Matching Items (10)
Filtering by
- All Subjects: artificial intelligence
Description
Deep neural networks (DNN) have shown tremendous success in various cognitive tasks, such as image classification, speech recognition, etc. However, their usage on resource-constrained edge devices has been limited due to high computation and large memory requirement.
To overcome these challenges, recent works have extensively investigated model compression techniques such as element-wise sparsity, structured sparsity and quantization. While most of these works have applied these compression techniques in isolation, there have been very few studies on application of quantization and structured sparsity together on a DNN model.
This thesis co-optimizes structured sparsity and quantization constraints on DNN models during training. Specifically, it obtains optimal setting of 2-bit weight and 2-bit activation coupled with 4X structured compression by performing combined exploration of quantization and structured compression settings. The optimal DNN model achieves 50X weight memory reduction compared to floating-point uncompressed DNN. This memory saving is significant since applying only structured sparsity constraints achieves 2X memory savings and only quantization constraints achieves 16X memory savings. The algorithm has been validated on both high and low capacity DNNs and on wide-sparse and deep-sparse DNN models. Experiments demonstrated that deep-sparse DNN outperforms shallow-dense DNN with varying level of memory savings depending on DNN precision and sparsity levels. This work further proposed a Pareto-optimal approach to systematically extract optimal DNN models from a huge set of sparse and dense DNN models. The resulting 11 optimal designs were further evaluated by considering overall DNN memory which includes activation memory and weight memory. It was found that there is only a small change in the memory footprint of the optimal designs corresponding to the low sparsity DNNs. However, activation memory cannot be ignored for high sparsity DNNs.
To overcome these challenges, recent works have extensively investigated model compression techniques such as element-wise sparsity, structured sparsity and quantization. While most of these works have applied these compression techniques in isolation, there have been very few studies on application of quantization and structured sparsity together on a DNN model.
This thesis co-optimizes structured sparsity and quantization constraints on DNN models during training. Specifically, it obtains optimal setting of 2-bit weight and 2-bit activation coupled with 4X structured compression by performing combined exploration of quantization and structured compression settings. The optimal DNN model achieves 50X weight memory reduction compared to floating-point uncompressed DNN. This memory saving is significant since applying only structured sparsity constraints achieves 2X memory savings and only quantization constraints achieves 16X memory savings. The algorithm has been validated on both high and low capacity DNNs and on wide-sparse and deep-sparse DNN models. Experiments demonstrated that deep-sparse DNN outperforms shallow-dense DNN with varying level of memory savings depending on DNN precision and sparsity levels. This work further proposed a Pareto-optimal approach to systematically extract optimal DNN models from a huge set of sparse and dense DNN models. The resulting 11 optimal designs were further evaluated by considering overall DNN memory which includes activation memory and weight memory. It was found that there is only a small change in the memory footprint of the optimal designs corresponding to the low sparsity DNNs. However, activation memory cannot be ignored for high sparsity DNNs.
ContributorsSrivastava, Gaurav (Author) / Seo, Jae-Sun (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Berisha, Visar (Committee member) / Arizona State University (Publisher)
Created2018
Description
The rapid improvement in computation capability has made deep convolutional neural networks (CNNs) a great success in recent years on many computer vision tasks with significantly improved accuracy. During the inference phase, many applications demand low latency processing of one image with strict power consumption requirement, which reduces the efficiency of GPU and other general-purpose platform, bringing opportunities for specific acceleration hardware, e.g. FPGA, by customizing the digital circuit specific for the deep learning algorithm inference. However, deploying CNNs on portable and embedded systems is still challenging due to large data volume, intensive computation, varying algorithm structures, and frequent memory accesses. This dissertation proposes a complete design methodology and framework to accelerate the inference process of various CNN algorithms on FPGA hardware with high performance, efficiency and flexibility.
As convolution contributes most operations in CNNs, the convolution acceleration scheme significantly affects the efficiency and performance of a hardware CNN accelerator. Convolution involves multiply and accumulate (MAC) operations with four levels of loops. Without fully studying the convolution loop optimization before the hardware design phase, the resulting accelerator can hardly exploit the data reuse and manage data movement efficiently. This work overcomes these barriers by quantitatively analyzing and optimizing the design objectives (e.g. memory access) of the CNN accelerator based on multiple design variables. An efficient dataflow and hardware architecture of CNN acceleration are proposed to minimize the data communication while maximizing the resource utilization to achieve high performance.
Although great performance and efficiency can be achieved by customizing the FPGA hardware for each CNN model, significant efforts and expertise are required leading to long development time, which makes it difficult to catch up with the rapid development of CNN algorithms. In this work, we present an RTL-level CNN compiler that automatically generates customized FPGA hardware for the inference tasks of various CNNs, in order to enable high-level fast prototyping of CNNs from software to FPGA and still keep the benefits of low-level hardware optimization. First, a general-purpose library of RTL modules is developed to model different operations at each layer. The integration and dataflow of physical modules are predefined in the top-level system template and reconfigured during compilation for a given CNN algorithm. The runtime control of layer-by-layer sequential computation is managed by the proposed execution schedule so that even highly irregular and complex network topology, e.g. GoogLeNet and ResNet, can be compiled. The proposed methodology is demonstrated with various CNN algorithms, e.g. NiN, VGG, GoogLeNet and ResNet, on two different standalone FPGAs achieving state-of-the art performance.
Based on the optimized acceleration strategy, there are still a lot of design options, e.g. the degree and dimension of computation parallelism, the size of on-chip buffers, and the external memory bandwidth, which impact the utilization of computation resources and data communication efficiency, and finally affect the performance and energy consumption of the accelerator. The large design space of the accelerator makes it impractical to explore the optimal design choice during the real implementation phase. Therefore, a performance model is proposed in this work to quantitatively estimate the accelerator performance and resource utilization. By this means, the performance bottleneck and design bound can be identified and the optimal design option can be explored early in the design phase.
As convolution contributes most operations in CNNs, the convolution acceleration scheme significantly affects the efficiency and performance of a hardware CNN accelerator. Convolution involves multiply and accumulate (MAC) operations with four levels of loops. Without fully studying the convolution loop optimization before the hardware design phase, the resulting accelerator can hardly exploit the data reuse and manage data movement efficiently. This work overcomes these barriers by quantitatively analyzing and optimizing the design objectives (e.g. memory access) of the CNN accelerator based on multiple design variables. An efficient dataflow and hardware architecture of CNN acceleration are proposed to minimize the data communication while maximizing the resource utilization to achieve high performance.
Although great performance and efficiency can be achieved by customizing the FPGA hardware for each CNN model, significant efforts and expertise are required leading to long development time, which makes it difficult to catch up with the rapid development of CNN algorithms. In this work, we present an RTL-level CNN compiler that automatically generates customized FPGA hardware for the inference tasks of various CNNs, in order to enable high-level fast prototyping of CNNs from software to FPGA and still keep the benefits of low-level hardware optimization. First, a general-purpose library of RTL modules is developed to model different operations at each layer. The integration and dataflow of physical modules are predefined in the top-level system template and reconfigured during compilation for a given CNN algorithm. The runtime control of layer-by-layer sequential computation is managed by the proposed execution schedule so that even highly irregular and complex network topology, e.g. GoogLeNet and ResNet, can be compiled. The proposed methodology is demonstrated with various CNN algorithms, e.g. NiN, VGG, GoogLeNet and ResNet, on two different standalone FPGAs achieving state-of-the art performance.
Based on the optimized acceleration strategy, there are still a lot of design options, e.g. the degree and dimension of computation parallelism, the size of on-chip buffers, and the external memory bandwidth, which impact the utilization of computation resources and data communication efficiency, and finally affect the performance and energy consumption of the accelerator. The large design space of the accelerator makes it impractical to explore the optimal design choice during the real implementation phase. Therefore, a performance model is proposed in this work to quantitatively estimate the accelerator performance and resource utilization. By this means, the performance bottleneck and design bound can be identified and the optimal design option can be explored early in the design phase.
ContributorsMa, Yufei (Author) / Vrudhula, Sarma (Thesis advisor) / Seo, Jae-Sun (Thesis advisor) / Cao, Yu (Committee member) / Barnaby, Hugh (Committee member) / Arizona State University (Publisher)
Created2018
Description
Deep learning (DL) has proved itself be one of the most important developements till date with far reaching impacts in numerous fields like robotics, computer vision, surveillance, speech processing, machine translation, finance, etc. They are now widely used for countless applications because of their ability to generalize real world data, robustness to noise in previously unseen data and high inference accuracy. With the ability to learn useful features from raw sensor data, deep learning algorithms have out-performed tradinal AI algorithms and pushed the boundaries of what can be achieved with AI. In this work, we demonstrate the power of deep learning by developing a neural network to automatically detect cough instances from audio recorded in un-constrained environments. For this, 24 hours long recordings from 9 dierent patients is collected and carefully labeled by medical personel. A pre-processing algorithm is proposed to convert event based cough dataset to a more informative dataset with start and end of coughs and also introduce data augmentation for regularizing the training procedure. The proposed neural network achieves 92.3% leave-one-out accuracy on data captured in real world.
Deep neural networks are composed of multiple layers that are compute/memory intensive. This makes it difficult to execute these algorithms real-time with low power consumption using existing general purpose computers. In this work, we propose hardware accelerators for a traditional AI algorithm based on random forest trees and two representative deep convolutional neural networks (AlexNet and VGG). With the proposed acceleration techniques, ~ 30x performance improvement was achieved compared to CPU for random forest trees. For deep CNNS, we demonstrate that much higher performance can be achieved with architecture space exploration using any optimization algorithms with system level performance and area models for hardware primitives as inputs and goal of minimizing latency with given resource constraints. With this method, ~30GOPs performance was achieved for Stratix V FPGA boards.
Hardware acceleration of DL algorithms alone is not always the most ecient way and sucient to achieve desired performance. There is a huge headroom available for performance improvement provided the algorithms are designed keeping in mind the hardware limitations and bottlenecks. This work achieves hardware-software co-optimization for Non-Maximal Suppression (NMS) algorithm. Using the proposed algorithmic changes and hardware architecture
With CMOS scaling coming to an end and increasing memory bandwidth bottlenecks, CMOS based system might not scale enough to accommodate requirements of more complicated and deeper neural networks in future. In this work, we explore RRAM crossbars and arrays as compact, high performing and energy efficient alternative to CMOS accelerators for deep learning training and inference. We propose and implement RRAM periphery read and write circuits and achieved ~3000x performance improvement in online dictionary learning compared to CPU.
This work also examines the realistic RRAM devices and their non-idealities. We do an in-depth study of the effects of RRAM non-idealities on inference accuracy when a pretrained model is mapped to RRAM based accelerators. To mitigate this issue, we propose Random Sparse Adaptation (RSA), a novel scheme aimed at tuning the model to take care of the faults of the RRAM array on which it is mapped. Our proposed method can achieve inference accuracy much higher than what traditional Read-Verify-Write (R-V-W) method could achieve. RSA can also recover lost inference accuracy 100x ~ 1000x faster compared to R-V-W. Using 32-bit high precision RSA cells, we achieved ~10% higher accuracy using fautly RRAM arrays compared to what can be achieved by mapping a deep network to an 32 level RRAM array with no variations.
Deep neural networks are composed of multiple layers that are compute/memory intensive. This makes it difficult to execute these algorithms real-time with low power consumption using existing general purpose computers. In this work, we propose hardware accelerators for a traditional AI algorithm based on random forest trees and two representative deep convolutional neural networks (AlexNet and VGG). With the proposed acceleration techniques, ~ 30x performance improvement was achieved compared to CPU for random forest trees. For deep CNNS, we demonstrate that much higher performance can be achieved with architecture space exploration using any optimization algorithms with system level performance and area models for hardware primitives as inputs and goal of minimizing latency with given resource constraints. With this method, ~30GOPs performance was achieved for Stratix V FPGA boards.
Hardware acceleration of DL algorithms alone is not always the most ecient way and sucient to achieve desired performance. There is a huge headroom available for performance improvement provided the algorithms are designed keeping in mind the hardware limitations and bottlenecks. This work achieves hardware-software co-optimization for Non-Maximal Suppression (NMS) algorithm. Using the proposed algorithmic changes and hardware architecture
With CMOS scaling coming to an end and increasing memory bandwidth bottlenecks, CMOS based system might not scale enough to accommodate requirements of more complicated and deeper neural networks in future. In this work, we explore RRAM crossbars and arrays as compact, high performing and energy efficient alternative to CMOS accelerators for deep learning training and inference. We propose and implement RRAM periphery read and write circuits and achieved ~3000x performance improvement in online dictionary learning compared to CPU.
This work also examines the realistic RRAM devices and their non-idealities. We do an in-depth study of the effects of RRAM non-idealities on inference accuracy when a pretrained model is mapped to RRAM based accelerators. To mitigate this issue, we propose Random Sparse Adaptation (RSA), a novel scheme aimed at tuning the model to take care of the faults of the RRAM array on which it is mapped. Our proposed method can achieve inference accuracy much higher than what traditional Read-Verify-Write (R-V-W) method could achieve. RSA can also recover lost inference accuracy 100x ~ 1000x faster compared to R-V-W. Using 32-bit high precision RSA cells, we achieved ~10% higher accuracy using fautly RRAM arrays compared to what can be achieved by mapping a deep network to an 32 level RRAM array with no variations.
ContributorsMohanty, Abinash (Author) / Cao, Yu (Thesis advisor) / Seo, Jae-Sun (Committee member) / Vrudhula, Sarma (Committee member) / Chakrabarti, Chaitali (Committee member) / Arizona State University (Publisher)
Created2018
Description
Cognitive technology has been at the forefront of the minds of many technology, government, and business leaders, because of its potential to completely revolutionize their fields. Furthermore, individuals in financial statement auditor roles are especially focused on the impact of cognitive technology because of its potential to eliminate many of the tedious, repetitive tasks involved in their profession. Adopting new technologies that can autonomously collect more data from a broader range of sources, turn the data into business intelligence, and even make decisions based on that data begs the question of whether human roles in accounting will be completely replaced. A partial answer: If the ramifications of past technological advances are any indicator, cognitive technology will replace some human audit operations and grow some new and higher order roles for humans. It will shift the focus of accounting professionals to more complex judgment and analysis.
The next question: What do these changes in the roles and responsibilities look like for the auditors of the future? Cognitive technology will assuredly present new issues for which humans will have to find solutions.
• How will humans be able to test the accuracy and completeness of the decisions derived by cognitive systems?
• If cognitive computing systems rely on supervised learning, what is the most effective way to train systems?
• How will cognitive computing fair in an industry that experiences ever-changing industry regulations?
• Will cognitive technology enhance the quality of audits?
In order to answer these questions and many more, I plan on examining how cognitive technologies evolved into their use today. Based on this historic trajectory, stakeholder interviews, and industry research, I will forecast what auditing jobs may look like in the near future taking into account rapid advances in cognitive computing.
The conclusions forecast a future in auditing that is much more accurate, timely, and pleasant. Cognitive technologies allow auditors to test entire populations of transactions, to tackle audit issues on a more continuous basis, to alleviate the overload of work that occurs after fiscal year-end, and to focus on client interaction.
The next question: What do these changes in the roles and responsibilities look like for the auditors of the future? Cognitive technology will assuredly present new issues for which humans will have to find solutions.
• How will humans be able to test the accuracy and completeness of the decisions derived by cognitive systems?
• If cognitive computing systems rely on supervised learning, what is the most effective way to train systems?
• How will cognitive computing fair in an industry that experiences ever-changing industry regulations?
• Will cognitive technology enhance the quality of audits?
In order to answer these questions and many more, I plan on examining how cognitive technologies evolved into their use today. Based on this historic trajectory, stakeholder interviews, and industry research, I will forecast what auditing jobs may look like in the near future taking into account rapid advances in cognitive computing.
The conclusions forecast a future in auditing that is much more accurate, timely, and pleasant. Cognitive technologies allow auditors to test entire populations of transactions, to tackle audit issues on a more continuous basis, to alleviate the overload of work that occurs after fiscal year-end, and to focus on client interaction.
ContributorsWitkop, David (Author) / Dawson, Gregory (Thesis director) / Munshi, Perseus (Committee member) / School of Accountancy (Contributor) / Department of Information Systems (Contributor) / Barrett, The Honors College (Contributor)
Created2018-05
Description
Speech recognition and keyword detection are becoming increasingly popular applications for mobile systems. While deep neural network (DNN) implementation of these systems have very good performance,
they have large memory and compute resource requirements, making their implementation on a mobile device quite challenging. In this thesis, techniques to reduce the memory and computation cost
of keyword detection and speech recognition networks (or DNNs) are presented.
The first technique is based on representing all weights and biases by a small number of bits and mapping all nodal computations into fixed-point ones with minimal degradation in the
accuracy. Experiments conducted on the Resource Management (RM) database show that for the keyword detection neural network, representing the weights by 5 bits results in a 6 fold reduction in memory compared to a floating point implementation with very little loss in performance. Similarly, for the speech recognition neural network, representing the weights by 6 bits results in a 5 fold reduction in memory while maintaining an error rate similar to a floating point implementation. Additional reduction in memory is achieved by a technique called weight pruning,
where the weights are classified as sensitive and insensitive and the sensitive weights are represented with higher precision. A combination of these two techniques helps reduce the memory
footprint by 81 - 84% for speech recognition and keyword detection networks respectively.
Further reduction in memory size is achieved by judiciously dropping connections for large blocks of weights. The corresponding technique, termed coarse-grain sparsification, introduces
hardware-aware sparsity during DNN training, which leads to efficient weight memory compression and significant reduction in the number of computations during classification without
loss of accuracy. Keyword detection and speech recognition DNNs trained with 75% of the weights dropped and classified with 5-6 bit weight precision effectively reduced the weight memory
requirement by ~95% compared to a fully-connected network with double precision, while showing similar performance in keyword detection accuracy and word error rate.
they have large memory and compute resource requirements, making their implementation on a mobile device quite challenging. In this thesis, techniques to reduce the memory and computation cost
of keyword detection and speech recognition networks (or DNNs) are presented.
The first technique is based on representing all weights and biases by a small number of bits and mapping all nodal computations into fixed-point ones with minimal degradation in the
accuracy. Experiments conducted on the Resource Management (RM) database show that for the keyword detection neural network, representing the weights by 5 bits results in a 6 fold reduction in memory compared to a floating point implementation with very little loss in performance. Similarly, for the speech recognition neural network, representing the weights by 6 bits results in a 5 fold reduction in memory while maintaining an error rate similar to a floating point implementation. Additional reduction in memory is achieved by a technique called weight pruning,
where the weights are classified as sensitive and insensitive and the sensitive weights are represented with higher precision. A combination of these two techniques helps reduce the memory
footprint by 81 - 84% for speech recognition and keyword detection networks respectively.
Further reduction in memory size is achieved by judiciously dropping connections for large blocks of weights. The corresponding technique, termed coarse-grain sparsification, introduces
hardware-aware sparsity during DNN training, which leads to efficient weight memory compression and significant reduction in the number of computations during classification without
loss of accuracy. Keyword detection and speech recognition DNNs trained with 75% of the weights dropped and classified with 5-6 bit weight precision effectively reduced the weight memory
requirement by ~95% compared to a fully-connected network with double precision, while showing similar performance in keyword detection accuracy and word error rate.
ContributorsArunachalam, Sairam (Author) / Chakrabarti, Chaitali (Thesis advisor) / Seo, Jae-Sun (Thesis advisor) / Cao, Yu (Committee member) / Arizona State University (Publisher)
Created2016
Description
Achieving human level intelligence is a long-term goal for many Artificial Intelligence (AI) researchers. Recent developments in combining deep learning and reinforcement learning helped us to move a step forward in achieving this goal. Reinforcement learning using a delayed reward mechanism is an approach to machine intelligence which studies decision making with control and how a decision making agent can learn to act optimally in an environment-unaware conditions.
Q-learning is one of the model-free reinforcement directed learning strategies which uses temporal differences to estimate the performances of state-action pairs called Q values. A simple implementation of Q-learning algorithm can be done using a Q table memory to store and update the Q values. However, with an increase in state space data due to a complex environment, and with an increase in possible number of actions an agent can perform, Q table reaches its space limit and would be difficult to scale well. Q-learning with neural networks eliminates the use of Q table by approximating the Q function using neural networks.
Autonomous agents need to develop cognitive properties and become self-adaptive to be deployable in any environment. Reinforcement learning with Q-learning have been very efficient in solving such problems. However, embedded systems like space rovers and autonomous robots rarely implement such techniques due to the constraints faced like processing power, chip area, convergence rate and cost of the chip. These problems present a need for a portable, low power, area efficient hardware accelerator to accelerate the process of such learning.
This problem is targeted by implementing a hardware schematic architecture for Q-learning using Artificial Neural networks. This architecture exploits the massive parallelism provided by neural network with a dedicated fine grain parallelism provided by a Field Programmable Gate Array (FPGA) thereby processing the Q values at a high throughput. Mars exploration rovers currently use Xilinx-Space-grade FPGA devices for image processing, pyrotechnic operation control and obstacle avoidance. The hardware resource consumption for the architecture has been synthesized considering Xilinx Virtex7 FPGA as the target device.
Q-learning is one of the model-free reinforcement directed learning strategies which uses temporal differences to estimate the performances of state-action pairs called Q values. A simple implementation of Q-learning algorithm can be done using a Q table memory to store and update the Q values. However, with an increase in state space data due to a complex environment, and with an increase in possible number of actions an agent can perform, Q table reaches its space limit and would be difficult to scale well. Q-learning with neural networks eliminates the use of Q table by approximating the Q function using neural networks.
Autonomous agents need to develop cognitive properties and become self-adaptive to be deployable in any environment. Reinforcement learning with Q-learning have been very efficient in solving such problems. However, embedded systems like space rovers and autonomous robots rarely implement such techniques due to the constraints faced like processing power, chip area, convergence rate and cost of the chip. These problems present a need for a portable, low power, area efficient hardware accelerator to accelerate the process of such learning.
This problem is targeted by implementing a hardware schematic architecture for Q-learning using Artificial Neural networks. This architecture exploits the massive parallelism provided by neural network with a dedicated fine grain parallelism provided by a Field Programmable Gate Array (FPGA) thereby processing the Q values at a high throughput. Mars exploration rovers currently use Xilinx-Space-grade FPGA devices for image processing, pyrotechnic operation control and obstacle avoidance. The hardware resource consumption for the architecture has been synthesized considering Xilinx Virtex7 FPGA as the target device.
ContributorsGankidi, Pranay Reddy (Author) / Thangavelautham, Jekanthan (Thesis advisor) / Ren, Fengbo (Committee member) / Seo, Jae-Sun (Committee member) / Arizona State University (Publisher)
Created2016
Description
Adversarial threats of deep learning are increasingly becoming a concern due to the ubiquitous deployment of deep neural networks(DNNs) in many security-sensitive domains. Among the existing threats, adversarial weight perturbation is an emerging class of threats that attempts to perturb the weight parameters of DNNs to breach security and privacy.In this thesis, the first weight perturbation attack introduced is called Bit-Flip Attack (BFA), which can maliciously flip a small number of bits within a computer’s main memory system storing the DNN weight parameter to achieve malicious objectives. Our developed algorithm can achieve three specific attack objectives: I) Un-targeted accuracy degradation attack, ii) Targeted attack, & iii) Trojan attack. Moreover, BFA utilizes the rowhammer technique to demonstrate the bit-flip attack in an actual computer prototype. While the bit-flip attack is conducted in a white-box setting, the subsequent contribution of this thesis is to develop another novel weight perturbation attack in a black-box setting. Consequently, this thesis discusses a new study of DNN model vulnerabilities in a multi-tenant Field Programmable Gate Array (FPGA) cloud under a strict black-box framework. This newly developed attack framework injects faults in the malicious tenant by duplicating specific DNN weight packages during data transmission between off-chip memory and on-chip buffer of a victim FPGA. The proposed attack is also experimentally validated in a multi-tenant cloud FPGA prototype. In the final part, the focus shifts toward deep learning model privacy, popularly known as model extraction, that can steal partial DNN weight parameters remotely with the aid of a memory side-channel attack. In addition, a novel training algorithm is designed to utilize the partially leaked DNN weight bit information, making the model extraction attack more effective. The algorithm effectively leverages the partial leaked bit information and generates a substitute prototype of the victim model with almost identical performance to the victim.
ContributorsRakin, Adnan Siraj (Author) / Fan, Deliang (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Seo, Jae-Sun (Committee member) / Cao, Yu (Committee member) / Arizona State University (Publisher)
Created2022
Description
Deep neural networks (DNNs) have had tremendous success in a variety of
statistical learning applications due to their vast expressive power. Most
applications run DNNs on the cloud on parallelized architectures. There is a need
for for efficient DNN inference on edge with low precision hardware and analog
accelerators. To make trained models more robust for this setting, quantization and
analog compute noise are modeled as weight space perturbations to DNNs and an
information theoretic regularization scheme is used to penalize the KL-divergence
between perturbed and unperturbed models. This regularizer has similarities to
both natural gradient descent and knowledge distillation, but has the advantage of
explicitly promoting the network to and a broader minimum that is robust to
weight space perturbations. In addition to the proposed regularization,
KL-divergence is directly minimized using knowledge distillation. Initial validation
on FashionMNIST and CIFAR10 shows that the information theoretic regularizer
and knowledge distillation outperform existing quantization schemes based on the
straight through estimator or L2 constrained quantization.
statistical learning applications due to their vast expressive power. Most
applications run DNNs on the cloud on parallelized architectures. There is a need
for for efficient DNN inference on edge with low precision hardware and analog
accelerators. To make trained models more robust for this setting, quantization and
analog compute noise are modeled as weight space perturbations to DNNs and an
information theoretic regularization scheme is used to penalize the KL-divergence
between perturbed and unperturbed models. This regularizer has similarities to
both natural gradient descent and knowledge distillation, but has the advantage of
explicitly promoting the network to and a broader minimum that is robust to
weight space perturbations. In addition to the proposed regularization,
KL-divergence is directly minimized using knowledge distillation. Initial validation
on FashionMNIST and CIFAR10 shows that the information theoretic regularizer
and knowledge distillation outperform existing quantization schemes based on the
straight through estimator or L2 constrained quantization.
ContributorsKadambi, Pradyumna (Author) / Berisha, Visar (Thesis advisor) / Dasarathy, Gautam (Committee member) / Seo, Jae-Sun (Committee member) / Cao, Yu (Committee member) / Arizona State University (Publisher)
Created2019
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
The rapid growth of Internet-of-things (IoT) and artificial intelligence applications have called forth a new computing paradigm--edge computing. Edge computing applications, such as video surveillance, autonomous driving, and augmented reality, are highly computationally intensive and require real-time processing. Current edge systems are typically based on commodity general-purpose hardware such as Central Processing Units (CPUs) and Graphical Processing Units (GPUs) , which are mainly designed for large, non-time-sensitive jobs in the cloud and do not match the needs of the edge workloads. Also, these systems are usually power hungry and are not suitable for resource-constrained edge deployments. Such application-hardware mismatch calls forth a new computing backbone to support the high-bandwidth, low-latency, and energy-efficient requirements. Also, the new system should be able to support a variety of edge applications with different characteristics. This thesis addresses the above challenges by studying the use of Field Programmable Gate Array (FPGA) -based computing systems for accelerating the edge workloads, from three critical angles. First, it investigates the feasibility of FPGAs for edge computing, in comparison to conventional CPUs and GPUs. Second, it studies the acceleration of common algorithmic characteristics, identified as loop patterns, using FPGAs, and develops a benchmark tool for analyzing the performance of these patterns on different accelerators. Third, it designs a new edge computing platform using multiple clustered FPGAs to provide high-bandwidth and low-latency acceleration of convolutional neural networks (CNNs) widely used in edge applications. Finally, it studies the acceleration of the emerging neural networks, randomly-wired neural networks, on the multi-FPGA platform. The experimental results from this work show that the new generation of workloads requires rethinking the current edge-computing architecture. First, through the acceleration of common loops, it demonstrates that FPGAs can outperform GPUs in specific loops types up to 14 times. Second, it shows the linear scalability of multi-FPGA platforms in accelerating neural networks. Third, it demonstrates the superiority of the new scheduler to optimally place randomly-wired neural networks on multi-FPGA platforms with 81.1 times better throughput than the available scheduling mechanisms.
ContributorsBiookaghazadeh, Saman (Author) / Zhao, Ming (Thesis advisor) / Ren, Fengbo (Thesis advisor) / Li, Baoxin (Committee member) / Seo, Jae-Sun (Committee member) / Arizona State University (Publisher)
Created2021