Matching Items (102)
Filtering by
- Genre: Academic theses
- Creators: Chakrabarti, Chaitali
- Status: Published
Description
QR decomposition (QRD) of a matrix is one of the most common linear algebra operationsused for the decomposition of a square
on-square matrix. It has a wide range
of applications especially in Multiple Input-Multiple Output (MIMO) communication
systems. Unfortunately it has high computation complexity { for matrix size of nxn,
QRD has O(n3) complexity and back substitution, which is used to solve a system
of linear equations, has O(n2) complexity. Thus, as the matrix size increases, the
hardware resource requirement for QRD and back substitution increases signicantly.
This thesis presents the design and implementation of a
exible QRD and back substitution accelerator using a folded architecture. It can support matrix sizes of
4x4, 8x8, 12x12, 16x16, and 20x20 with low hardware resource requirement.
The proposed architecture is based on the systolic array implementation of the
Givens algorithm for QRD. It is built with three dierent types of computation blocks
which are connected in a 2-D array structure. These blocks are controlled by a
scheduler which facilitates reusability of the blocks to perform computation for any
input matrix size which is a multiple of 4. These blocks are designed using two
basic programming elements which support both the forward and backward paths to
compute matrix R in QRD and column-matrix X in back substitution computation.
The proposed architecture has been mapped to Xilinx Zynq Ultrascale+ FPGA
(Field Programmable Gate Array), ZCU102. All inputs are complex with precision
of 40 bits (38 fractional bits and 1 signed bit). The architecture can be clocked at
50 MHz. The synthesis results of the folded architecture for dierent matrix sizes
are presented. The results show that the folded architecture can support QRD and
back substitution for inputs of large sizes which otherwise cannot t on an FPGA
when implemented using a
at architecture. The memory sizes required for dierent
matrix sizes are also presented.
on-square matrix. It has a wide range
of applications especially in Multiple Input-Multiple Output (MIMO) communication
systems. Unfortunately it has high computation complexity { for matrix size of nxn,
QRD has O(n3) complexity and back substitution, which is used to solve a system
of linear equations, has O(n2) complexity. Thus, as the matrix size increases, the
hardware resource requirement for QRD and back substitution increases signicantly.
This thesis presents the design and implementation of a
exible QRD and back substitution accelerator using a folded architecture. It can support matrix sizes of
4x4, 8x8, 12x12, 16x16, and 20x20 with low hardware resource requirement.
The proposed architecture is based on the systolic array implementation of the
Givens algorithm for QRD. It is built with three dierent types of computation blocks
which are connected in a 2-D array structure. These blocks are controlled by a
scheduler which facilitates reusability of the blocks to perform computation for any
input matrix size which is a multiple of 4. These blocks are designed using two
basic programming elements which support both the forward and backward paths to
compute matrix R in QRD and column-matrix X in back substitution computation.
The proposed architecture has been mapped to Xilinx Zynq Ultrascale+ FPGA
(Field Programmable Gate Array), ZCU102. All inputs are complex with precision
of 40 bits (38 fractional bits and 1 signed bit). The architecture can be clocked at
50 MHz. The synthesis results of the folded architecture for dierent matrix sizes
are presented. The results show that the folded architecture can support QRD and
back substitution for inputs of large sizes which otherwise cannot t on an FPGA
when implemented using a
at architecture. The memory sizes required for dierent
matrix sizes are also presented.
ContributorsKanagala, Srimayee (Author) / Chakrabarti, Chaitali (Thesis advisor) / Bliss, Daniel (Committee member) / Cao, Yu (Kevin) (Committee member) / Arizona State University (Publisher)
Created2020
Description
Lattice-based Cryptography is an up and coming field of cryptography that utilizes the difficulty of lattice problems to design lattice-based cryptosystems that are resistant to quantum attacks and applicable to Fully Homomorphic Encryption schemes (FHE). In this thesis, the parallelization of the Residue Number System (RNS) and algorithmic efficiency of the Number Theoretic Transform (NTT) are combined to tackle the most significant bottleneck of polynomial ring multiplication with the hardware design of an optimized RNS-based NTT polynomial multiplier. The design utilizes Negative Wrapped Convolution, the NTT, RNS Montgomery reduction with Bajard and Shenoy extensions, and optimized modular 32-bit channel arithmetic for nine RNS channels to accomplish an RNS polynomial multiplication. In addition to a full software implementation of the whole system, a pipelined and optimized RNS-based NTT unit with 4 RNS butterflies is implemented on the Xilinx Artix-7 FPGA(xc7a200tlffg1156-2L) for size and delay estimates. The hardware implementation achieves an operating frequency of 47.043 MHz and utilizes 13239 LUT's, 4010 FF's, and 330 DSP blocks, allowing for multiple simultaneously operating NTT units depending on FGPA size constraints.
ContributorsBrist, Logan Alan (Author) / Chakrabarti, Chaitali (Thesis advisor) / Papandreou-Suppappola, Antonia (Committee member) / Bliss, Daniel (Committee member) / Arizona State University (Publisher)
Created2020
Description
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.
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.
ContributorsDu, Xiaocong (Author) / Cao, Yu (Thesis advisor) / Seo, Jae-Sun (Committee member) / Chakrabarti, Chaitali (Committee member) / Fan, Deliang (Committee member) / Arizona State University (Publisher)
Created2020
Description
As device and voltage scaling cease, ever-increasing performance targets can only be achieved through the design of parallel, heterogeneous architectures. The workloads targeted by these domain-specific architectures must be designed to leverage the strengths of the platform: a task that has proven to be extremely difficult and expensive.
Machine learning has the potential to automate this process by understanding the features of computation that optimize device utilization and throughput.
Unfortunately, applications of this technique have utilized small data-sets and specific feature extraction, limiting the impact of their contributions.
To address this problem I present Dash-Database; a repository of C and C++ programs for software-defined radio applications and its neighboring fields; a methodology for structuring the features of computation using kernels, and a set of evaluation metrics to standardize computation data sets. Dash-Database contributes a general data set that supports machine understanding of computation and standardizes the input corpus utilized for machine learning of computation; currently only a small set of benchmarks and features are being used.
I present an evaluation of Dash-Database using three novel metrics: breadth, depth and richness; and compare its results to a data set largely representative of those used in prior work, indicating a 5x increase in breadth, 40x increase in depth, and a rich set of sample features.
Using Dash-Database, the broader community can work toward a general machine understanding of computation that can automate the design of workloads for domain-specific computation.
Machine learning has the potential to automate this process by understanding the features of computation that optimize device utilization and throughput.
Unfortunately, applications of this technique have utilized small data-sets and specific feature extraction, limiting the impact of their contributions.
To address this problem I present Dash-Database; a repository of C and C++ programs for software-defined radio applications and its neighboring fields; a methodology for structuring the features of computation using kernels, and a set of evaluation metrics to standardize computation data sets. Dash-Database contributes a general data set that supports machine understanding of computation and standardizes the input corpus utilized for machine learning of computation; currently only a small set of benchmarks and features are being used.
I present an evaluation of Dash-Database using three novel metrics: breadth, depth and richness; and compare its results to a data set largely representative of those used in prior work, indicating a 5x increase in breadth, 40x increase in depth, and a rich set of sample features.
Using Dash-Database, the broader community can work toward a general machine understanding of computation that can automate the design of workloads for domain-specific computation.
ContributorsWillis, Benjamin Roy (Author) / Brunhaver, John S (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Shrivastava, Aviral (Committee member) / Arizona State University (Publisher)
Created2020
Description
The following document describes the hardware implementation and analysis of Temporal Interference Mitigation using High-Level Synthesis. As the problem of spectral congestion becomes more chronic and widespread, Electromagnetic radio frequency (RF) based systems are posing as viable solution to this problem. Among the existing RF methods Cooperation based systems have been a solution to a host of congestion problems. One of the most important elements of RF receiver is the spatially adaptive part of the receiver. Temporal Mitigation is vital technique employed at the receiver for signal recovery and future propagation along the radar chain.
The computationally intensive parts of temporal mitigation are identified and hardware accelerated. The hardware implementation is based on sequential approach with optimizations applied on the individual components for better performance.
An extensive analysis using a range of fixed point data types is performed to find the optimal data type necessary.
Finally a hybrid combination of data types for different components of temporal mitigation is proposed based on results from the above analysis.
The computationally intensive parts of temporal mitigation are identified and hardware accelerated. The hardware implementation is based on sequential approach with optimizations applied on the individual components for better performance.
An extensive analysis using a range of fixed point data types is performed to find the optimal data type necessary.
Finally a hybrid combination of data types for different components of temporal mitigation is proposed based on results from the above analysis.
ContributorsSiddiqui, Saquib Ahmad (Author) / Bliss, Daniel (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Ogras, Umit Y. (Committee member) / Jayasuriya, Suren (Committee member) / Arizona State University (Publisher)
Created2020
Description
Movement disorders are becoming one of the leading causes of functional disability due to aging populations and extended life expectancy. Diagnosis, treatment, and rehabilitation currently depend on the behavior observed in a clinical environment. After the patient leaves the clinic, there is no standard approach to continuously monitor the patient and report potential problems. Furthermore, self-recording is inconvenient and unreliable. To address these challenges, wearable health monitoring is emerging as an effective way to augment clinical care for movement disorders.
Wearable devices are being used in many health, fitness, and activity monitoring applications. However, their widespread adoption has been hindered by several adaptation and technical challenges. First, conventional rigid devices are uncomfortable to wear for long periods. Second, wearable devices must operate under very low-energy budgets due to their small battery capacities. Small batteries create a need for frequent recharging, which in turn leads users to stop using them. Third, the usefulness of wearable devices must be demonstrated through high impact applications such that users can get value out of them.
This dissertation presents solutions to solving the challenges faced by wearable devices. First, it presents an open-source hardware/software platform for wearable health monitoring. The proposed platform uses flexible hybrid electronics to enable devices that conform to the shape of the user’s body. Second, it proposes an algorithm to enable recharge-free operation of wearable devices that harvest energy from the environment. The proposed solution maximizes the performance of the wearable device under minimum energy constraints. The results of the proposed algorithm are, on average, within 3% of the optimal solution computed offline. Third, a comprehensive framework for human activity recognition (HAR), one of the first steps towards a solution for movement disorders is presented. It starts with an online learning framework for HAR. Experiments on a low power IoT device (TI-CC2650 MCU) with twenty-two users show 95% accuracy in identifying seven activities and their transitions with less than 12.5 mW power consumption. The online learning framework is accompanied by a transfer learning approach for HAR that determines the number of neural network layers to transfer among uses to enable efficient online learning. Next, a technique to co-optimize the accuracy and active time of wearable applications by utilizing multiple design points with different energy-accuracy trade-offs is presented. The proposed technique switches between the design points at runtime to maximize a generalized objective function under tight harvested energy budget constraints. Finally, we present the first ultra-low-energy hardware accelerator that makes it practical to perform HAR on energy harvested from wearable devices. The accelerator consumes 22.4 microjoules per operation using a commercial 65 nm technology. In summary, the solutions presented in this dissertation can enable the wider adoption of wearable devices.
Wearable devices are being used in many health, fitness, and activity monitoring applications. However, their widespread adoption has been hindered by several adaptation and technical challenges. First, conventional rigid devices are uncomfortable to wear for long periods. Second, wearable devices must operate under very low-energy budgets due to their small battery capacities. Small batteries create a need for frequent recharging, which in turn leads users to stop using them. Third, the usefulness of wearable devices must be demonstrated through high impact applications such that users can get value out of them.
This dissertation presents solutions to solving the challenges faced by wearable devices. First, it presents an open-source hardware/software platform for wearable health monitoring. The proposed platform uses flexible hybrid electronics to enable devices that conform to the shape of the user’s body. Second, it proposes an algorithm to enable recharge-free operation of wearable devices that harvest energy from the environment. The proposed solution maximizes the performance of the wearable device under minimum energy constraints. The results of the proposed algorithm are, on average, within 3% of the optimal solution computed offline. Third, a comprehensive framework for human activity recognition (HAR), one of the first steps towards a solution for movement disorders is presented. It starts with an online learning framework for HAR. Experiments on a low power IoT device (TI-CC2650 MCU) with twenty-two users show 95% accuracy in identifying seven activities and their transitions with less than 12.5 mW power consumption. The online learning framework is accompanied by a transfer learning approach for HAR that determines the number of neural network layers to transfer among uses to enable efficient online learning. Next, a technique to co-optimize the accuracy and active time of wearable applications by utilizing multiple design points with different energy-accuracy trade-offs is presented. The proposed technique switches between the design points at runtime to maximize a generalized objective function under tight harvested energy budget constraints. Finally, we present the first ultra-low-energy hardware accelerator that makes it practical to perform HAR on energy harvested from wearable devices. The accelerator consumes 22.4 microjoules per operation using a commercial 65 nm technology. In summary, the solutions presented in this dissertation can enable the wider adoption of wearable devices.
ContributorsBhat, Ganapati (Author) / Ogras, Umit Y. (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Nedić, Angelia (Committee member) / Marculescu, Radu (Committee member) / Arizona State University (Publisher)
Created2020
Description
The past decade has seen a tremendous surge in running machine learning (ML) functions on mobile devices, from mere novelty applications to now indispensable features for the next generation of devices.
While the mobile platform capabilities range widely, long battery life and reliability are common design concerns that are crucial to remain competitive.
Consequently, state-of-the-art mobile platforms have become highly heterogeneous by combining a powerful CPUs with GPUs to accelerate the computation of deep neural networks (DNNs), which are the most common structures to perform ML operations.
But traditional von Neumann architectures are not optimized for the high memory bandwidth and massively parallel computation demands required by DNNs.
Hence, propelling research into non-von Neumann architectures to support the demands of DNNs.
The re-imagining of computer architectures to perform efficient DNN computations requires focusing on the prohibitive demands presented by DNNs and alleviating them. The two central challenges for efficient computation are (1) large memory storage and movement due to weights of the DNN and (2) massively parallel multiplications to compute the DNN output.
Introducing sparsity into the DNNs, where certain percentage of either the weights or the outputs of the DNN are zero, greatly helps with both challenges. This along with algorithm-hardware co-design to compress the DNNs is demonstrated to provide efficient solutions to greatly reduce the power consumption of hardware that compute DNNs. Additionally, exploring emerging technologies such as non-volatile memories and 3-D stacking of silicon in conjunction with algorithm-hardware co-design architectures will pave the way for the next generation of mobile devices.
Towards the objectives stated above, our specific contributions include (a) an architecture based on resistive crosspoint array that can update all values stored and compute matrix vector multiplication in parallel within a single cycle, (b) a framework of training DNNs with a block-wise sparsity to drastically reduce memory storage and total number of computations required to compute the output of DNNs, (c) the exploration of hardware implementations of sparse DNNs and architectural guidelines to reduce power consumption for the implementations in monolithic 3D integrated circuits, and (d) a prototype chip in 65nm CMOS accelerator for long-short term memory networks trained with the proposed block-wise sparsity scheme.
While the mobile platform capabilities range widely, long battery life and reliability are common design concerns that are crucial to remain competitive.
Consequently, state-of-the-art mobile platforms have become highly heterogeneous by combining a powerful CPUs with GPUs to accelerate the computation of deep neural networks (DNNs), which are the most common structures to perform ML operations.
But traditional von Neumann architectures are not optimized for the high memory bandwidth and massively parallel computation demands required by DNNs.
Hence, propelling research into non-von Neumann architectures to support the demands of DNNs.
The re-imagining of computer architectures to perform efficient DNN computations requires focusing on the prohibitive demands presented by DNNs and alleviating them. The two central challenges for efficient computation are (1) large memory storage and movement due to weights of the DNN and (2) massively parallel multiplications to compute the DNN output.
Introducing sparsity into the DNNs, where certain percentage of either the weights or the outputs of the DNN are zero, greatly helps with both challenges. This along with algorithm-hardware co-design to compress the DNNs is demonstrated to provide efficient solutions to greatly reduce the power consumption of hardware that compute DNNs. Additionally, exploring emerging technologies such as non-volatile memories and 3-D stacking of silicon in conjunction with algorithm-hardware co-design architectures will pave the way for the next generation of mobile devices.
Towards the objectives stated above, our specific contributions include (a) an architecture based on resistive crosspoint array that can update all values stored and compute matrix vector multiplication in parallel within a single cycle, (b) a framework of training DNNs with a block-wise sparsity to drastically reduce memory storage and total number of computations required to compute the output of DNNs, (c) the exploration of hardware implementations of sparse DNNs and architectural guidelines to reduce power consumption for the implementations in monolithic 3D integrated circuits, and (d) a prototype chip in 65nm CMOS accelerator for long-short term memory networks trained with the proposed block-wise sparsity scheme.
ContributorsKadetotad, Deepak Vinayak (Author) / Seo, Jae-Sun (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Vrudhula, Sarma (Committee member) / Cao, Yu (Committee member) / Arizona State University (Publisher)
Created2019
Description
Heterogenous SoCs are in development that marry multiple architectural patterns together. In order for software to be run on such a platform, it must be broken down into its constituent parts, kernels, and scheduled for execution on the hardware. Although this can be done by hand, it would be arduous and time consuming; rather, a tool should be developed that analyzes the source binary, extracts the kernels, schedules the kernels, and optimizes the scheduled kernels for their target component. This dissertation proposes a decidable kernel definition that enables an algorithmic approach to detecting kernels from arbitrary programs. This definition is built upon four constraints that can be tested using basic graph theory. In addition, two algorithms are proposed that successfully extract kernels based upon runtime information. The first utilizes dynamic traces, which are generated using a collection of novel optimizations. The second utilizes a simple affinity matrix, which has no runtime overhead during program execution. Finally, a Dense Neural Network is proposed that is capable of detecting a kernel's archetype based upon only the composition of the source program and the number of times individual basic blocks execute. The contributions proposed in this dissertation provide the necessary infrastructure to perform a litany of other optimizations on kernels. By detecting kernels algorithmically, any program can be analyzed and optimized with techniques that have heretofore required kernels be written in a compatible form. Computational kernels can be extracted from any program with no constraints. The innovations describes here will form the foundation for automated kernel optimization in the future, helping optimize the code of the future.
ContributorsUhrie, Richard Lawrence (Author) / Brunhaver, John (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Shrivastiva, Aviral (Committee member) / Wu, Carole-Jean (Committee member) / Arizona State University (Publisher)
Created2021
Description
Due to high DRAM access latency and energy, several convolutional neural network(CNN) accelerators face performance and energy efficiency challenges, which are critical for embedded implementations. As these applications exploit larger datasets, memory accesses of these emerging applications are increasing. As a result, it is difficult to predict the combined dynamic random access memory (DRAM) workload behavior, which can sabotage memory optimizations in software. To understand the impact of external memory access on CNN accelerators which reduces the high DRAMaccess latency and energy, simulators such as RAMULATOR and VAMPIRE have been proposed in prior work. In this work, we utilize these simulators to benchmark external memory access in CNN accelerators. Experiments are performed generating trace files based on the number of parameters and data precision and also using trace file generated for CNN Accelerator Altera Arria 10 GX 1150 FPGA data to complete the end to end workflow using the mentioned simulators. Besides that, certain modifications were made in the default VAMPIRE code to implement certain functionalities such as PREA(Precharge All) and REF(Refresh). Then, precalculated energies were computed for DDR3, DDR4, and HBM based on the micron model to mention it in the dram specification file inputted to the VAMPIRE tool. An experimental study was performed and a comparison is made between DDR3, DDR4, and HBM, it was proved that DDR4 is nearly 31% more energy-efficient than DDR3 and HBMis 54% energy-efficient than DDR3. Performed modeling and experimental analysis on a large set of data and then split it into a set of data and compared the results of the small sets multiplied with the number of sets and the large data set and concluded that the results were nearly the same. Finally, a GUI is developed by wrapping both the simulators. GUI provides user-friendly access and can analyze the parameters without much prior knowledge and understanding of the working.
ContributorsPannala, Manvitha (Author) / Cao, Yu (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Seo, Jae-Sun (Committee member) / Arizona State University (Publisher)
Created2021
Description
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.
ContributorsCherupally, Sai Kiran (Author) / Seo, Jae-Sun (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Cao, Yu (Kevin) (Committee member) / Fan, Deliang (Committee member) / Arizona State University (Publisher)
Created2022