Theses and Dissertations
This collection includes both ASU Theses and Dissertations, submitted by graduate students, and the Barrett, Honors College theses submitted by undergraduate students.
Displaying 1 - 2 of 2
Filtering by
- All Subjects: DNN structured sparsity
- All Subjects: Radar targets
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
As the demand for spectrum sharing between radar and communications systems is steadily increasing, the coexistence between the two systems is a growing and very challenging problem. Radar tracking in the presence of strong communications interference can result in low probability of detection even when sequential Monte Carlo
tracking methods such as the particle filter (PF) are used that better match the target kinematic model. In particular, the tracking performance can fluctuate as the power level of the communications interference can vary dynamically and unpredictably.
This work proposes to integrate the interacting multiple model (IMM) selection approach with the PF tracker to allow for dynamic variations in the power spectral density of the communications interference. The model switching allows for a necessary transition between different communications interference power spectral density (CI-PSD) values in order to reduce prediction errors. Simulations demonstrate the high performance of the integrated approach with as many as six dynamic CI-PSD value changes during the target track. For low signal-to-interference-plus-noise ratios, the derivation for estimating the high power levels of the communications interference is provided; the estimated power levels would be dynamically used in the IMM when integrated with a track-before-detect filter that is better matched to low SINR tracking applications.
tracking methods such as the particle filter (PF) are used that better match the target kinematic model. In particular, the tracking performance can fluctuate as the power level of the communications interference can vary dynamically and unpredictably.
This work proposes to integrate the interacting multiple model (IMM) selection approach with the PF tracker to allow for dynamic variations in the power spectral density of the communications interference. The model switching allows for a necessary transition between different communications interference power spectral density (CI-PSD) values in order to reduce prediction errors. Simulations demonstrate the high performance of the integrated approach with as many as six dynamic CI-PSD value changes during the target track. For low signal-to-interference-plus-noise ratios, the derivation for estimating the high power levels of the communications interference is provided; the estimated power levels would be dynamically used in the IMM when integrated with a track-before-detect filter that is better matched to low SINR tracking applications.
ContributorsZhou, Jian (Author) / Papandreou-Suppappola, Antonia (Thesis advisor) / Kovvali, Narayan (Committee member) / Berisha, Visar (Committee member) / Arizona State University (Publisher)
Created2015