Convolutional neural networks(CNNs) achieve high accuracy on large datasets but requires significant computation and storage requirement for training/testing. While many applications demand low latency and energy-efficient processing of the images, deploying these complex algorithms on the hardware is a challenging task. This dissertation first presents a compiler-based CNN training accelerator using DDR3 and HBM2 memory. An optimized RTL library is implemented to perform training-specific tasks and an RTL compiler is developed to generate FPGA-synthesizable RTL based on user-defined constraints. High Bandwidth Memory(HBM) provides efficient off-chip communication and improves the training performance. The impact of HBM2 on CNN training workloads is analyzed and compressively compared with DDR3. For training ResNet-20/VGG-like CNNs for the CIFAR-10 dataset, the proposed CNN training accelerator on Stratix-10 GX FPGA(DDR3) demonstrates 479 GOPS performance, and on Stratix-10 MX FPGA(HBM) shows 4.5/9.7 X energy-efficiency improvement compared to Tesla V100 GPU. Next, the FPGA online learning accelerator is presented. Adopting model segmentation techniques from Progressive Segmented Training(PST), the online learning accelerator achieved a 4.2X reduction in training latency. Furthermore, this dissertation presents an 8-bit floating-point (FP8) training processor which implements (1) Highly parallel tensor cores that maintain high PE utilization, (2) Hardware-efficient channel gating for dynamic output activation sparsity (3) Dynamic weight sparsity based on group Lasso (4) Gradient skipping based on FP prediction error. The 28nm prototype chip demonstrates significant improvements in FLOPs reduction (7.3×), energy efficiency (16.4 TFLOPS/W), and overall training latency speedup (4.7×) for both supervised training and self-supervised training tasks. In addition to the training accelerators, this dissertation also presents a CNN inference accelerator on ASIC(FixyNN) and FPGA(FixyFPGA). FixyNN consists of a fixed-weight feature extractor that generates ubiquitous CNN features and a conventional programmable CNN accelerator. In the fixed-weight feature extractor, the network weights are hard-coded into hardware and used as a fixed operand for the multiplication. Experimental results demonstrate FixyNN can achieve very high energy efficiencies up to 26.6 TOPS/W, and FixyFPGA achieves $2.34\times$ higher GOPS on ImageNet classification. In summary, this dissertation comprehensively discusses novel architectures of high-performance and energy-efficient ASIC/FPGA CNN inference/training accelerators.

Edge computing applications have recently gained prominence as the world of internet-of-things becomes increasingly embedded into people's lives. Performing computations at the edge addresses multiple issues, such as memory bandwidth-latency bottlenecks, exposure of sensitive data to external attackers, etc. It is important to protect the data collected and processed by edge devices, and also to prevent unauthorized access to such data. It is also important to ensure that the computing hardware fits well within the tight energy and area budgets for the edge devices which are being progressively scaled-down in size. Firstly, a novel low-power smart security prototype chip that combines multiple entropy sources, such as real-time electrocardiogram (ECG) data, and SRAM-based physical unclonable functions (PUF), for authentication and cryptography applications is proposed. Up to ~12X improvement in the equal error rate compared to a prior ECG-only authentication system is achieved by combining feature vectors obtained from ECG, heart rate variability, and SRAM PUF. The resulting vectors can also be utilized for secure cryptography applications. Secondly, a novel in-memory computing (IMC) hardware noise-aware training algorithms that make DNNs more robust to hardware noise is developed and evaluated. Up to 17% accuracy was recovered in deep neural networks (DNNs) deployed on IMC prototype hardware. The noise-aware training principles are also used to improve the adversarial robustness of DNNs, and successfully defend against both adversarial input and weight attacks. Up to ~10\% improvement in robustness against adversarial input attacks, and up to 33% improvement in robustness against adversarial weight attacks are achieved. Finally, a DNN training algorithm that pursues and optimises both activation and weight sparsity simultaneously is proposed and evaluated to obtain highly compressed DNNs. This lead to up to 4.7x reduction in the total number of flops required to perform complex image recognition tasks. A custom sparse inference accelerator is designed and synthesized to evaluate the benefits of the above flop reduction. A speedup of 4.24x is achieved. In summary, this dissertation contains innovative algorithm and hardware design techniques aided by machine learning, which enhance the security and efficiency of edge computing applications.

In recent years, there has been significant progress in deep learning and computer vision, with many models proposed that have achieved state-of-art results on various image recognition tasks. However, to explore the full potential of the advances in this field, there is an urgent need to push the processing of deep networks from the cloud to edge devices. Unfortunately, many deep learning models cannot be efficiently implemented on edge devices as these devices are severely resource-constrained. In this thesis, I present QU-Net, a lightweight binary segmentation model based on the U-Net architecture. Traditionally, neural networks consider the entire image to be significant. However, in real-world scenarios, many regions in an image do not contain any objects of significance. These regions can be removed from the original input allowing a network to focus on the relevant regions and thus reduce computational costs. QU-Net proposes the salient regions (binary mask) that the deeper models can use as the input. Experiments show that QU-Net helped achieve a computational reduction of 25% on the Microsoft Common Objects in Context (MS COCO) dataset and 57% on the Cityscapes dataset. Moreover, QU-Net is a generalizable model that outperforms other similar works, such as Dynamic Convolutions.

The growth in speed and density of programmable logic devices, such as Field programmable gate arrays (FPGA), enables sophisticated designs to be created within a short time frame. The flexibility of a programmable device alleviates the difficulty of the integration of a design with a wide range of components on a single chip. FPGAs bring both performance and power efficiency, especially for compute or data-intensive applications. Efficient and accurate mRNA quantification is an essential step for molecular signature identification, disease outcome prediction, and drug development, which is a typical compute- and data-intensive compute workload. In this work, I propose to accelerate mRNA quantification with FPGA implementation. I analyze the performance of mRNA Quantification with FPGA, which shows better or similar performance compared to that of CPU implementation.

Edge networks pose unique challenges for machine learning and network management. The primary objective of this dissertation is to study deep learning and adaptive control aspects of edge networks and to address some of the unique challenges therein. This dissertation explores four particular problems of interest at the intersection of edge intelligence, deep learning and network management. The first problem explores the learning of generative models in edge learning setting. Since the learning tasks in similar environments share model similarity, it is plausible to leverage pre-trained generative models from other edge nodes. Appealing to optimal transport theory tailored towards Wasserstein-1 generative adversarial networks, this part aims to develop a framework which systematically optimizes the generative model learning performance using local data at the edge node while exploiting the adaptive coalescence of pre-trained generative models from other nodes. In the second part, a many-to-one wireless architecture for federated learning at the network edge, where multiple edge devices collaboratively train a model using local data, is considered. The unreliable nature of wireless connectivity, togetherwith the constraints in computing resources at edge devices, dictates that the local updates at edge devices should be carefully crafted and compressed to match the wireless communication resources available and should work in concert with the receiver. Therefore, a Stochastic Gradient Descent based bandlimited coordinate descent algorithm is designed for such settings. The third part explores the adaptive traffic engineering algorithms in a dynamic network environment. The ages of traffic measurements exhibit significant variation due to asynchronization and random communication delays between routers and controllers. Inspired by the software defined networking architecture, a controller-assisted distributed routing scheme with recursive link weight reconfigurations, accounting for the impact of measurement ages and routing instability, is devised. The final part focuses on developing a federated learning based framework for traffic reshaping of electric vehicle (EV) charging. The absence of private EV owner information and scattered EV charging data among charging stations motivates the utilization of a federated learning approach. Federated learning algorithms are devised to minimize peak EV charging demand both spatially and temporarily, while maximizing the charging station profit.

Over the past decades, the amount of data required to be processed and analyzed by computing systems has been increasing dramatically to exascale (10^18 bytes/s or ops). However, modern computing platforms' inability to deliver both energy-efficient and high-performance computing solutions leads to a gap between meets and needs, especially in resource-constraint Internet of Things (IoT) devices. Unfortunately, such a gap will keep widening mainly due to limitations in both devices and architectures. With this motivation, this dissertation's focus is on cross-layer (device/circuit/architecture/application) co-design of energy-efficient and high-performance Processing-in-Memory (PIM) platforms for implementing complex big data applications, i.e., deep learning, bioinformatics, graph processing tasks, and data encryption. The dissertation shows how to leverage innovations from device, circuit, and architecture to integrate memory and logic to break the existing memory and power walls and dramatically increase computing efficiency of today’s non-Von-Neumann computing systems.The proposed PIM platforms transform current volatile and non-volatile random access memory arrays to computational units capable of working as both memory and low-area-overhead, massively parallel, fast, reconfigurable in-memory logic. Instead of integrating complex logic units in cost-sensitive memory, the explored designs exploit hardware-friendly bit-line computing methods to implement complete Boolean logic functions between operands within a memory array in a reduced clock cycle, overcoming the multi-cycle logic issue in modern PIM platforms. Besides, new customized in-memory algorithms and mapping methods are developed to convert the crucial iteratively-used big data application's functions to bit-wise PIM-supported logic. To quantitatively analyze the performance of various PIM platforms running big data applications, a generic and comprehensive evaluation framework is presented. The overall system computing performance (throughput, latency, energy efficiency) for each application is explored through the developed framework. The device-to-algorithm co-simulation results on neural network acceleration demonstrate that the proposed platforms can obtain 36.8× higher energy-efficiency and 22× speed-up compared to state-of-the-art Graphics Processing Unit (GPU). In accelerating bioinformatics tasks such as biological sequence alignment, the presented PIM designs result in ~2×, 43.8×, 458× more throughput per Watt compared to state-of-the-art Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), and GPU platforms, respectively.

Rapid development of computer vision applications such as image recognition and object detection has been enabled by the emerging deep learning technologies. To improve the accuracy further, deeper and wider neural networks with diverse architecture are proposed for better feature extraction. Though the performance boost is impressive, only marginal improvement can be achieved with significantly increased computational overhead. One solution is to compress the exploding-sized model by dropping less important weights or channels. This is an effective solution that has been well explored. However, by utilizing the rich relation information of the data, one can also improve the accuracy with reasonable overhead. This work makes progress toward efficient and accurate visual tasks including detection, prediction and understanding by using relations.
For object detection, a novel approach, Graph Assisted Reasoning (GAR), is proposed to utilize a heterogeneous graph to model object-object relations and object-scene relations. GAR fuses the features from neighboring object nodes as well as scene nodes. In this way, GAR produces better recognition than that produced from individual object nodes. Moreover, compared to previous approaches using Recurrent Neural Network (RNN), GAR's light-weight and low-coupling architecture further facilitate its integration into the object detection module.

For trajectories prediction, a novel approach, namely Diverse Attention RNN (DAT-RNN), is proposed to handle the diversity of trajectories and modeling of neighboring relations. DAT-RNN integrates both temporal and spatial relations to improve the prediction under various circumstances.

Last but not least, this work presents a novel relation implication-enhanced (RIE) approach that improves relation detection through relation direction and implication. With the relation implication, the SGG model is exposed to more ground truth information and thus mitigates the overfitting problem of the biased datasets. Moreover, the enhancement with relation implication is compatible with various context encoding schemes.

Comprehensive experiments on benchmarking datasets demonstrate the efficacy of the proposed approaches.

The rapid advancement of Deep Neural Networks (DNNs), computing, and sensing technology has enabled many new applications, such as the self-driving vehicle, the surveillance drone, and the robotic system. Compared to conventional edge devices (e.g. cell phone or smart home devices), these emerging devices are required to deal with much more complicated and dynamic situations in real-time with bounded computation resources. However, there are several challenges, including but not limited to efficiency, real-time adaptation, model stability, and automation of architecture design.

To tackle the challenges mentioned above, model plasticity and stability are leveraged to achieve efficient and online deep learning, especially in the scenario of learning streaming data at the edge:

First, a dynamic training scheme named Continuous Growth and Pruning (CGaP) is proposed to compress the DNNs through growing important parameters and pruning unimportant ones, achieving up to 98.1% reduction in the number of parameters.

Second, this dissertation presents Progressive Segmented Training (PST), which targets catastrophic forgetting problems in continual learning through importance sampling, model segmentation, and memory-assisted balancing. PST achieves state-of-the-art accuracy with 1.5X FLOPs reduction in the complete inference path.

Third, to facilitate online learning in real applications, acquisitive learning (AL) is further proposed to emphasize both knowledge inheritance and acquisition: the majority of the knowledge is first pre-trained in the inherited model and then adapted to acquire new knowledge. The inherited model's stability is monitored by noise injection and the landscape of the loss function, while the acquisition is realized by importance sampling and model segmentation. Compared to a conventional scheme, AL reduces accuracy drop by >10X on CIFAR-100 dataset, with 5X reduction in latency per training image and 150X reduction in training FLOPs.

Finally, this dissertation presents evolutionary neural architecture search in light of model stability (ENAS-S). ENAS-S uses a novel fitness score, which addresses not only the accuracy but also the model stability, to search for an optimal inherited model for the application of continual learning. ENAS-S outperforms hand-designed DNNs when learning from a data stream at the edge.

In summary, in this dissertation, several algorithms exploiting model plasticity and model stability are presented to improve the efficiency and accuracy of deep neural networks, especially for the scenario of continual learning.

The advances of Deep Learning (DL) achieved recently have successfully demonstrated its great potential of surpassing or close to human-level performance across multiple domains. Consequently, there exists a rising demand to deploy state-of-the-art DL algorithms, e.g., Deep Neural Networks (DNN), in real-world applications to release labors from repetitive work. On the one hand, the impressive performance achieved by the DNN normally accompanies with the drawbacks of intensive memory and power usage due to enormous model size and high computation workload, which significantly hampers their deployment on the resource-limited cyber-physical systems or edge devices. Thus, the urgent demand for enhancing the inference efficiency of DNN has also great research interests across various communities. On the other hand, scientists and engineers still have insufficient knowledge about the principles of DNN which makes it mostly be treated as a black-box. Under such circumstance, DNN is like "the sword of Damocles" where its security or fault-tolerance capability is an essential concern which cannot be circumvented.

Motivated by the aforementioned concerns, this dissertation comprehensively investigates the emerging efficiency and security issues of DNNs, from both software and hardware design perspectives. From the efficiency perspective, as the foundation technique for efficient inference of target DNN, the model compression via quantization is elaborated. In order to maximize the inference performance boost, the deployment of quantized DNN on the revolutionary Computing-in-Memory based neural accelerator is presented in a cross-layer (device/circuit/system) fashion. From the security perspective, the well known adversarial attack is investigated spanning from its original input attack form (aka. Adversarial example generation) to its parameter attack variant.

Technological advances in low power wearable electronics and energy optimization techniques

make motion energy harvesting a viable energy source. However, it has not been

widely adopted due to bulky energy harvester designs that are uncomfortable to wear. This

work addresses this problem by analyzing the feasibility of powering low wearable power

devices using piezoelectric energy generated at the human knee. We start with a novel

mathematical model for estimating the power generated from human knee joint movements.

This thesis’s major contribution is to analyze the feasibility of human motion energy harvesting

and validating this analytical model using a commercially available piezoelectric

module. To this end, we implemented an experimental setup that replicates a human knee.

Then, we performed experiments at different excitation frequencies and amplitudes with

two commercially available Macro Fiber Composite (MFC) modules. These experimental

results are used to validate the analytical model and predict the energy harvested as a function

of the number of steps taken in a day. The model estimates that 13μWcan be generated

on an average while walking with a 4.8% modeling error. The obtained results show that

piezoelectricity is indeed a viable approach for powering low-power wearable devices.