privacy concerns shown by IoT users, little has been done to reduce and protect individual information exposure. It is extremely difficult to mitigate IoT devices from reidentification threats which is why it is still a major challenge for IoT users to securely protect their information. The trust controls how we regulate privacy in our IoT platforms in the same way that it governs personal relationships. As IoT devices become increasingly linked, more data is shared across individuals, businesses, governments, and ecosystems. Technologies, sensors, machines, data, and cloud connections all rely largely on trust relationships that have been formed. With the rapid growth of additional types of IoT devices that are being introduced, it, therefore, expands privacy concerns and is difficult to develop trust with an IoT system or device without the option to regulate information privacy settings. Privacy has always been a barrier for many devices as they race for the early adoption of IoT technologies. Several Internet of Things devices or systems will continue to pose privacy threats. As a result, the main objective of this study was to examine the individual understanding of privacy and to promote information privacy protection awareness not only to IoT users but also to organizations that use IoT devices or platforms to run their day-to-day business operations. Furthermore, the objective extends to compare user knowledge and concerns about IoT privacy, as well as to identify any common attitudes and variances. However, in terms of enhancing individuals’ knowledge, an artifact was developed to educate and enhance information privacy awareness among IoT users. A pre- and postquestionnaire was generated to test and validate user knowledge regarding information privacy protection in IoT. The study was conducted using a quantitative research method. Findings indicate that IoT users’ awareness of information privacy protection turned out to be average, suggesting a need for education and awareness. Several participants stated that information privacy protection awareness is required within the community to educate, raise awareness, eliminate human error, and enable individuals to be conscious of their privacy when surfing the Internet.

Internet of Things (IoT) technology has been incorporated into the majority of people’s everyday lives and places of employment due to the quick development in information technology. Modern agricultural techniques increasingly use the well-known and superior approach of managing a farm known as “smart farming”. Utilizing a variety of information and agricultural technologies, crops are observed for their general health and productivity. This requires monitoring the condition of field crops and looking at many other indicators. The goal of smart agriculture is to reduce the amount of money spent on agricultural inputs while keeping the quality of the final product constant. The Internet of Things (IoT) has made smart agriculture possible through data collection and storage techniques. For example, modern irrigation systems use effective sensor networks to collect field data for the best plant irrigation. Smart agriculture will become more susceptible to cyber-attacks as its reliance on the IoT ecosystem grows, because IoT networks have a large number of nodes but limited resources, which makes security a difficult issue. Hence, it is crucial to have an intrusion detection system (IDS) that can address such challenges. In this manuscript, an IoT-based privacy-preserving anomaly detection model for smart agriculture has been proposed. The motivation behind this work is twofold. Firstly, ensuring data privacy in IoT-based agriculture is of the utmost importance due to the large volumes of sensitive information collected by IoT devices, including on environmental conditions, crop health, and resource utilization data. Secondly, the timely detection of anomalies in smart agriculture systems is critical to enable proactive interventions, such as preventing crop damage, optimizing resource allocation, and ensuring sustainable farming practices. In this paper, we propose a privacy-encoding-based enhanced deep learning framework for the difficulty of data encryption and intrusion detection. In terms of data encoding, a novel method of a sparse capsule-auto encoder (SCAE) is proposed along with feature selection, feature mapping, and feature normalization. An SCAE is used to convert information into a new encrypted format in order to prevent deduction attacks. An attention-based gated recurrent unit neural network model is proposed to detect the intrusion. An AGRU is an advanced version of a GRU which is enhanced by an attention mechanism. In the results section, the proposed model is compared with existing deep learning models using two public datasets. Parameters such as recall, precision, accuracy, and F1-score are considered. The proposed model has accuracy, recall, precision, and F1-score of 99.9%, 99.7%, 99.9%, and 99.8%, respectively. The proposed method is compared using a variety of machine learning techniques such as the deep neural network (DNN), convolutional neural network (CNN), recurrent neural network (RNN), and long short-term memory (LSTM).

<p>Deep neural networks have achieved tremendous success in many domains (e.g., computer vision~\cite{Alexnet12,vggnet15,fastrcnn15}, speech recognition~\cite{hinton2012deep,dahl2012context}, natural language processing~\cite{dahl2012context,collobert2011natural}, games~\cite{silver2017mastering,silver2016mastering}), however, there are still many challenges in deep learning comunity such as how to speed up training large deep neural networks, how to compress large nerual networks for mobile/embed device without performance loss, how to automatically design the optimal network structures for a certain task, and how to further design the optimal networks with improved performance and certain model size with reduced computation cost.</p> <p>To speed up training large neural networks, we propose to use multinomial sampling for dropout, i.e., sampling features or neurons according to a multinomial distribution with different probabilities for different features/neurons. To exhibit the optimal dropout probabilities, we analyze the shallow learning with multinomial dropout and establish the risk bound for stochastic optimization. By minimizing a sampling dependent factor in the risk bound, we obtain a distribution-dependent dropout with sampling probabilities dependent on the second order statistics of the data distribution. To tackle the issue of evolving distribution of neurons in deep learning, we propose an efficient adaptive dropout (named evolutional dropout) that computes the sampling probabilities on-the-fly from a mini-batch of examples.</p> <p>To compress large neural network structures, we propose a simple yet powerful method for compressing the size of deep Convolutional Neural Networks (CNNs) based on parameter binarization. The striking difference from most previous work on parameter binarization/quantization lies at different treatments of $1\times 1$ convolutions and $k\times k$ convolutions ($k>1$), where we only binarize $k\times k$ convolutions into binary patterns. By doing this, we show that previous deep CNNs such as GoogLeNet and Inception-type Nets can be compressed dramatically with marginal drop in performance. Second, in light of the different functionalities of $1\times 1$ (data projection/transformation) and $k\times k$ convolutions (pattern extraction), we propose a new block structure codenamed the pattern residual block that adds transformed feature maps generated by $1\times 1$ convolutions to the pattern feature maps generated by $k\times k$ convolutions, based on which we design a small network with $\sim 1$ million parameters. Combining with our parameter binarization, we achieve better performance on ImageNet than using similar sized networks including recently released Google MobileNets.</p> <p>To automatically design neural networks, we study how to design a genetic programming approach for optimizing the structure of a CNN for a given task under limited computational resources yet without imposing strong restrictions on the search space. To reduce the computational costs, we propose two general strategies that are observed to be helpful: (i) aggressively selecting strongest individuals for survival and reproduction, and killing weaker individuals at a very early age; (ii) increasing mutation frequency to encourage diversity and faster evolution. The combined strategy with additional optimization techniques allows us to explore a large search space but with affordable computational costs.</p> <p>To further design the optimal networks with improved performance and certain model size under reduced computation cost, we propose an ecologically inspired genetic approach for neural network structure search , that includes two types of succession: primary and secondary succession as well as accelerated extinction. Specifically, we first use primary succession to rapidly evolve a community of poor initialized neural network structures into a more diverse community, followed by a secondary succession stage for fine-grained searching based on the networks from the primary succession. Accelerated extinction is applied in both stages to reduce computational cost. In addition, we also introduce the gene duplication to further utilize the novel block of layers that appeared in the discovered network structure.</p>

The Internet of Things (IoT) platform consists of numerous IoT devices and a small number ofservers. In order for the platform to operate efficiently and reliably, servers need to have completecontrol over many IoT devices. The advantage of using network boot in this environment is thatthe server has control over the device’s boot process. When applying network boot on an IoT platform,the server needs to be able to handle the network boot that many IoT devices request. If theserver does not have enough processing power for the network boot request, IoT device boot wouldbe delayed. In this paper, we propose a resource-efficient network boot to solve this problem. Theproposed framework has three mechanism. First, the server reserves the right to control which systemimage to transfer during the network boot process on the platform. Second, the server appliesdeduplication to the system image to minimize the amount of data required for transmission. Third,the server monitors network boot requests and performs file transfers appropriate to the situation,minimizing time spent on transfers. The proposed system allows the server to efficiently control numerousIoT devices when using network boot on the IoT platform.

IoT devices rely on authentication mechanisms to render secure message exchange. During data transmission, scalability, data integrity, and processing time have been considered challenging aspects for a system constituted by IoT devices. The application of physical unclonable functions (PUFs) ensures secure data transmission among the internet of things (IoT) devices in a simplified network with an efficient time-stamped agreement. This paper proposes a secure, lightweight, cost-efficient reinforcement machine learning framework (SLCR-MLF) to achieve decentralization and security, thus enabling scalability, data integrity, and optimized processing time in IoT devices. PUF has been integrated into SLCR-MLF to improve the security of the cluster head node in the IoT platform during transmission by providing the authentication service for device-to-device communication. An IoT network gathers information of interest from multiple cluster members selected by the proposed framework. In addition, the software-defined secured (SDS) technique is integrated with SLCR-MLF to improve data integrity and optimize processing time in the IoT platform. Simulation analysis shows that the proposed framework outperforms conventional methods regarding the network’s lifetime, energy, secured data retrieval rate, and performance ratio. By enabling the proposed framework, number of residual nodes is reduced to 16%, energy consumption is reduced by up to 50%, almost 30% improvement in data retrieval rate, and network lifetime is improved by up to 1000 msec.

Nowadays deep neural networks are widely used to accurately classify input data. An interesting application area is the Internet of Things (IoT), where a massive amount of sensor data has to be classified. The processing power of the cloud is attractive, however the variable latency imposes a major drawback in situations where near real-time classification is required. In order to exploit the apparent trade-off between utilizing the stable but limited embedded computing power of IoT devices and the seemingly unlimited computing power of Cloud computing at the cost of higher and variable latency, we propose a Big-Little architecture for deep neural networks. A small neural network trained to a subset of prioritized output classes is running on the embedded device, while a more specific classification is calculated when required by a large neural network in the cloud. We show the applicability of this concept in the IoT domain by evaluating our approach for state of the art neural network classification problems on popular embedded devices such as the Raspberry Pi and Intel Edison.

Using low-rank adaptation to fine-tune pre-trained neural network models has attracted widespread attention due to its advantages of low resource requirements, high precision, and no additional inference delay. However, most existing methods are designed for large language models based on the Transformer structure and lack adaptation to convolutional neural networks (CNN) widely used on Internet of Things (IoT) devices. In addition, IoT devices are usually deployed outdoors and collect large amounts of data that are affected by environmental conditions. Providing a highly robust CNN model is a prerequisite for providing high-quality services. To this end, this paper proposes a highly robust seed-filter adaptation method (SeFA) for pre-trained CNNs. SeFA introduces an adaptation branch with the same structure as the backbone network. In the adaptation branch, some filters are first designated seed filters and grouped. Then, additional filters are generated from the grouped seed filters and nonlinear transformation functions (NLFs) with different hyperparameters. The seed filters' parameters are updated during model training, and the NLF hyperparameters are randomly initialized and frozen. Both grouping seed filters and configuring NLFs with non-learnable hyperparameters can improve the model's robustness. This is because grouping seed filters can generate diverse filters on demand without increasing the model's complexity, and the NLFs' rules can regularize the model. It is worth mentioning that the number of fine-tuning parameters of the pre-trained model that can adapt to downstream tasks can be flexibly controlled by specifying the number of seed filters. The key idea of this paper is to propose SeFA with flexible, controllable, learnable parameters, high robustness, and adaptability to pretrained CNN models, thereby facilitating fine-tuning on resource-constrained IoT devices and providing highly robust visual services. Experimental results on the CIFAR-10, CIFAR-10-C, CIFAR-100, CIFAR-100-C, and Icons-50 datasets demonstrate that the proposed SeFA outperforms other state-of-the-art methods. Specifically, on the ResNet-152 model and the CIFAR-10-C dataset, the accuracy of our SeFA is about 7% higher than that of the full fine-tuning method. In addition, we verify the efficiency of SeFA-based models for fine-tuning and inference on the device.

Self-mixing interferometry (SMI) is superior to other laser interferometry methods due to its simplicity and compactness. However, SMI signals are often complex and difficult to process due to interference in the form of variations in the effective reflectivity of the target, noisy signals, complex signal shapes, and other dependencies. Deep neural networks have been a very popular area of research in computer artificial intelligence in recent years, allowing more implicit features to be uncovered than traditional shallow machine learning. It has been shown that convolutional and back-propagation neural networks can be used for SMI signal processing. There are also studies that have used machine learning genetic algorithms for absolute distance measurement. Based on the above, this study used convolutional neural networks to form a deep neural network for absolute distance measurement based on SMI technology. We first trained the deep convolutional neural network at different feedback strengths. The results of the Convolutional Neural Network (CNN) model showed a coefficient of determination of 0.9987. which is consistent with the required model. The trained network was then used to estimate absolute distances with and without the addition of noise. The comparison proves that the proposed method is noise-proof and has high adaptability for measurements under different conditions.

<abstract> <p>With the widespread use of Internet, Internet of Things (IoT) devices have exponentially increased. These devices become vulnerable to malware attacks with the enormous amount of data on IoT devices; as a result, malware detection becomes a major problem in IoT devices. A reliable and effective mechanism is essential for malware detection. In recent years, research workers have developed various techniques for the complex detection of malware, but accurate detection continues to be a problem. Ransomware attacks pose major security risks to corporate and personal information and data. The owners of computer-based resources can be influenced by monetary losses, reputational damage, and privacy and verification violations due to successful assaults of ransomware. Therefore, there is a need to swiftly and accurately detect the ransomware. With this motivation, the study designs an Ebola optimization search algorithm for enhanced deep learning-based ransomware detection (EBSAEDL-RD) technique in IoT security. The purpose of the EBSAEDL-RD method is to recognize and classify the ransomware to achieve security in the IoT platform. To accomplish this, the EBSAEDL-RD technique employs min-max normalization to scale the input data into a useful format. Also, the EBSAEDL-RD technique makes use of the EBSA technique to select an optimum set of features. Meanwhile, the classification of ransomware takes place using the bidirectional gated recurrent unit (BiGRU) model. Then, the sparrow search algorithm (SSA) can be applied for optimum hyperparameter selection of the BiGRU model. The wide-ranging experiments of the EBSAEDL-RD approach are performed on benchmark data. The obtained results highlighted that the EBSAEDL-RD algorithm reaches better performance over other models on IoT security.</p> </abstract>

Feature data processing: Making medical data fit deep neural networks

The exponential growth of Internet of Things (IoT) and 5G networks has resulted in maximum users, and the role of cognitive radio has become pivotal in handling the crowded users. In this scenario, cognitive radio techniques such as spectrum sensing, spectrum sharing and dynamic spectrum access will become essential components in Wireless IoT communication. IoT devices must learn adaptively to the environment and extract the spectrum knowledge and inferred spectrum knowledge by appropriately changing communication parameters such as modulation index, frequency bands, coding rate etc., to accommodate the above characteristics. Implementing the above learning methods on the embedded chip leads to high latency, high power consumption and more chip area utilisation. To overcome the problems mentioned above, we present DEEP HOLE Radio systems, the intelligent system enabling the spectrum knowledge extraction from the unprocessed samples by the optimized deep learning models directly from the Radio Frequency (RF) environment. DEEP HOLE Radio provides (i) an optimized deep learning framework with a good trade-off between latency, power and utilization. (ii) Complete Hardware-Software architecture where the SoC’s coupled with radio transceivers for maximum performance. The experimentation has been carried out using GNURADIO software interfaced with Zynq-7000 devices mounting on ESP8266 radio transceivers with inbuilt Omni directional antennas. The whole spectrum of knowledge has been extracted using GNU radio. These extracted features are used to train the proposed optimized deep learning models, which run parallel on Zynq-SoC 7000, consuming less area, power, latency and less utilization area. The proposed framework has been evaluated and compared with the existing frameworks such as RFLearn, Long Term Short Memory (LSTM), Convolutional Neural Networks (CNN) and Deep Neural Networks (DNN). The outcome shows that the proposed framework has outperformed the existing framework regarding the area, power and time. Moreover, the experimental results show that the proposed framework decreases the delay, power and area by 15%, 20% 25% concerning the existing RFlearn and other hardware constraint frameworks.

Internet of Things (IoT) platforms enable users to deploy home automation applications. Meanwhile, privacy issues arise as large amounts of sensitive device data flow out to IoT platforms. Most of the data flowing out to a platform actually do not trigger automation actions, while homeowners currently have no control once devices are bound to the platform. We present PFirewall, a customizable data-flow control system to enhance the privacy of IoT platform users. PFirewall automatically generates data-minimization policies, which only disclose minimum amount of data to fulfill automation. In addition, PFirewall provides interfaces for homeowners to customize individual privacy preferences by defining user-specified policies. To enforce these policies, PFirewall transparently intervenes and mediates the communication between IoT devices and the platform, without modifying the platform, IoT devices, or hub. Evaluation results on four real-world testbeds show that PFirewall reduces IoT data sent to the platform by 97% without impairing home automation, and effectively mitigates user-activity inference/tracking attacks and other privacy risks.

With the emergence of cloud computing, large amounts of private data are stored and processed in the cloud. On the other hand, data owners (users) may not want to reveal data information to cloud providers to protect their privacy. Therefore, users may upload encrypted data to the cloud or third-party platforms, such as Google Cloud, Amazon Web Service, and Microsoft Azure. Conventionally, data must be decrypted before being analyzed in the cloud, which raises privacy concerns. Moreover, decryption of big data such as images requires enormous computation resources, which is unsuitable for energy-constrained devices, particularly Internet of Things (IoT) devices. Data privacy in most popular applications, such as image query or classification, can be preserved if encrypted images can be directly classified on the cloud or IoT devices without decryption. This paper proposes a high-speed double random phase encoding (DRPE) technique of encrypting images into white-noise images. DRPE-encrypted images are then uploaded and stored in the cloud. Images that are encrypted without being decrypted are classified using deep convolutional neural networks in the cloud. The simulation results indicated the feasibility and good performance of the proposed approach. The proposed privacy-preserving image classification method can be useful in data-sensitive fields, such as medicine and transportation.

In this paper, we tackle the problem of deploying face recognition (FR) solutions in heterogeneous Internet of Things (IoT) platforms. The main challenges are the optimal deployment of deep neural networks (DNNs) in the high variety of IoT devices (e.g., robots, tablets, smartphones, etc.), the secure management of biometric data while respecting the users’ privacy, and the design of appropriate user interaction with facial verification mechanisms for all kinds of users. We analyze different approaches to solving all these challenges and propose a knowledge-driven methodology for the automated deployment of DNN-based FR solutions in IoT devices, with the secure management of biometric data, and real-time feedback for improved interaction. We provide some practical examples and experimental results with state-of-the-art DNNs for FR in Intel’s and NVIDIA’s hardware platforms as IoT devices.

The rapid development in the wearable sensors helped develop the healthcare-based Internet of Things (IoT) platforms for continuous monitoring of chronic disease. As per the World Health Organization (WHO) 85% of the Cardiovascular Disease (CVD) is preventable with proper monitoring for symptoms and awareness. The electrocardiogram (ECG) signal refers to a recording of electrical activity that is generated by the human heart. An analysis of the ECG signal helps in building models that can make an automatic classification of various CVDs. Deep neural network is found to be efficient to make an automatic classification of the primary signals of the ECG. These ECG signals to the arrhythmia classes were grouped by using the Convolutional Neural Networks (CNN). The 9 Layers of the CNN techniques are employed in state-of-the-art modern research to classify the signal. The Tabu Search (TS) algorithm with the CNN, Hybrid Optimization Algorithm known as the CNN with ABC-TS, and CNN with Artificial Bee Colony (ABC) is used to optimize the CNN in this work for classifying ECG signals. An evaluation of the technique was made on the basis of metrics of performance evaluation that are capable of producing enhanced outcomes.