Alan D. George
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Alan D. George has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Ryad Benosman
Biomedical Science Tower 3, University of Pittsburgh, Pittsburgh, PA, USA
5 shared publications - Himanshu Akolkar
Biomedical Science Tower 3, University of Pittsburgh, Pittsburgh, PA, USA
4 shared publications Seth Roffe
NASA Goddard
3 shared publications- Antony Gillette
University of Pittsburgh, Pittsburgh, Pennsylvania 15213
2 shared publications - Calvin B. Gealy
Department of Electrical and Computer Engineering, University of Pittsburgh, Pittsburgh, PA, USA
2 shared publications Christopher Wilson
Science Data Processing Branch, NASA Goddard Space Flight Center, Greenbelt, MD, USA
2 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (21) , 32464-32478
Characterizing, Modeling, and Mitigating Radiation-Induced Noise on Event-Based Vision Sensors for Space Applications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (21) , 32464-32478
Characterizing, Modeling, and Mitigating Radiation-Induced Noise on Event-Based Vision Sensors for Space Applications
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 132202-132222
A Scalable, Ultra-Low-Latency FPGA Accelerator for Neuromorphic Feature Extraction
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 132202-132222
Open AccessA Scalable, Ultra-Low-Latency FPGA Accelerator for Neuromorphic Feature Extraction
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- p. 1
Reliable Spaceflight Computing with Commercial Off-the-Shelf RISC-V FPGA SoCs
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- p. 1
Open AccessReliable Spaceflight Computing with Commercial Off-the-Shelf RISC-V FPGA SoCs
- Open Access
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Comparative Analysis of Parallel Communication Models for Regular and Irregular GPU Workloads
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Open AccessComparative Analysis of Parallel Communication Models for Regular and Irregular GPU Workloads
- Open Access
- 20 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 53420-53432
SPOC: Model Metrics and Inference Scalability
The growth of deep learning has increased the challenge of understanding model performance and scalability. Model metrics are coarse-grained values that help to capture the complexity of the model. In this research, we propose SPOC (size, parameters, operations, and Critical Datapath Length) as a suite of metrics that describe much of the complexity in deep-learning architecture and therefore help to characterize their scalability. We utilize small, specially structured “micro-models” to showcase overall scalability trends across four devices: theARMCortex-A76 embedded CPU, theAMDEPYC 9374F server CPU, the NVIDIA Jetson Orin Nano embedded GPU, and the NVIDIA A100 server GPU. Through these micro-models, we vary components of SPOC and highlight different scaling behavior across these classes of devices. Finally, we provide examples of how SPOC can be used with standard CNNs and MobileViT to capture expected and measured performance differences during inference. Through this study, we provide deep-learning practitioners with an overview of how metrics can guide our understanding of inference performance.
- 20 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 53420-53432
Open AccessSPOC: Model Metrics and Inference Scalability
The growth of deep learning has increased the challenge of understanding model performance and scalability. Model metrics are coarse-grained values that help to capture the complexity of the model. In this research, we propose SPOC (size, parameters, operations, and Critical Datapath Length) as a suite of metrics that describe much of the complexity in deep-learning architecture and therefore help to characterize their scalability. We utilize small, specially structured “micro-models” to showcase overall scalability trends across four devices: theARMCortex-A76 embedded CPU, theAMDEPYC 9374F server CPU, the NVIDIA Jetson Orin Nano embedded GPU, and the NVIDIA A100 server GPU. Through these micro-models, we vary components of SPOC and highlight different scaling behavior across these classes of devices. Finally, we provide examples of how SPOC can be used with standard CNNs and MobileViT to capture expected and measured performance differences during inference. Through this study, we provide deep-learning practitioners with an overview of how metrics can guide our understanding of inference performance.
- 10 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 61 (3) , 6629-6646
Dependable DPU Architectures on AMD-Xilinx Versal Adaptive SoCs for Space Applications
Space-computing platforms have considerable performance restrictions that are imposed by the limited onboard-processing capabilities provided by heritage flight computers. Conversely, there is a growing need for increased system autonomy enabled by deep learning (DL) to maximize performance and minimize the burden of ground-based processing. To address these limitations, domain-specific architectures with specialized acceleration hardware, such as the AMD-Xilinx Versal Adaptive System-on-Chip (SoC), have been developed. This heterogeneous platform contains significant energy-efficient compute capabilities, but it is susceptible to radiation-induced effects. Therefore, the dependability of the device must be characterized prior to inclusion on future space-computing platforms. In addition, several popular DL models exist, but each model provides unique accuracy, performance, energy-efficiency, and dependability characteristics that must be thoroughly understood. In this research, we propose a methodology for evaluating and analyzing dependable computing on AMD-Xilinx Deep Learning Processing Unit (DPU) architectures on Versal SoCs using simulated radiation-induced single-event effects (SEEs) through memory-mapped data fault injection. Using our proposed methodology, we perform this fault injection on three Versal AI Core and two Versal AI Edge DPU architectures and evaluate system performance, power consumption, energy efficiency, resource utilization, and dependability on three deployed DL models. Due to innate DPU configurability, our analysis also explores adding varying degrees of triple modular redundancy (TMR) through different DPU architectural features for increased dependability. We leveraged our fault-injection methodology to demonstrate a 24.65× average reduction in critical bits of our TMR DPU architectures compared to the unmitigated baseline, showcasing a significant increase in system dependability.
- 10 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 61 (3) , 6629-6646
Dependable DPU Architectures on AMD-Xilinx Versal Adaptive SoCs for Space Applications
Space-computing platforms have considerable performance restrictions that are imposed by the limited onboard-processing capabilities provided by heritage flight computers. Conversely, there is a growing need for increased system autonomy enabled by deep learning (DL) to maximize performance and minimize the burden of ground-based processing. To address these limitations, domain-specific architectures with specialized acceleration hardware, such as the AMD-Xilinx Versal Adaptive System-on-Chip (SoC), have been developed. This heterogeneous platform contains significant energy-efficient compute capabilities, but it is susceptible to radiation-induced effects. Therefore, the dependability of the device must be characterized prior to inclusion on future space-computing platforms. In addition, several popular DL models exist, but each model provides unique accuracy, performance, energy-efficiency, and dependability characteristics that must be thoroughly understood. In this research, we propose a methodology for evaluating and analyzing dependable computing on AMD-Xilinx Deep Learning Processing Unit (DPU) architectures on Versal SoCs using simulated radiation-induced single-event effects (SEEs) through memory-mapped data fault injection. Using our proposed methodology, we perform this fault injection on three Versal AI Core and two Versal AI Edge DPU architectures and evaluate system performance, power consumption, energy efficiency, resource utilization, and dependability on three deployed DL models. Due to innate DPU configurability, our analysis also explores adding varying degrees of triple modular redundancy (TMR) through different DPU architectural features for increased dependability. We leveraged our fault-injection methodology to demonstrate a 24.65× average reduction in critical bits of our TMR DPU architectures compared to the unmitigated baseline, showcasing a significant increase in system dependability.
- 26 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (24) , 40229-40243
Flow-Based Visual Stream Compression for Event Cameras
As the use of neuromorphic, event-based vision sensors expands, the need for compression of their output streams has increased. While their operational principle ensures event streams are spatially sparse, the high-temporal resolution of the sensors can result in high-data rates from the sensor depending on scene dynamics. For systems operating in communication-bandwidth-constrained and power-constrained environments, it is essential to compress these streams before transmitting them to a remote receiver. Therefore, we introduce a flow-based method for the real-time asynchronous compression of event streams as they are generated. This method leverages real-time optical flow estimates to predict future events without needing to transmit them, therefore, drastically reducing the amount of data transmitted. The flow-based compression introduced is evaluated using a variety of methods, including spatiotemporal distance between event streams. The introduced method itself is shown to achieve an average compression ratio (CR) of 2.70 on a variety of event-camera data sets with the evaluation configuration used. That compression is achieved with a median temporal error of 0.31 ms and an average spatiotemporal event-stream distance of 4.72. When combined with Lempel-Ziv–Markov chain algorithm compression for non-real-time applications, our method can achieve state-of-the-art average CRs ranging from 9.29 to 13.01. Additionally, we demonstrate that the proposed prediction algorithm is capable of performing real time, low-latency event prediction.
- 26 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (24) , 40229-40243
Flow-Based Visual Stream Compression for Event Cameras
As the use of neuromorphic, event-based vision sensors expands, the need for compression of their output streams has increased. While their operational principle ensures event streams are spatially sparse, the high-temporal resolution of the sensors can result in high-data rates from the sensor depending on scene dynamics. For systems operating in communication-bandwidth-constrained and power-constrained environments, it is essential to compress these streams before transmitting them to a remote receiver. Therefore, we introduce a flow-based method for the real-time asynchronous compression of event streams as they are generated. This method leverages real-time optical flow estimates to predict future events without needing to transmit them, therefore, drastically reducing the amount of data transmitted. The flow-based compression introduced is evaluated using a variety of methods, including spatiotemporal distance between event streams. The introduced method itself is shown to achieve an average compression ratio (CR) of 2.70 on a variety of event-camera data sets with the evaluation configuration used. That compression is achieved with a median temporal error of 0.31 ms and an average spatiotemporal event-stream distance of 4.72. When combined with Lempel-Ziv–Markov chain algorithm compression for non-real-time applications, our method can achieve state-of-the-art average CRs ranging from 9.29 to 13.01. Additionally, we demonstrate that the proposed prediction algorithm is capable of performing real time, low-latency event prediction.
- Open Access
- 11 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 123276-123293
Benchmarking Inference of Transformer-Based Transcription Models With Clustering on Embedded GPUs
Early awareness of inference performance ensures the feasibility of machine learning for embedded deployment. Often, ML model selection often focuses first on training performance and accuracy, with inference considered second. While prioritizing training is necessary, model inference performance is often impacted by these optimizations. Knowing whether a model will run meaningfully faster or be more energy efficient than its un-optimized-counterpart is required for resource-constrained embedded environments. Training-time optimizations may incur real performance losses at inference time. This paper benchmarks the effect of one such training optimization, clustered attention, and examines its effect on the inference performance of the transformer-based transcription model wav2vec2. The execution time and energy consumption of this model is evaluated on NVIDIA Jetson-embedded GPU devices. Clustered attention is specific to the transformer self-attention mechanism, which exhibits poor memory and execution-time scaling with respect to a variable input size. These scaling characteristics make it a potentially critical bottleneck that must be observed under realistic conditions. Our research considers three model variants: reference (original), clustered, and improved-clustered attention models. The reference model was faster for small input sizes, but the clustered model was faster for input sizes longer than 10 seconds. The clustered model had the lowest maximum energy per inference for input sizes greater than about 12 seconds of audio. With optimal configuration, the improved-clustered model takes 26.34% more time to execute than the reference model. With these operational differences, we show that inference performance and energy consumption of deployment should not be overlooked in selecting model optimizations.
- 11 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 123276-123293
Open AccessBenchmarking Inference of Transformer-Based Transcription Models With Clustering on Embedded GPUs
Early awareness of inference performance ensures the feasibility of machine learning for embedded deployment. Often, ML model selection often focuses first on training performance and accuracy, with inference considered second. While prioritizing training is necessary, model inference performance is often impacted by these optimizations. Knowing whether a model will run meaningfully faster or be more energy efficient than its un-optimized-counterpart is required for resource-constrained embedded environments. Training-time optimizations may incur real performance losses at inference time. This paper benchmarks the effect of one such training optimization, clustered attention, and examines its effect on the inference performance of the transformer-based transcription model wav2vec2. The execution time and energy consumption of this model is evaluated on NVIDIA Jetson-embedded GPU devices. Clustered attention is specific to the transformer self-attention mechanism, which exhibits poor memory and execution-time scaling with respect to a variable input size. These scaling characteristics make it a potentially critical bottleneck that must be observed under realistic conditions. Our research considers three model variants: reference (original), clustered, and improved-clustered attention models. The reference model was faster for small input sizes, but the clustered model was faster for input sizes longer than 10 seconds. The clustered model had the lowest maximum energy per inference for input sizes greater than about 12 seconds of audio. With optimal configuration, the improved-clustered model takes 26.34% more time to execute than the reference model. With these operational differences, we show that inference performance and energy consumption of deployment should not be overlooked in selecting model optimizations.
- Open Access
- 11 May 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Embedded Computing Systems
- Vol. 23 (3) , 1-35
Characterizing Parameter Scaling with Quantization for Deployment of CNNs on Real-Time Systems
Modern deep-learning models tend to include billions of parameters, reducing real-time performance. Embedded systems are compute-constrained while frequently used to deploy these models for real-time systems given size, weight, and power requirements. Tools like parameter-scaling methods help to shrink models to ease deployment. This research compares two scaling methods for convolutional neural networks, uniform scaling and NeuralScale, and analyzes their impact on inference latency, memory utilization, and power. Uniform scaling scales the number of filters evenly across a network. NeuralScale adaptively scales the model to theoretically achieve the highest accuracy for a target parameter count. In this study, VGG-11, MobileNetV2, and ResNet-50 models were scaled to four ratios: 0.25×, 0.50×, 0.75×, 1.00×. These models were benchmarked on an ARM Cortex-A72 CPU, an NVIDIA Jetson AGX Xavier GPU, and a Xilinx ZCU104 FPGA. Additionally, quantization was applied to meet real-time objectives. The CIFAR-10 and tinyImageNet datasets were studied. On CIFAR-10, NeuralScale creates more computationally intensive models than uniform scaling for the same parameter count, with relative speeds of 41% on the CPU, 72% on the GPU, and 96% on the FPGA. The additional computational complexity is a tradeoff for accuracy improvements in VGG-11 and MobileNetV2 NeuralScale models but reduced ResNet-50 NeuralScale accuracy. Furthermore, quantization alone achieves similar or better performance on the CPU and GPU devices when compared with models scaled to 0.50×, despite slight reductions in accuracy. On the GPU, quantization reduces latency by 2.7× and memory consumption by 4.3×. Uniform-scaling models are 1.8× faster and use 2.8× less memory. NeuralScale reduces latency by 1.3× and dropped memory by 1.1×. We find quantization to be a practical first tool for improved performance. Uniform scaling can easily be applied for additional improvements. NeuralScale may improve accuracy but tends to negatively impact performance, so more care must be taken with it.
- 11 May 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Embedded Computing Systems
- Vol. 23 (3) , 1-35
Open AccessCharacterizing Parameter Scaling with Quantization for Deployment of CNNs on Real-Time Systems
Modern deep-learning models tend to include billions of parameters, reducing real-time performance. Embedded systems are compute-constrained while frequently used to deploy these models for real-time systems given size, weight, and power requirements. Tools like parameter-scaling methods help to shrink models to ease deployment. This research compares two scaling methods for convolutional neural networks, uniform scaling and NeuralScale, and analyzes their impact on inference latency, memory utilization, and power. Uniform scaling scales the number of filters evenly across a network. NeuralScale adaptively scales the model to theoretically achieve the highest accuracy for a target parameter count. In this study, VGG-11, MobileNetV2, and ResNet-50 models were scaled to four ratios: 0.25×, 0.50×, 0.75×, 1.00×. These models were benchmarked on an ARM Cortex-A72 CPU, an NVIDIA Jetson AGX Xavier GPU, and a Xilinx ZCU104 FPGA. Additionally, quantization was applied to meet real-time objectives. The CIFAR-10 and tinyImageNet datasets were studied. On CIFAR-10, NeuralScale creates more computationally intensive models than uniform scaling for the same parameter count, with relative speeds of 41% on the CPU, 72% on the GPU, and 96% on the FPGA. The additional computational complexity is a tradeoff for accuracy improvements in VGG-11 and MobileNetV2 NeuralScale models but reduced ResNet-50 NeuralScale accuracy. Furthermore, quantization alone achieves similar or better performance on the CPU and GPU devices when compared with models scaled to 0.50×, despite slight reductions in accuracy. On the GPU, quantization reduces latency by 2.7× and memory consumption by 4.3×. Uniform-scaling models are 1.8× faster and use 2.8× less memory. NeuralScale reduces latency by 1.3× and dropped memory by 1.1×. We find quantization to be a practical first tool for improved performance. Uniform scaling can easily be applied for additional improvements. NeuralScale may improve accuracy but tends to negatively impact performance, so more care must be taken with it.
- 26 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 60 (4) , 5481-5498
Soft-Error Characterization and Mitigation Strategies for Edge Tensor Processing Units in Space
The Google Coral Edge Tensor Processing Unit (Edge TPU) offers low-power, high-performance capabilities ideal for enabling deep learning in space. However, as a commercial product, no reliability considerations are made in its design, and little is known about its radiation profile. As a device targeted by current and future space computing platforms, it is vital to mission success to understand the vulnerabilities and possible failure modes prior to flight. In this research, we evaluate the vulnerabilities of the Edge TPU and propose fault-mitigation techniques to improve device reliability. Several Edge TPUs were irradiated at the Los Alamos Neutron Science Center's (LANSCE) Weapons Neutron Research Facility using a wide-spectrum neutron beam. We evaluate the reliability of two machine-learning applications with common use cases within the space domain: image classification and semantic segmentation. Through experimentation, a vulnerability within the onboard memory is identified. Responsible for caching model parameters for increased performance, the onboard memory represents a critical device area. Any upsets within the cache risk compromising data integrity and model determinism. Across a variety of models tested under neutron radiation, fault accumulation and persistence are consistently observed resulting in the degradation of model accuracy and confidence. For image classification, class confidence scores experience frequent fluctuations between inferences leading to potential misclassifications. Semantic segmentation shows decreases in pixel accuracy and mean intersection-over-union (mIoU). To alleviate the impact of radiation, we propose two fault-mitigation techniques: Naive Refreshing (NR) and Golden Batch Refreshing (GBR). Naive Refreshing forces the Edge TPU to reload model parameters after every $r$ inferences. Unfortunately, reloading incurs an overhead that significantly reduces performance by increasing total inference time by up to $4.4\times$ . Finally, Golden Batch Refreshing is proposed as an alternative method to reload model parameters based on detected anomalies in the output of a pretested batch of model inputs. Periodically, the inputs from the golden batch are passed into the model and the host validates the output. If an anomaly is detected, the model parameters are reloaded prior to the next inference. GBR improves performance over NR by only incurring the overhead of the inference time to examine the model for errors rather than forcing expensive memory transactions. The golden inputs can also be interleaved between critical inferences for additional overhead reduction. By leveraging knowledge of the cache vulnerabilities and applying one or more mitigation strategies, Edge TPUs can be properly considered for integration into existing and future flight hardware.
- 26 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 60 (4) , 5481-5498
Soft-Error Characterization and Mitigation Strategies for Edge Tensor Processing Units in Space
The Google Coral Edge Tensor Processing Unit (Edge TPU) offers low-power, high-performance capabilities ideal for enabling deep learning in space. However, as a commercial product, no reliability considerations are made in its design, and little is known about its radiation profile. As a device targeted by current and future space computing platforms, it is vital to mission success to understand the vulnerabilities and possible failure modes prior to flight. In this research, we evaluate the vulnerabilities of the Edge TPU and propose fault-mitigation techniques to improve device reliability. Several Edge TPUs were irradiated at the Los Alamos Neutron Science Center's (LANSCE) Weapons Neutron Research Facility using a wide-spectrum neutron beam. We evaluate the reliability of two machine-learning applications with common use cases within the space domain: image classification and semantic segmentation. Through experimentation, a vulnerability within the onboard memory is identified. Responsible for caching model parameters for increased performance, the onboard memory represents a critical device area. Any upsets within the cache risk compromising data integrity and model determinism. Across a variety of models tested under neutron radiation, fault accumulation and persistence are consistently observed resulting in the degradation of model accuracy and confidence. For image classification, class confidence scores experience frequent fluctuations between inferences leading to potential misclassifications. Semantic segmentation shows decreases in pixel accuracy and mean intersection-over-union (mIoU). To alleviate the impact of radiation, we propose two fault-mitigation techniques: Naive Refreshing (NR) and Golden Batch Refreshing (GBR). Naive Refreshing forces the Edge TPU to reload model parameters after every $r$ inferences. Unfortunately, reloading incurs an overhead that significantly reduces performance by increasing total inference time by up to $4.4\times$ . Finally, Golden Batch Refreshing is proposed as an alternative method to reload model parameters based on detected anomalies in the output of a pretested batch of model inputs. Periodically, the inputs from the golden batch are passed into the model and the host validates the output. If an anomaly is detected, the model parameters are reloaded prior to the next inference. GBR improves performance over NR by only incurring the overhead of the inference time to examine the model for errors rather than forcing expensive memory transactions. The golden inputs can also be interleaved between critical inferences for additional overhead reduction. By leveraging knowledge of the cache vulnerabilities and applying one or more mitigation strategies, Edge TPUs can be properly considered for integration into existing and future flight hardware.