Mohammad Javad Shafiee

Jia Cheng Shang, Yuhao Chen, Mohammad Javad Shafiee et al.

Hockey rink registration is a useful tool for aiding and automating sports analysis. When combined with player tracking, it can provide location information of players on the rink by estimating a homography matrix that can warp broadcast video frames onto an overhead template of the rink, or vice versa. However, most existing techniques require accurate ground truth information, which can take many hours to annotate, and only work on the trained rink types. In this paper, we propose a generalized rink registration pipeline that, once trained, can be applied to both seen and unseen rink types with only an overhead rink template and the video frame as inputs. Our pipeline uses domain adaptation techniques, semi-supervised learning, and synthetic data during training to achieve this ability and overcome the lack of non-NHL training data. The proposed method is evaluated on both NHL (source) and non-NHL (target) rink data and the results demonstrate that our approach can generalize to non-NHL rinks, while maintaining competitive performance on NHL rinks.

Alexander Wong, Yifan Wu, Saad Abbasi et al.

Multi-task learning has shown considerable promise for improving the performance of deep learning-driven vision systems for the purpose of robotic grasping. However, high architectural and computational complexity can result in poor suitability for deployment on embedded devices that are typically leveraged in robotic arms for real-world manufacturing and warehouse environments. As such, the design of highly efficient multi-task deep neural network architectures tailored for computer vision tasks for robotic grasping on the edge is highly desired for widespread adoption in manufacturing environments. Motivated by this, we propose Fast GraspNeXt, a fast self-attention neural network architecture tailored for embedded multi-task learning in computer vision tasks for robotic grasping. To build Fast GraspNeXt, we leverage a generative network architecture search strategy with a set of architectural constraints customized to achieve a strong balance between multi-task learning performance and embedded inference efficiency. Experimental results on the MetaGraspNet benchmark dataset show that the Fast GraspNeXt network design achieves the highest performance (average precision (AP), accuracy, and mean squared error (MSE)) across multiple computer vision tasks when compared to other efficient multi-task network architecture designs, while having only 17.8M parameters (about >5x smaller), 259 GFLOPs (as much as >5x lower) and as much as >3.15x faster on a NVIDIA Jetson TX2 embedded processor.

Saeejith Nair, Saad Abbasi, Alexander Wong et al.

Neural Architecture Search (NAS) has enabled automatic discovery of more efficient neural network architectures, especially for mobile and embedded vision applications. Although recent research has proposed ways of quickly estimating latency on unseen hardware devices with just a few samples, little focus has been given to the challenges of estimating latency on runtimes using optimized graphs, such as TensorRT and specifically for edge devices. In this work, we propose MAPLE-Edge, an edge device-oriented extension of MAPLE, the state-of-the-art latency predictor for general purpose hardware, where we train a regression network on architecture-latency pairs in conjunction with a hardware-runtime descriptor to effectively estimate latency on a diverse pool of edge devices. Compared to MAPLE, MAPLE-Edge can describe the runtime and target device platform using a much smaller set of CPU performance counters that are widely available on all Linux kernels, while still achieving up to +49.6% accuracy gains against previous state-of-the-art baseline methods on optimized edge device runtimes, using just 10 measurements from an unseen target device. We also demonstrate that unlike MAPLE which performs best when trained on a pool of devices sharing a common runtime, MAPLE-Edge can effectively generalize across runtimes by applying a trick of normalizing performance counters by the operator latency, in the measured hardware-runtime descriptor. Lastly, we show that for runtimes exhibiting lower than desired accuracy, performance can be boosted by collecting additional samples from the target device, with an extra 90 samples translating to gains of nearly +40%.

Alexander Wong, Mohammad Javad Shafiee, Saad Abbasi et al.

With the growing adoption of deep learning for on-device TinyML applications, there has been an ever-increasing demand for efficient neural network backbones optimized for the edge. Recently, the introduction of attention condenser networks have resulted in low-footprint, highly-efficient, self-attention neural networks that strike a strong balance between accuracy and speed. In this study, we introduce a faster attention condenser design called double-condensing attention condensers that allow for highly condensed feature embeddings. We further employ a machine-driven design exploration strategy that imposes design constraints based on best practices for greater efficiency and robustness to produce the macro-micro architecture constructs of the backbone. The resulting backbone (which we name AttendNeXt) achieves significantly higher inference throughput on an embedded ARM processor when compared to several other state-of-the-art efficient backbones (>10x faster than FB-Net C at higher accuracy and speed and >10x faster than MobileOne-S1 at smaller size) while having a small model size (>1.37x smaller than MobileNetv3-L at higher accuracy and speed) and strong accuracy (1.1% higher top-1 accuracy than MobileViT XS on ImageNet at higher speed). These promising results demonstrate that exploring different efficient architecture designs and self-attention mechanisms can lead to interesting new building blocks for TinyML applications.

Carol Xu, Mahmoud Famouri, Gautam Bathla et al.

Photovoltaic cells are electronic devices that convert light energy to electricity, forming the backbone of solar energy harvesting systems. An essential step in the manufacturing process for photovoltaic cells is visual quality inspection using electroluminescence imaging to identify defects such as cracks, finger interruptions, and broken cells. A big challenge faced by industry in photovoltaic cell visual inspection is the fact that it is currently done manually by human inspectors, which is extremely time consuming, laborious, and prone to human error. While deep learning approaches holds great potential to automating this inspection, the hardware resource-constrained manufacturing scenario makes it challenging for deploying complex deep neural network architectures. In this work, we introduce CellDefectNet, a highly efficient attention condenser network designed via machine-driven design exploration specifically for electroluminesence-based photovoltaic cell defect detection on the edge. We demonstrate the efficacy of CellDefectNet on a benchmark dataset comprising of a diversity of photovoltaic cells captured using electroluminescence imagery, achieving an accuracy of ~86.3% while possessing just 410K parameters (~13$\times$ lower than EfficientNet-B0, respectively) and ~115M FLOPs (~12$\times$ lower than EfficientNet-B0) and ~13$\times$ faster on an ARM Cortex A-72 embedded processor when compared to EfficientNet-B0.

Hossein Aboutalebi, Maya Pavlova, Mohammad Javad Shafiee et al.

Since the World Health Organization declared COVID-19 a pandemic in 2020, the global community has faced ongoing challenges in controlling and mitigating the transmission of the SARS-CoV-2 virus, as well as its evolving subvariants and recombinants. A significant challenge during the pandemic has not only been the accurate detection of positive cases but also the efficient prediction of risks associated with complications and patient survival probabilities. These tasks entail considerable clinical resource allocation and attention.In this study, we introduce COVID-Net Biochem, a versatile and explainable framework for constructing machine learning models. We apply this framework to predict COVID-19 patient survival and the likelihood of developing Acute Kidney Injury during hospitalization, utilizing clinical and biochemical data in a transparent, systematic approach. The proposed approach advances machine learning model design by seamlessly integrating domain expertise with explainability tools, enabling model decisions to be based on key biomarkers. This fosters a more transparent and interpretable decision-making process made by machines specifically for medical applications.

Amir Nazemi, Mohammad Javad Shafiee, Zahra Gharaee et al.

Recent state-of-the-art semi-supervised Video Object Segmentation (VOS) methods have shown significant improvements in target object segmentation accuracy when information from preceding frames is used in segmenting the current frame. In particular, such memory-based approaches can help a model to more effectively handle appearance changes (representation drift) or occlusions. Ideally, for maximum performance, Online VOS methods would need all or most of the preceding frames (or their extracted information) to be stored in memory and be used for online learning in later frames. Such a solution is not feasible for long videos, as the required memory size grows without bound, and such methods can fail when memory is limited and a target object experiences repeated representation drifts throughout a video. We propose two novel techniques to reduce the memory requirement of Online VOS methods while improving modeling accuracy and generalization on long videos. Motivated by the success of continual learning techniques in preserving previously-learned knowledge, here we propose Gated-Regularizer Continual Learning (GRCL), which improves the performance of any Online VOS subject to limited memory, and a Reconstruction-based Memory Selection Continual Learning (RMSCL), which empowers Online VOS methods to efficiently benefit from stored information in memory. We also analyze the performance of a hybrid combination of the two proposed methods. Experimental results show that the proposed methods are able to improve the performance of Online VOS models by more than 8%, with improved robustness on long-video datasets while maintaining comparable performance on short-video datasets such as DAVIS16, DAVIS17, and YouTube-VOS18.

Carol Xu, Mahmoud Famouri, Gautam Bathla et al.

Light guide plates are essential optical components widely used in a diverse range of applications ranging from medical lighting fixtures to back-lit TV displays. In this work, we introduce a fully-integrated, high-throughput, high-performance deep learning-driven workflow for light guide plate surface visual quality inspection (VQI) tailored for real-world manufacturing environments. To enable automated VQI on the edge computing within the fully-integrated VQI system, a highly compact deep anti-aliased attention condenser neural network (which we name LightDefectNet) tailored specifically for light guide plate surface defect detection in resource-constrained scenarios was created via machine-driven design exploration with computational and "best-practices" constraints as well as L_1 paired classification discrepancy loss. Experiments show that LightDetectNet achieves a detection accuracy of ~98.2% on the LGPSDD benchmark while having just 770K parameters (~33X and ~6.9X lower than ResNet-50 and EfficientNet-B0, respectively) and ~93M FLOPs (~88X and ~8.4X lower than ResNet-50 and EfficientNet-B0, respectively) and ~8.8X faster inference speed than EfficientNet-B0 on an embedded ARM processor. As such, the proposed deep learning-driven workflow, integrated with the aforementioned LightDefectNet neural network, is highly suited for high-throughput, high-performance light plate surface VQI within real-world manufacturing environments.

Saeejith Nair, Yuhao Chen, Mohammad Javad Shafiee et al.

Neural radiance fields (NeRFs) enable high-quality novel view synthesis, but their high computational complexity limits deployability. While existing neural-based solutions strive for efficiency, they use one-size-fits-all architectures regardless of scene complexity. The same architecture may be unnecessarily large for simple scenes but insufficient for complex ones. Thus, there is a need to dynamically optimize the neural network component of NeRFs to achieve a balance between computational complexity and specific targets for synthesis quality. We introduce NAS-NeRF, a generative neural architecture search strategy that generates compact, scene-specialized NeRF architectures by balancing architecture complexity and target synthesis quality metrics. Our method incorporates constraints on target metrics and budgets to guide the search towards architectures tailored for each scene. Experiments on the Blender synthetic dataset show the proposed NAS-NeRF can generate architectures up to 5.74$\times$ smaller, with 4.19$\times$ fewer FLOPs, and 1.93$\times$ faster on a GPU than baseline NeRFs, without suffering a drop in SSIM. Furthermore, we illustrate that NAS-NeRF can also achieve architectures up to 23$\times$ smaller, with 22$\times$ fewer FLOPs, and 4.7$\times$ faster than baseline NeRFs with only a 5.3% average SSIM drop. Our source code is also made publicly available at https://saeejithnair.github.io/NAS-NeRF.

Saad Abbasi, Alexander Wong, Mohammad Javad Shafiee

Deep neural network (DNN) latency characterization is a time-consuming process and adds significant cost to Neural Architecture Search (NAS) processes when searching for efficient convolutional neural networks for embedded vision applications. DNN Latency is a hardware dependent metric and requires direct measurement or inference on target hardware. A recently introduced latency estimation technique known as MAPLE predicts DNN execution time on previously unseen hardware devices by using hardware performance counters. Leveraging these hardware counters in the form of an implicit prior, MAPLE achieves state-of-the-art performance in latency prediction. Here, we propose MAPLE-X which extends MAPLE by incorporating explicit prior knowledge of hardware devices and DNN architecture latency to better account for model stability and robustness. First, by identifying DNN architectures that exhibit a similar latency to each other, we can generate multiple virtual examples to significantly improve the accuracy over MAPLE. Secondly, the hardware specifications are used to determine the similarity between training and test hardware to emphasize training samples captured from comparable devices (domains) and encourages improved domain alignment. Experimental results using a convolution neural network NAS benchmark across different types of devices, including an Intel processor that is now used for embedded vision applications, demonstrate a 5% improvement over MAPLE and 9% over HELP. Furthermore, we include ablation studies to independently assess the benefits of virtual examples and hardware-based sample importance.

Carol Xu, Mahmoud Famouri, Gautam Bathla et al.

Light guide plates are essential optical components widely used in a diverse range of applications ranging from medical lighting fixtures to back-lit TV displays. An essential step in the manufacturing of light guide plates is the quality inspection of defects such as scratches, bright/dark spots, and impurities. This is mainly done in industry through manual visual inspection for plate pattern irregularities, which is time-consuming and prone to human error and thus act as a significant barrier to high-throughput production. Advances in deep learning-driven computer vision has led to the exploration of automated visual quality inspection of light guide plates to improve inspection consistency, accuracy, and efficiency. However, given the cost constraints in visual inspection scenarios, the widespread adoption of deep learning-driven computer vision methods for inspecting light guide plates has been greatly limited due to high computational requirements. In this study, we explore the utilization of machine-driven design exploration with computational and "best-practices" constraints as well as L$_1$ paired classification discrepancy loss to create LightDefectNet, a highly compact deep anti-aliased attention condenser neural network architecture tailored specifically for light guide plate surface defect detection in resource-constrained scenarios. Experiments show that LightDetectNet achieves a detection accuracy of $\sim$98.2% on the LGPSDD benchmark while having just 770K parameters ($\sim$33$\times$ and $\sim$6.9$\times$ lower than ResNet-50 and EfficientNet-B0, respectively) and $\sim$93M FLOPs ($\sim$88$\times$ and $\sim$8.4$\times$ lower than ResNet-50 and EfficientNet-B0, respectively) and $\sim$8.8$\times$ faster inference speed than EfficientNet-B0 on an embedded ARM processor.

Hossein Aboutalebi, Maya Pavlova, Mohammad Javad Shafiee et al.

The world is still struggling in controlling and containing the spread of the COVID-19 pandemic caused by the SARS-CoV-2 virus. The medical conditions associated with SARS-CoV-2 infections have resulted in a surge in the number of patients at clinics and hospitals, leading to a significantly increased strain on healthcare resources. As such, an important part of managing and handling patients with SARS-CoV-2 infections within the clinical workflow is severity assessment, which is often conducted with the use of chest x-ray (CXR) images. In this work, we introduce COVID-Net CXR-S, a convolutional neural network for predicting the airspace severity of a SARS-CoV-2 positive patient based on a CXR image of the patient's chest. More specifically, we leveraged transfer learning to transfer representational knowledge gained from over 16,000 CXR images from a multinational cohort of over 15,000 patient cases into a custom network architecture for severity assessment. Experimental results with a multi-national patient cohort curated by the Radiological Society of North America (RSNA) RICORD initiative showed that the proposed COVID-Net CXR-S has potential to be a powerful tool for computer-aided severity assessment of CXR images of COVID-19 positive patients. Furthermore, radiologist validation on select cases by two board-certified radiologists with over 10 and 19 years of experience, respectively, showed consistency between radiologist interpretation and critical factors leveraged by COVID-Net CXR-S for severity assessment. While not a production-ready solution, the ultimate goal for the open source release of COVID-Net CXR-S is to act as a catalyst for clinical scientists, machine learning researchers, as well as citizen scientists to develop innovative new clinical decision support solutions for helping clinicians around the world manage the continuing pandemic.

Saeejith Nair, Mohammad Javad Shafiee, Alexander Wong

We present DARLEI, a framework that combines evolutionary algorithms with parallelized reinforcement learning for efficiently training and evolving populations of UNIMAL agents. Our approach utilizes Proximal Policy Optimization (PPO) for individual agent learning and pairs it with a tournament selection-based generational learning mechanism to foster morphological evolution. By building on Nvidia's Isaac Gym, DARLEI leverages GPU accelerated simulation to achieve over 20x speedup using just a single workstation, compared to previous work which required large distributed CPU clusters. We systematically characterize DARLEI's performance under various conditions, revealing factors impacting diversity of evolved morphologies. For example, by enabling inter-agent collisions within the simulator, we find that we can simulate some multi-agent interactions between the same morphology, and see how it influences individual agent capabilities and long-term evolutionary adaptation. While current results demonstrate limited diversity across generations, we hope to extend DARLEI in future work to include interactions between diverse morphologies in richer environments, and create a platform that allows for coevolving populations and investigating emergent behaviours in them. Our source code is also made publicly at https://saeejithnair.github.io/darlei.

Saad Abbasi, Alexander Wong, Mohammad Javad Shafiee

Modern deep neural networks must demonstrate state-of-the-art accuracy while exhibiting low latency and energy consumption. As such, neural architecture search (NAS) algorithms take these two constraints into account when generating a new architecture. However, efficiency metrics such as latency are typically hardware dependent requiring the NAS algorithm to either measure or predict the architecture latency. Measuring the latency of every evaluated architecture adds a significant amount of time to the NAS process. Here we propose Microprocessor A Priori for Latency Estimation MAPLE that does not rely on transfer learning or domain adaptation but instead generalizes to new hardware by incorporating a prior hardware characteristics during training. MAPLE takes advantage of a novel quantitative strategy to characterize the underlying microprocessor by measuring relevant hardware performance metrics, yielding a fine-grained and expressive hardware descriptor. Moreover, the proposed MAPLE benefits from the tightly coupled I/O between the CPU and GPU and their dependency to predict DNN latency on GPUs while measuring microprocessor performance hardware counters from the CPU feeding the GPU hardware. Through this quantitative strategy as the hardware descriptor, MAPLE can generalize to new hardware via a few shot adaptation strategy where with as few as 3 samples it exhibits a 6% improvement over state-of-the-art methods requiring as much as 10 samples. Experimental results showed that, increasing the few shot adaptation samples to 10 improves the accuracy significantly over the state-of-the-art methods by 12%. Furthermore, it was demonstrated that MAPLE exhibiting 8-10% better accuracy, on average, compared to relevant baselines at any number of adaptation samples.

Mohammad Javad Shafiee, Mahmoud Famouri, Gautam Bathla et al.

A critical aspect in the manufacturing process is the visual quality inspection of manufactured components for defects and flaws. Human-only visual inspection can be very time-consuming and laborious, and is a significant bottleneck especially for high-throughput manufacturing scenarios. Given significant advances in the field of deep learning, automated visual quality inspection can lead to highly efficient and reliable detection of defects and flaws during the manufacturing process. However, deep learning-driven visual inspection methods often necessitate significant computational resources, thus limiting throughput and act as a bottleneck to widespread adoption for enabling smart factories. In this study, we investigated the utilization of a machine-driven design exploration approach to create TinyDefectNet, a highly compact deep convolutional network architecture tailored for high-throughput manufacturing visual quality inspection. TinyDefectNet comprises of just ~427K parameters and has a computational complexity of ~97M FLOPs, yet achieving a detection accuracy of a state-of-the-art architecture for the task of surface defect detection on the NEU defect benchmark dataset. As such, TinyDefectNet can achieve the same level of detection performance at 52$\times$ lower architectural complexity and 11x lower computational complexity. Furthermore, TinyDefectNet was deployed on an AMD EPYC 7R32, and achieved 7.6x faster throughput using the native Tensorflow environment and 9x faster throughput using AMD ZenDNN accelerator library. Finally, explainability-driven performance validation strategy was conducted to ensure correct decision-making behaviour was exhibited by TinyDefectNet to improve trust in its usage by operators and inspectors.

Hossein Aboutalebi, Maya Pavlova, Hayden Gunraj et al.

Medical image analysis continues to hold interesting challenges given the subtle characteristics of certain diseases and the significant overlap in appearance between diseases. In this work, we explore the concept of self-attention for tackling such subtleties in and between diseases. To this end, we introduce MEDUSA, a multi-scale encoder-decoder self-attention mechanism tailored for medical image analysis. While self-attention deep convolutional neural network architectures in existing literature center around the notion of multiple isolated lightweight attention mechanisms with limited individual capacities being incorporated at different points in the network architecture, MEDUSA takes a significant departure from this notion by possessing a single, unified self-attention mechanism with significantly higher capacity with multiple attention heads feeding into different scales in the network architecture. To the best of the authors' knowledge, this is the first "single body, multi-scale heads" realization of self-attention and enables explicit global context amongst selective attention at different levels of representational abstractions while still enabling differing local attention context at individual levels of abstractions. With MEDUSA, we obtain state-of-the-art performance on multiple challenging medical image analysis benchmarks including COVIDx, RSNA RICORD, and RSNA Pneumonia Challenge when compared to previous work. Our MEDUSA model is publicly available.

Saad Abbasi, Mohammad Javad Shafiee, Ellick Chan et al.

The fine-grained relationship between form and function with respect to deep neural network architecture design and hardware-specific acceleration is one area that is not well studied in the research literature, with form often dictated by accuracy as opposed to hardware function. In this study, a comprehensive empirical exploration is conducted to investigate the impact of deep neural network architecture design on the degree of inference speedup that can be achieved via hardware-specific acceleration. More specifically, we empirically study the impact of a variety of commonly used macro-architecture design patterns across different architectural depths through the lens of OpenVINO microprocessor-specific and GPU-specific acceleration. Experimental results showed that while leveraging hardware-specific acceleration achieved an average inference speed-up of 380%, the degree of inference speed-up varied drastically depending on the macro-architecture design pattern, with the greatest speedup achieved on the depthwise bottleneck convolution design pattern at 550%. Furthermore, we conduct an in-depth exploration of the correlation between FLOPs requirement, level 3 cache efficacy, and network latency with increasing architectural depth and width. Finally, we analyze the inference time reductions using hardware-specific acceleration when compared to native deep learning frameworks across a wide variety of hand-crafted deep convolutional neural network architecture designs as well as ones found via neural architecture search strategies. We found that the DARTS-derived architecture to benefit from the greatest improvement from hardware-specific software acceleration (1200%) while the depthwise bottleneck convolution-based MobileNet-V2 to have the lowest overall inference time of around 2.4 ms.

Hossein Aboutalebi, Mohammad Javad Shafiee, Michelle Karg et al.

Despite the significant advances in deep learning over the past decade, a major challenge that limits the wide-spread adoption of deep learning has been their fragility to adversarial attacks. This sensitivity to making erroneous predictions in the presence of adversarially perturbed data makes deep neural networks difficult to adopt for certain real-world, mission-critical applications. While much of the research focus has revolved around adversarial example creation and adversarial hardening, the area of performance measures for assessing adversarial robustness is not well explored. Motivated by this, this study presents the concept of residual error, a new performance measure for not only assessing the adversarial robustness of a deep neural network at the individual sample level, but also can be used to differentiate between adversarial and non-adversarial examples to facilitate for adversarial example detection. Furthermore, we introduce a hybrid model for approximating the residual error in a tractable manner. Experimental results using the case of image classification demonstrates the effectiveness and efficacy of the proposed residual error metric for assessing several well-known deep neural network architectures. These results thus illustrate that the proposed measure could be a useful tool for not only assessing the robustness of deep neural networks used in mission-critical scenarios, but also in the design of adversarially robust models.

Hossein Aboutalebi, Saad Abbasi, Mohammad Javad Shafiee et al.

The health and socioeconomic difficulties caused by the COVID-19 pandemic continues to cause enormous tensions around the world. In particular, this extraordinary surge in the number of cases has put considerable strain on health care systems around the world. A critical step in the treatment and management of COVID-19 positive patients is severity assessment, which is challenging even for expert radiologists given the subtleties at different stages of lung disease severity. Motivated by this challenge, we introduce COVID-Net CT-S, a suite of deep convolutional neural networks for predicting lung disease severity due to COVID-19 infection. More specifically, a 3D residual architecture design is leveraged to learn volumetric visual indicators characterizing the degree of COVID-19 lung disease severity. Experimental results using the patient cohort collected by the China National Center for Bioinformation (CNCB) showed that the proposed COVID-Net CT-S networks, by leveraging volumetric features, can achieve significantly improved severity assessment performance when compared to traditional severity assessment networks that learn and leverage 2D visual features to characterize COVID-19 severity.

Saad Abbasi, Mahmoud Famouri, Mohammad Javad Shafiee et al.

Human operators often diagnose industrial machinery via anomalous sounds. Automated acoustic anomaly detection can lead to reliable maintenance of machinery. However, deep learning-driven anomaly detection methods often require an extensive amount of computational resources which prohibits their deployment in factories. Here we explore a machine-driven design exploration strategy to create OutlierNets, a family of highly compact deep convolutional autoencoder network architectures featuring as few as 686 parameters, model sizes as small as 2.7 KB, and as low as 2.8 million FLOPs, with a detection accuracy matching or exceeding published architectures with as many as 4 million parameters. Furthermore, CPU-accelerated latency experiments show that the OutlierNet architectures can achieve as much as 21x lower latency than published networks.

Ahmadreza Jeddi, Mohammad Javad Shafiee, Alexander Wong

Adversarial Training (AT) with Projected Gradient Descent (PGD) is an effective approach for improving the robustness of the deep neural networks. However, PGD AT has been shown to suffer from two main limitations: i) high computational cost, and ii) extreme overfitting during training that leads to reduction in model generalization. While the effect of factors such as model capacity and scale of training data on adversarial robustness have been extensively studied, little attention has been paid to the effect of a very important parameter in every network optimization on adversarial robustness: the learning rate. In particular, we hypothesize that effective learning rate scheduling during adversarial training can significantly reduce the overfitting issue, to a degree where one does not even need to adversarially train a model from scratch but can instead simply adversarially fine-tune a pre-trained model. Motivated by this hypothesis, we propose a simple yet very effective adversarial fine-tuning approach based on a $\textit{slow start, fast decay}$ learning rate scheduling strategy which not only significantly decreases computational cost required, but also greatly improves the accuracy and robustness of a deep neural network. Experimental results show that the proposed adversarial fine-tuning approach outperforms the state-of-the-art methods on CIFAR-10, CIFAR-100 and ImageNet datasets in both test accuracy and the robustness, while reducing the computational cost by 8-10$\times$. Furthermore, a very important benefit of the proposed adversarial fine-tuning approach is that it enables the ability to improve the robustness of any pre-trained deep neural network without needing to train the model from scratch, which to the best of the authors' knowledge has not been previously demonstrated in research literature.

Alexander Wong, Mahmoud Famouri, Mohammad Javad Shafiee

While significant advances in deep learning has resulted in state-of-the-art performance across a large number of complex visual perception tasks, the widespread deployment of deep neural networks for TinyML applications involving on-device, low-power image recognition remains a big challenge given the complexity of deep neural networks. In this study, we introduce AttendNets, low-precision, highly compact deep neural networks tailored for on-device image recognition. More specifically, AttendNets possess deep self-attention architectures based on visual attention condensers, which extends on the recently introduced stand-alone attention condensers to improve spatial-channel selective attention. Furthermore, AttendNets have unique machine-designed macroarchitecture and microarchitecture designs achieved via a machine-driven design exploration strategy. Experimental results on ImageNet$_{50}$ benchmark dataset for the task of on-device image recognition showed that AttendNets have significantly lower architectural and computational complexity when compared to several deep neural networks in research literature designed for efficiency while achieving highest accuracies (with the smallest AttendNet achieving $\sim$7.2% higher accuracy, while requiring $\sim$3$\times$ fewer multiply-add operations, $\sim$4.17$\times$ fewer parameters, and $\sim$16.7$\times$ lower weight memory requirements than MobileNet-V1). Based on these promising results, AttendNets illustrate the effectiveness of visual attention condensers as building blocks for enabling various on-device visual perception tasks for TinyML applications.

Hossein Aboutalebi, Mohammad Javad Shafiee, Michelle Karg et al.

Prior studies have unveiled the vulnerability of the deep neural networks in the context of adversarial machine learning, leading to great recent attention into this area. One interesting question that has yet to be fully explored is the bias-variance relationship of adversarial machine learning, which can potentially provide deeper insights into this behaviour. The notion of bias and variance is one of the main approaches to analyze and evaluate the generalization and reliability of a machine learning model. Although it has been extensively used in other machine learning models, it is not well explored in the field of deep learning and it is even less explored in the area of adversarial machine learning. In this study, we investigate the effect of adversarial machine learning on the bias and variance of a trained deep neural network and analyze how adversarial perturbations can affect the generalization of a network. We derive the bias-variance trade-off for both classification and regression applications based on two main loss functions: (i) mean squared error (MSE), and (ii) cross-entropy. Furthermore, we perform quantitative analysis with both simulated and real data to empirically evaluate consistency with the derived bias-variance tradeoffs. Our analysis sheds light on why the deep neural networks have poor performance under adversarial perturbation from a bias-variance point of view and how this type of perturbation would change the performance of a network. Moreover, given these new theoretical findings, we introduce a new adversarial machine learning algorithm with lower computational complexity than well-known adversarial machine learning strategies (e.g., PGD) while providing a high success rate in fooling deep neural networks in lower perturbation magnitudes.

Mohammad Javad Shafiee, Ahmadreza Jeddi, Amir Nazemi et al.

This paper analyzes the robustness of deep learning models in autonomous driving applications and discusses the practical solutions to address that.

Ahmadreza Jeddi, Mohammad Javad Shafiee, Michelle Karg et al.

While deep neural networks have been achieving state-of-the-art performance across a wide variety of applications, their vulnerability to adversarial attacks limits their widespread deployment for safety-critical applications. Alongside other adversarial defense approaches being investigated, there has been a very recent interest in improving adversarial robustness in deep neural networks through the introduction of perturbations during the training process. However, such methods leverage fixed, pre-defined perturbations and require significant hyper-parameter tuning that makes them very difficult to leverage in a general fashion. In this study, we introduce Learn2Perturb, an end-to-end feature perturbation learning approach for improving the adversarial robustness of deep neural networks. More specifically, we introduce novel perturbation-injection modules that are incorporated at each layer to perturb the feature space and increase uncertainty in the network. This feature perturbation is performed at both the training and the inference stages. Furthermore, inspired by the Expectation-Maximization, an alternating back-propagation training algorithm is introduced to train the network and noise parameters consecutively. Experimental results on CIFAR-10 and CIFAR-100 datasets show that the proposed Learn2Perturb method can result in deep neural networks which are $4-7\%$ more robust on $l_{\infty}$ FGSM and PDG adversarial attacks and significantly outperforms the state-of-the-art against $l_2$ $C\&W$ attack and a wide range of well-known black-box attacks.

Zhong Qiu Lin, Mohammad Javad Shafiee, Stanislav Bochkarev et al.

There has been a significant surge of interest recently around the concept of explainable artificial intelligence (XAI), where the goal is to produce an interpretation for a decision made by a machine learning algorithm. Of particular interest is the interpretation of how deep neural networks make decisions, given the complexity and `black box' nature of such networks. Given the infancy of the field, there has been very limited exploration into the assessment of the performance of explainability methods, with most evaluations centered around subjective visual interpretation of the produced interpretations. In this study, we explore a more machine-centric strategy for quantifying the performance of explainability methods on deep neural networks via the notion of decision-making impact analysis. We introduce two quantitative performance metrics: i) Impact Score, which assesses the percentage of critical factors with either strong confidence reduction impact or decision changing impact, and ii) Impact Coverage, which assesses the percentage coverage of adversarially impacted factors in the input. A comprehensive analysis using this approach was conducted on several state-of-the-art explainability methods (LIME, SHAP, Expected Gradients, GSInquire) on a ResNet-50 deep convolutional neural network using a subset of ImageNet for the task of image classification. Experimental results show that the critical regions identified by LIME within the tested images had the lowest impact on the decision-making process of the network (~38%), with progressive increase in decision-making impact for SHAP (~44%), Expected Gradients (~51%), and GSInquire (~76%). While by no means perfect, the hope is that the proposed machine-centric strategy helps push the conversation forward towards better metrics for evaluating explainability methods and improve trust in deep neural networks.

Mohammad Javad Shafiee, Andrew Hryniowski, Francis Li et al.

The design of compact deep neural networks is a crucial task to enable widespread adoption of deep neural networks in the real-world, particularly for edge and mobile scenarios. Due to the time-consuming and challenging nature of manually designing compact deep neural networks, there has been significant recent research interest into algorithms that automatically search for compact network architectures. A particularly interesting class of compact architecture search algorithms are those that are guided by baseline network architectures. Such algorithms have been shown to be significantly more computationally efficient than unguided methods. In this study, we explore the current state of compact architecture search for deep neural networks through both theoretical and empirical analysis of four different state-of-the-art compact architecture search algorithms: i) group lasso regularization, ii) variational dropout, iii) MorphNet, and iv) Generative Synthesis. We examine these methods in detail based on a number of different factors such as efficiency, effectiveness, and scalability. Furthermore, empirical evaluations are conducted to compare the efficacy of these compact architecture search algorithms across three well-known benchmark datasets. While by no means an exhaustive exploration, we hope that this study helps provide insights into the interesting state of this relatively new area of research in terms of diversity and real, tangible gains already achieved in architecture design improvements. Furthermore, the hope is that this study would help in pushing the conversation forward towards a deeper theoretical and empirical understanding where the research community currently stands in the landscape of compact architecture search for deep neural networks, and the practical challenges and considerations in leveraging such approaches for operational usage.

Alexander Wong, Mahmoud Famuori, Mohammad Javad Shafiee et al.

Object detection remains an active area of research in the field of computer vision, and considerable advances and successes has been achieved in this area through the design of deep convolutional neural networks for tackling object detection. Despite these successes, one of the biggest challenges to widespread deployment of such object detection networks on edge and mobile scenarios is the high computational and memory requirements. As such, there has been growing research interest in the design of efficient deep neural network architectures catered for edge and mobile usage. In this study, we introduce YOLO Nano, a highly compact deep convolutional neural network for the task of object detection. A human-machine collaborative design strategy is leveraged to create YOLO Nano, where principled network design prototyping, based on design principles from the YOLO family of single-shot object detection network architectures, is coupled with machine-driven design exploration to create a compact network with highly customized module-level macroarchitecture and microarchitecture designs tailored for the task of embedded object detection. The proposed YOLO Nano possesses a model size of ~4.0MB (>15.1x and >8.3x smaller than Tiny YOLOv2 and Tiny YOLOv3, respectively) and requires 4.57B operations for inference (>34% and ~17% lower than Tiny YOLOv2 and Tiny YOLOv3, respectively) while still achieving an mAP of ~69.1% on the VOC 2007 dataset (~12% and ~10.7% higher than Tiny YOLOv2 and Tiny YOLOv3, respectively). Experiments on inference speed and power efficiency on a Jetson AGX Xavier embedded module at different power budgets further demonstrate the efficacy of YOLO Nano for embedded scenarios.

Mohammad Javad Shafiee, Mirko Nentwig, Yohannes Kassahun et al.

An effective deep learning development process is critical for widespread industrial adoption, particularly in the automotive sector. A typical industrial deep learning development cycle involves customizing and re-designing an off-the-shelf network architecture to meet the operational requirements of the target application, leading to considerable trial and error work by a machine learning practitioner. This approach greatly impedes development with a long turnaround time and the unsatisfactory quality of the created models. As a result, a development platform that can aid engineers in greatly accelerating the design and production of compact, optimized deep neural networks is highly desirable. In this joint industrial case study, we study the efficacy of the GenSynth AI-assisted AI design platform for accelerating the design of custom, optimized deep neural networks for autonomous driving through human-machine collaborative design. We perform a quantitative examination by evaluating 10 different compact deep neural networks produced by GenSynth for the purpose of object detection via a NASNet-based user network prototype design, targeted at a low-cost GPU-based accelerated embedded system. Furthermore, we quantitatively assess the talent hours and GPU processing hours used by the GenSynth process and three other approaches based on the typical industrial development process. In addition, we quantify the annual cloud cost savings for comprehensive testing using networks produced by GenSynth. Finally, we assess the usability and merits of the GenSynth process through user feedback. The findings of this case study showed that GenSynth is easy to use and can be effective at accelerating the design and production of compact, customized deep neural network.

Mohammad Saeed Shafiee, Mohammad Javad Shafiee, Alexander Wong

While deep neural networks extract rich features from the input data, the current trade-off between depth and computational cost makes it difficult to adopt deep neural networks for many industrial applications, especially when computing power is limited. Here, we are inspired by the idea that, while deeper embeddings are needed to discriminate difficult samples (i.e., fine-grained discrimination), a large number of samples can be well discriminated via much shallower embeddings (i.e., coarse-grained discrimination). In this study, we introduce the simple yet effective concept of decision gates (d-gate), modules trained to decide whether a sample needs to be projected into a deeper embedding or if an early prediction can be made at the d-gate, thus enabling the computation of dynamic representations at different depths. The proposed d-gate modules can be integrated with any deep neural network and reduces the average computational cost of the deep neural networks while maintaining modeling accuracy. The proposed d-gate framework is examined via different network architectures and datasets, with experimental results showing that leveraging the proposed d-gate modules led to a ~43% speed-up and 44% FLOPs reduction on ResNet-101 and 55% speed-up and 39% FLOPs reduction on DenseNet-201 trained on the CIFAR10 dataset with only ~2% drop in accuracy. Furthermore, experiments where d-gate modules are integrated into ResNet-101 trained on the ImageNet dataset demonstrate that it is possible to reduce the computational cost of the network by 1.5 GFLOPs without any drop in the modeling accuracy.

Alexander Wong, Mohammad Javad Shafiee, Brendan Chwyl et al.

The tremendous potential exhibited by deep learning is often offset by architectural and computational complexity, making widespread deployment a challenge for edge scenarios such as mobile and other consumer devices. To tackle this challenge, we explore the following idea: Can we learn generative machines to automatically generate deep neural networks with efficient network architectures? In this study, we introduce the idea of generative synthesis, which is premised on the intricate interplay between a generator-inquisitor pair that work in tandem to garner insights and learn to generate highly efficient deep neural networks that best satisfies operational requirements. What is most interesting is that, once a generator has been learned through generative synthesis, it can be used to generate not just one but a large variety of different, unique highly efficient deep neural networks that satisfy operational requirements. Experimental results for image classification, semantic segmentation, and object detection tasks illustrate the efficacy of generative synthesis in producing generators that automatically generate highly efficient deep neural networks (which we nickname FermiNets) with higher model efficiency and lower computational costs (reaching >10x more efficient and fewer multiply-accumulate operations than several tested state-of-the-art networks), as well as higher energy efficiency (reaching >4x improvements in image inferences per joule consumed on a Nvidia Tegra X2 mobile processor). As such, generative synthesis can be a powerful, generalized approach for accelerating and improving the building of deep neural networks for on-device edge scenarios.

Amir Nazemi, Mohammad Javad Shafiee, Zohreh Azimifar et al.

Vehicle Make and Model Recognition (MMR) systems provide a fully automatic framework to recognize and classify different vehicle models. Several approaches have been proposed to address this challenge, however they can perform in restricted conditions. Here, we formulate the vehicle make and model recognition as a fine-grained classification problem and propose a new configurable on-road vehicle make and model recognition framework. We benefit from the unsupervised feature learning methods and in more details we employ Locality constraint Linear Coding (LLC) method as a fast feature encoder for encoding the input SIFT features. The proposed method can perform in real environments of different conditions. This framework can recognize fifty models of vehicles and has an advantage to classify every other vehicle not belonging to one of the specified fifty classes as an unknown vehicle. The proposed MMR framework can be configured to become faster or more accurate based on the application domain. The proposed approach is examined on two datasets including Iranian on-road vehicle dataset and CompuCar dataset. The Iranian on-road vehicle dataset contains images of 50 models of vehicles captured in real situations by traffic cameras in different weather and lighting conditions. Experimental results show superiority of the proposed framework over the state-of-the-art methods on Iranian on-road vehicle datatset and comparable results on CompuCar dataset with 97.5% and 98.4% accuracies, respectively.

Alexander Wong, Mohammad Javad Shafiee, Michael St. Jules

Traffic sign recognition is a very important computer vision task for a number of real-world applications such as intelligent transportation surveillance and analysis. While deep neural networks have been demonstrated in recent years to provide state-of-the-art performance traffic sign recognition, a key challenge for enabling the widespread deployment of deep neural networks for embedded traffic sign recognition is the high computational and memory requirements of such networks. As a consequence, there are significant benefits in investigating compact deep neural network architectures for traffic sign recognition that are better suited for embedded devices. In this paper, we introduce MicronNet, a highly compact deep convolutional neural network for real-time embedded traffic sign recognition designed based on macroarchitecture design principles (e.g., spectral macroarchitecture augmentation, parameter precision optimization, etc.) as well as numerical microarchitecture optimization strategies. The resulting overall architecture of MicronNet is thus designed with as few parameters and computations as possible while maintaining recognition performance, leading to optimized information density of the proposed network. The resulting MicronNet possesses a model size of just ~1MB and ~510,000 parameters (~27x fewer parameters than state-of-the-art) while still achieving a human performance level top-1 accuracy of 98.9% on the German traffic sign recognition benchmark. Furthermore, MicronNet requires just ~10 million multiply-accumulate operations to perform inference, and has a time-to-compute of just 32.19 ms on a Cortex-A53 high efficiency processor. These experimental results show that highly compact, optimized deep neural network architectures can be designed for real-time traffic sign recognition that are well-suited for embedded scenarios.

Alexander Wong, Mohammad Javad Shafiee, Francis Li et al.

Object detection is a major challenge in computer vision, involving both object classification and object localization within a scene. While deep neural networks have been shown in recent years to yield very powerful techniques for tackling the challenge of object detection, one of the biggest challenges with enabling such object detection networks for widespread deployment on embedded devices is high computational and memory requirements. Recently, there has been an increasing focus in exploring small deep neural network architectures for object detection that are more suitable for embedded devices, such as Tiny YOLO and SqueezeDet. Inspired by the efficiency of the Fire microarchitecture introduced in SqueezeNet and the object detection performance of the single-shot detection macroarchitecture introduced in SSD, this paper introduces Tiny SSD, a single-shot detection deep convolutional neural network for real-time embedded object detection that is composed of a highly optimized, non-uniform Fire sub-network stack and a non-uniform sub-network stack of highly optimized SSD-based auxiliary convolutional feature layers designed specifically to minimize model size while maintaining object detection performance. The resulting Tiny SSD possess a model size of 2.3MB (~26X smaller than Tiny YOLO) while still achieving an mAP of 61.3% on VOC 2007 (~4.2% higher than Tiny YOLO). These experimental results show that very small deep neural network architectures can be designed for real-time object detection that are well-suited for embedded scenarios.

Mohammad Javad Shafiee, Brendan Chwyl, Francis Li et al.

The computational complexity of leveraging deep neural networks for extracting deep feature representations is a significant barrier to its widespread adoption, particularly for use in embedded devices. One particularly promising strategy to addressing the complexity issue is the notion of evolutionary synthesis of deep neural networks, which was demonstrated to successfully produce highly efficient deep neural networks while retaining modeling performance. Here, we further extend upon the evolutionary synthesis strategy for achieving efficient feature extraction via the introduction of a stress-induced evolutionary synthesis framework, where stress signals are imposed upon the synapses of a deep neural network during training to induce stress and steer the synthesis process towards the production of more efficient deep neural networks over successive generations and improved model fidelity at a greater efficiency. The proposed stress-induced evolutionary synthesis approach is evaluated on a variety of different deep neural network architectures (LeNet5, AlexNet, and YOLOv2) on different tasks (object classification and object detection) to synthesize efficient StressedNets over multiple generations. Experimental results demonstrate the efficacy of the proposed framework to synthesize StressedNets with significant improvement in network architecture efficiency (e.g., 40x for AlexNet and 33x for YOLOv2) and speed improvements (e.g., 5.5x inference speed-up for YOLOv2 on an Nvidia Tegra X1 mobile processor).

Mohammad Javad Shafiee, Francis Li, Brendan Chwyl et al.

While deep neural networks have been shown in recent years to outperform other machine learning methods in a wide range of applications, one of the biggest challenges with enabling deep neural networks for widespread deployment on edge devices such as mobile and other consumer devices is high computational and memory requirements. Recently, there has been greater exploration into small deep neural network architectures that are more suitable for edge devices, with one of the most popular architectures being SqueezeNet, with an incredibly small model size of 4.8MB. Taking further advantage of the notion that many applications of machine learning on edge devices are often characterized by a low number of target classes, this study explores the utility of combining architectural modifications and an evolutionary synthesis strategy for synthesizing even smaller deep neural architectures based on the more recent SqueezeNet v1.1 macroarchitecture for applications with fewer target classes. In particular, architectural modifications are first made to SqueezeNet v1.1 to accommodate for a 10-class ImageNet-10 dataset, and then an evolutionary synthesis strategy is leveraged to synthesize more efficient deep neural networks based on this modified macroarchitecture. The resulting SquishedNets possess model sizes ranging from 2.4MB to 0.95MB (~5.17X smaller than SqueezeNet v1.1, or 253X smaller than AlexNet). Furthermore, the SquishedNets are still able to achieve accuracies ranging from 81.2% to 77%, and able to process at speeds of 156 images/sec to as much as 256 images/sec on a Nvidia Jetson TX1 embedded chip. These preliminary results show that a combination of architectural modifications and an evolutionary synthesis strategy can be a useful tool for producing very small deep neural network architectures that are well-suited for edge device scenarios.

Mohammad Javad Shafiee, Alexander Wong

While skin cancer is the most diagnosed form of cancer in men and women, with more cases diagnosed each year than all other cancers combined, sufficiently early diagnosis results in very good prognosis and as such makes early detection crucial. While radiomics have shown considerable promise as a powerful diagnostic tool for significantly improving oncological diagnostic accuracy and efficiency, current radiomics-driven methods have largely rely on pre-defined, hand-crafted quantitative features, which can greatly limit the ability to fully characterize unique cancer phenotype that distinguish it from healthy tissue. Recently, the notion of discovery radiomics was introduced, where a large amount of custom, quantitative radiomic features are directly discovered from the wealth of readily available medical imaging data. In this study, we present a novel discovery radiomics framework for skin cancer detection, where we leverage novel deep multi-column radiomic sequencers for high-throughput discovery and extraction of a large amount of custom radiomic features tailored for characterizing unique skin cancer tissue phenotype. The discovered radiomic sequencer was tested against 9,152 biopsy-proven clinical images comprising of different skin cancers such as melanoma and basal cell carcinoma, and demonstrated sensitivity and specificity of 91% and 75%, respectively, thus achieving dermatologist-level performance and \break hence can be a powerful tool for assisting general practitioners and dermatologists alike in improving the efficiency, consistency, and accuracy of skin cancer diagnosis.

Mohammad Javad Shafiee, Brendan Chywl, Francis Li et al.

Object detection is considered one of the most challenging problems in this field of computer vision, as it involves the combination of object classification and object localization within a scene. Recently, deep neural networks (DNNs) have been demonstrated to achieve superior object detection performance compared to other approaches, with YOLOv2 (an improved You Only Look Once model) being one of the state-of-the-art in DNN-based object detection methods in terms of both speed and accuracy. Although YOLOv2 can achieve real-time performance on a powerful GPU, it still remains very challenging for leveraging this approach for real-time object detection in video on embedded computing devices with limited computational power and limited memory. In this paper, we propose a new framework called Fast YOLO, a fast You Only Look Once framework which accelerates YOLOv2 to be able to perform object detection in video on embedded devices in a real-time manner. First, we leverage the evolutionary deep intelligence framework to evolve the YOLOv2 network architecture and produce an optimized architecture (referred to as O-YOLOv2 here) that has 2.8X fewer parameters with just a ~2% IOU drop. To further reduce power consumption on embedded devices while maintaining performance, a motion-adaptive inference method is introduced into the proposed Fast YOLO framework to reduce the frequency of deep inference with O-YOLOv2 based on temporal motion characteristics. Experimental results show that the proposed Fast YOLO framework can reduce the number of deep inferences by an average of 38.13%, and an average speedup of ~3.3X for objection detection in video compared to the original YOLOv2, leading Fast YOLO to run an average of ~18FPS on a Nvidia Jetson TX1 embedded system.

Audrey Chung, Mohammad Javad Shafiee, Paul Fieguth et al.

Evolutionary deep intelligence was recently proposed as a method for achieving highly efficient deep neural network architectures over successive generations. Drawing inspiration from nature, we propose the incorporation of sexual evolutionary synthesis. Rather than the current asexual synthesis of networks, we aim to produce more compact feature representations by synthesizing more diverse and generalizable offspring networks in subsequent generations via the combination of two parent networks. Experimental results were obtained using the MNIST and CIFAR-10 datasets, and showed improved architectural efficiency and comparable testing accuracy relative to the baseline asexual evolutionary neural networks. In particular, the network synthesized via sexual evolutionary synthesis for MNIST had approximately double the architectural efficiency (cluster efficiency of 34.29X and synaptic efficiency of 258.37X) in comparison to the network synthesized via asexual evolutionary synthesis, with both networks achieving a testing accuracy of ~97%.

Mohammad Javad Shafiee, Francis Li, Alexander Wong

A key contributing factor to incredible success of deep neural networks has been the significant rise on massively parallel computing devices allowing researchers to greatly increase the size and depth of deep neural networks, leading to significant improvements in modeling accuracy. Although deeper, larger, or complex deep neural networks have shown considerable promise, the computational complexity of such networks is a major barrier to utilization in resource-starved scenarios. We explore the synaptogenesis of deep neural networks in the formation of efficient deep neural network architectures within an evolutionary deep intelligence framework, where a probabilistic generative modeling strategy is introduced to stochastically synthesize increasingly efficient yet effective offspring deep neural networks over generations, mimicking evolutionary processes such as heredity, random mutation, and natural selection in a probabilistic manner. In this study, we primarily explore the imposition of synaptic precision restrictions and its impact on the evolutionary synthesis of deep neural networks to synthesize more efficient network architectures tailored for resource-starved scenarios. Experimental results show significant improvements in synaptic efficiency (~10X decrease for GoogLeNet-based DetectNet) and inference speed (>5X increase for GoogLeNet-based DetectNet) while preserving modeling accuracy.

Mohammad Javad Shafiee, Audrey G. Chung, Farzad Khalvati et al.

While lung cancer is the second most diagnosed form of cancer in men and women, a sufficiently early diagnosis can be pivotal in patient survival rates. Imaging-based, or radiomics-driven, detection methods have been developed to aid diagnosticians, but largely rely on hand-crafted features which may not fully encapsulate the differences between cancerous and healthy tissue. Recently, the concept of discovery radiomics was introduced, where custom abstract features are discovered from readily available imaging data. We propose a novel evolutionary deep radiomic sequencer discovery approach based on evolutionary deep intelligence. Motivated by patient privacy concerns and the idea of operational artificial intelligence, the evolutionary deep radiomic sequencer discovery approach organically evolves increasingly more efficient deep radiomic sequencers that produce significantly more compact yet similarly descriptive radiomic sequences over multiple generations. As a result, this framework improves operational efficiency and enables diagnosis to be run locally at the radiologist's computer while maintaining detection accuracy. We evaluated the evolved deep radiomic sequencer (EDRS) discovered via the proposed evolutionary deep radiomic sequencer discovery framework against state-of-the-art radiomics-driven and discovery radiomics methods using clinical lung CT data with pathologically-proven diagnostic data from the LIDC-IDRI dataset. The evolved deep radiomic sequencer shows improved sensitivity (93.42%), specificity (82.39%), and diagnostic accuracy (88.78%) relative to previous radiomics approaches.

Mohammad Javad Shafiee, Elnaz Barshan, Alexander Wong

A promising paradigm for achieving highly efficient deep neural networks is the idea of evolutionary deep intelligence, which mimics biological evolution processes to progressively synthesize more efficient networks. A crucial design factor in evolutionary deep intelligence is the genetic encoding scheme used to simulate heredity and determine the architectures of offspring networks. In this study, we take a deeper look at the notion of synaptic cluster-driven evolution of deep neural networks which guides the evolution process towards the formation of a highly sparse set of synaptic clusters in offspring networks. Utilizing a synaptic cluster-driven genetic encoding, the probabilistic encoding of synaptic traits considers not only individual synaptic properties but also inter-synaptic relationships within a deep neural network. This process results in highly sparse offspring networks which are particularly tailored for parallel computational devices such as GPUs and deep neural network accelerator chips. Comprehensive experimental results using four well-known deep neural network architectures (LeNet-5, AlexNet, ResNet-56, and DetectNet) on two different tasks (object categorization and object detection) demonstrate the efficiency of the proposed method. Cluster-driven genetic encoding scheme synthesizes networks that can achieve state-of-the-art performance with significantly smaller number of synapses than that of the original ancestor network. ($\sim$125-fold decrease in synapses for MNIST). Furthermore, the improved cluster efficiency in the generated offspring networks ($\sim$9.71-fold decrease in clusters for MNIST and a $\sim$8.16-fold decrease in clusters for KITTI) is particularly useful for accelerated performance on parallel computing hardware architectures such as those in GPUs and deep neural network accelerator chips.

Mohammad Javad Shafiee, Alexander Wong

There has been significant recent interest towards achieving highly efficient deep neural network architectures. A promising paradigm for achieving this is the concept of evolutionary deep intelligence, which attempts to mimic biological evolution processes to synthesize highly-efficient deep neural networks over successive generations. An important aspect of evolutionary deep intelligence is the genetic encoding scheme used to mimic heredity, which can have a significant impact on the quality of offspring deep neural networks. Motivated by the neurobiological phenomenon of synaptic clustering, we introduce a new genetic encoding scheme where synaptic probability is driven towards the formation of a highly sparse set of synaptic clusters. Experimental results for the task of image classification demonstrated that the synthesized offspring networks using this synaptic cluster-driven genetic encoding scheme can achieve state-of-the-art performance while having network architectures that are not only significantly more efficient (with a ~125-fold decrease in synapses for MNIST) compared to the original ancestor network, but also tailored for GPU-accelerated machine learning applications.

Mohammad Javad Shafiee, Akshaya Mishra, Alexander Wong

Taking inspiration from biological evolution, we explore the idea of "Can deep neural networks evolve naturally over successive generations into highly efficient deep neural networks?" by introducing the notion of synthesizing new highly efficient, yet powerful deep neural networks over successive generations via an evolutionary process from ancestor deep neural networks. The architectural traits of ancestor deep neural networks are encoded using synaptic probability models, which can be viewed as the `DNA' of these networks. New descendant networks with differing network architectures are synthesized based on these synaptic probability models from the ancestor networks and computational environmental factor models, in a random manner to mimic heredity, natural selection, and random mutation. These offspring networks are then trained into fully functional networks, like one would train a newborn, and have more efficient, more diverse network architectures than their ancestor networks, while achieving powerful modeling capabilities. Experimental results for the task of visual saliency demonstrated that the synthesized `evolved' offspring networks can achieve state-of-the-art performance while having network architectures that are significantly more efficient (with a staggering $\sim$48-fold decrease in synapses by the fourth generation) compared to the original ancestor network.

Parthipan Siva, Mohammad Javad Shafiee, Mike Jamieson et al.

The problem of automated crowd segmentation and counting has garnered significant interest in the field of video surveillance. This paper proposes a novel scene invariant crowd segmentation and counting algorithm designed with high accuracy yet low computational complexity in mind, which is key for widespread industrial adoption. A novel low-complexity, scale-normalized feature called Histogram of Moving Gradients (HoMG) is introduced for highly effective spatiotemporal representation of individuals and crowds within a video. Real-time crowd segmentation is achieved via boosted cascade of weak classifiers based on sliding-window HoMG features, while linear SVM regression of crowd-region HoMG features is employed for real-time crowd counting. Experimental results using multi-camera crowd datasets show that the proposed algorithm significantly outperform state-of-the-art crowd counting algorithms, as well as achieve very promising crowd segmentation results, thus demonstrating the efficacy of the proposed method for highly-accurate, real-time video-driven crowd analysis.

Edward Li, Farzad Khalvati, Mohammad Javad Shafiee et al.

Magnetic Resonance Imaging (MRI) is a crucial medical imaging technology for the screening and diagnosis of frequently occurring cancers. However image quality may suffer by long acquisition times for MRIs due to patient motion, as well as result in great patient discomfort. Reducing MRI acquisition time can reduce patient discomfort and as a result reduces motion artifacts from the acquisition process. Compressive sensing strategies, when applied to MRI, have been demonstrated to be effective at decreasing acquisition times significantly by sparsely sampling the \emph{k}-space during the acquisition process. However, such a strategy requires advanced reconstruction algorithms to produce high quality and reliable images from compressive sensing MRI. This paper proposes a new reconstruction approach based on cross-domain stochastically fully connected conditional random fields (CD-SFCRF) for compressive sensing MRI. The CD-SFCRF introduces constraints in both \emph{k}-space and spatial domains within a stochastically fully connected graphical model to produce improved MRI reconstruction. Experimental results using T2-weighted (T2w) imaging and diffusion-weighted imaging (DWI) of the prostate show strong performance in preserving fine details and tissue structures in the reconstructed images when compared to other tested methods even at low sampling rates.

Mohammad Javad Shafiee, Parthipan Siva, Paul Fieguth et al.

Transfer learning is a recent field of machine learning research that aims to resolve the challenge of dealing with insufficient training data in the domain of interest. This is a particular issue with traditional deep neural networks where a large amount of training data is needed. Recently, StochasticNets was proposed to take advantage of sparse connectivity in order to decrease the number of parameters that needs to be learned, which in turn may relax training data size requirements. In this paper, we study the efficacy of transfer learning on StochasticNet frameworks. Experimental results show ~7% improvement on StochasticNet performance when the transfer learning is applied in training step.

Ameneh Boroomand, Mohammad Javad Shafiee, Farzad Khalvati et al.

Diffusion weighted magnetic resonance imaging (DW-MR) is a powerful tool in imaging-based prostate cancer screening and detection. Endorectal coils are commonly used in DW-MR imaging to improve the signal-to-noise ratio (SNR) of the acquisition, at the expense of significant intensity inhomogeneities (bias field) that worsens as we move away from the endorectal coil. The presence of bias field can have a significant negative impact on the accuracy of different image analysis tasks, as well as prostate tumor localization, thus leading to increased inter- and intra-observer variability. Retrospective bias correction approaches are introduced as a more efficient way of bias correction compared to the prospective methods such that they correct for both of the scanner and anatomy-related bias fields in MR imaging. Previously proposed retrospective bias field correction methods suffer from undesired noise amplification that can reduce the quality of bias-corrected DW-MR image. Here, we propose a unified data reconstruction approach that enables joint compensation of bias field as well as data noise in DW-MR imaging. The proposed noise-compensated, bias-corrected (NCBC) data reconstruction method takes advantage of a novel stochastically fully connected joint conditional random field (SFC-JCRF) model to mitigate the effects of data noise and bias field in the reconstructed MR data. The proposed NCBC reconstruction method was tested on synthetic DW-MR data, physical DW-phantom as well as real DW-MR data all acquired using endorectal MR coil. Both qualitative and quantitative analysis illustrated that the proposed NCBC method can achieve improved image quality when compared to other tested bias correction methods. As such, the proposed NCBC method may have potential as a useful retrospective approach for improving the consistency of image interpretations.

Mohammad Javad Shafiee, Parthipan Siva, Paul Fieguth et al.

Deep neural networks are a powerful tool for feature learning and extraction given their ability to model high-level abstractions in highly complex data. One area worth exploring in feature learning and extraction using deep neural networks is efficient neural connectivity formation for faster feature learning and extraction. Motivated by findings of stochastic synaptic connectivity formation in the brain as well as the brain's uncanny ability to efficiently represent information, we propose the efficient learning and extraction of features via StochasticNets, where sparsely-connected deep neural networks can be formed via stochastic connectivity between neurons. To evaluate the feasibility of such a deep neural network architecture for feature learning and extraction, we train deep convolutional StochasticNets to learn abstract features using the CIFAR-10 dataset, and extract the learned features from images to perform classification on the SVHN and STL-10 datasets. Experimental results show that features learned using deep convolutional StochasticNets, with fewer neural connections than conventional deep convolutional neural networks, can allow for better or comparable classification accuracy than conventional deep neural networks: relative test error decrease of ~4.5% for classification on the STL-10 dataset and ~1% for classification on the SVHN dataset. Furthermore, it was shown that the deep features extracted using deep convolutional StochasticNets can provide comparable classification accuracy even when only 10% of the training data is used for feature learning. Finally, it was also shown that significant gains in feature extraction speed can be achieved in embedded applications using StochasticNets. As such, StochasticNets allow for faster feature learning and extraction performance while facilitate for better or comparable accuracy performances.

Mohammad Javad Shafiee, Audrey G. Chung, Devinder Kumar et al.

Radiomics has proven to be a powerful prognostic tool for cancer detection, and has previously been applied in lung, breast, prostate, and head-and-neck cancer studies with great success. However, these radiomics-driven methods rely on pre-defined, hand-crafted radiomic feature sets that can limit their ability to characterize unique cancer traits. In this study, we introduce a novel discovery radiomics framework where we directly discover custom radiomic features from the wealth of available medical imaging data. In particular, we leverage novel StochasticNet radiomic sequencers for extracting custom radiomic features tailored for characterizing unique cancer tissue phenotype. Using StochasticNet radiomic sequencers discovered using a wealth of lung CT data, we perform binary classification on 42,340 lung lesions obtained from the CT scans of 93 patients in the LIDC-IDRI dataset. Preliminary results show significant improvement over previous state-of-the-art methods, indicating the potential of the proposed discovery radiomics framework for improving cancer screening and diagnosis.