To address potential gaps noted in patient monitoring in the hospital, a novel patient behavior detection system using mmWave radar and deep convolution neural network (CNN), which supports the simultaneous recognition of multiple patients' behaviors in real-time, is proposed. In this study, we use an mmWave radar to track multiple patients and detect the scattering point cloud of each one. For each patient, the Doppler pattern of the point cloud over a time period is collected as the behavior signature. A three-layer CNN model is created to classify the behavior for each patient. The tracking and point clouds detection algorithm was also implemented on an mmWave radar hardware platform with an embedded graphics processing unit (GPU) board to collect Doppler pattern and run the CNN model. A training dataset of six types of behavior were collected, over a long duration, to train the model using Adam optimizer with an objective to minimize cross-entropy loss function. Lastly, the system was tested for real-time operation and obtained a very good inference accuracy when predicting each patient's behavior in a two-patient scenario.
Even though the Convolutional Neural Networks (CNN) has shown superior results in the field of computer vision, it is still a challenging task to implement computer vision algorithms in real-time at the edge, especially using a low-cost IoT device due to high memory consumption and computation complexities in a CNN. Network compression methodologies such as weight pruning, filter pruning, and quantization are used to overcome the above mentioned problem. Even though filter pruning methodology has shown better performances compared to other techniques, irregularity of the number of filters pruned across different layers of a CNN might not comply with majority of the neural computing hardware architectures. In this paper, a novel greedy approach called cluster pruning has been proposed, which provides a structured way of removing filters in a CNN by considering the importance of filters and the underlying hardware architecture. The proposed methodology is compared with the conventional filter pruning algorithm on Pascal-VOC open dataset, and Head-Counting dataset, which is our own dataset developed to detect and count people entering a room. We benchmark our proposed method on three hardware architectures, namely CPU, GPU, and Intel Movidius Neural Computer Stick (NCS) using the popular SSD-MobileNet and SSD-SqueezeNet neural network architectures used for edge-AI vision applications. Results demonstrate that our method outperforms the conventional filter pruning methodology, using both datasets on above mentioned hardware architectures. Furthermore, a low cost IoT hardware setup consisting of an Intel Movidius-NCS is proposed to deploy an edge-AI application using our proposed pruning methodology.
This paper presents a fast and parsimonious parsing method to accurately and robustly detect a vectorized wireframe in an input image with a single forward pass. The proposed method is end-to-end trainable, consisting of three components: (i) line segment and junction proposal generation, (ii) line segment and junction matching, and (iii) line segment and junction verification. For computing line segment proposals, a novel exact dual representation is proposed which exploits a parsimonious geometric reparameterization for line segments and forms a holistic 4-dimensional attraction field map for an input image. Junctions can be treated as the "basins" in the attraction field. The proposed method is thus called Holistically-Attracted Wireframe Parser (HAWP). In experiments, the proposed method is tested on two benchmarks, the Wireframe dataset, and the YorkUrban dataset. On both benchmarks, it obtains state-of-the-art performance in terms of accuracy and efficiency. For example, on the Wireframe dataset, compared to the previous state-of-the-art method L-CNN, it improves the challenging mean structural average precision (msAP) by a large margin ($2.8\%$ absolute improvements) and achieves 29.5 FPS on single GPU ($89\%$ relative improvement). A systematic ablation study is performed to further justify the proposed method.
Deep neural networks yield promising results in a wide range of computer vision applications, including landmark detection. A major challenge for accurate anatomical landmark detection in volumetric images such as clinical CT scans is that large-scale data often constrain the capacity of the employed neural network architecture due to GPU memory limitations, which in turn can limit the precision of the output. We propose a multi-scale, end-to-end deep learning method that achieves fast and memory-efficient landmark detection in 3D images. Our architecture consists of blocks of shift-equivariant networks, each of which performs landmark detection at a different spatial scale. These blocks are connected from coarse to fine-scale, with differentiable resampling layers, so that all levels can be trained together. We also present a noise injection strategy that increases the robustness of the model and allows us to quantify uncertainty at test time. We evaluate our method for carotid artery bifurcations detection on 263 CT volumes and achieve a better than state-of-the-art accuracy with mean Euclidean distance error of 2.81mm.
The understanding of the surrounding environment plays a critical role in autonomous robotic systems, such as self-driving cars. Extensive research has been carried out concerning visual perception. Yet, to obtain a more complete perception of the environment, autonomous systems of the future should also take acoustic information into account. Recent sound event localization and detection (SELD) frameworks utilize convolutional recurrent neural networks (CRNNs). However, considering the recurrent nature of CRNNs, it becomes challenging to implement them efficiently on embedded hardware. Not only are their computations strenuous to parallelize, but they also require high memory bandwidth and large memory buffers. In this work, we develop a more robust and hardware-friendly novel architecture based on a temporal convolutional network(TCN). The proposed framework (SELD-TCN) outperforms the state-of-the-art SELDnet performance on four different datasets. Moreover, SELD-TCN achieves 4x faster training time per epoch and 40x faster inference time on an ordinary graphics processing unit (GPU).
Background and Objective: Object detection is a primary research interest in computer vision. Sperm-cell detection in a densely populated bull semen microscopic observation video presents challenges such as partial occlusion, vast number of objects in a single video frame, tiny size of the object, artifacts, low contrast, and blurry objects because of the rapid movement of the sperm cells. This study proposes an architecture, called DeepSperm, that solves the aforementioned challenges and is more accurate and faster than state-of-the-art architectures. Methods: In the proposed architecture, we use only one detection layer, which is specific for small object detection. For handling overfitting and increasing accuracy, we set a higher network resolution, use a dropout layer, and perform data augmentation on hue, saturation, and exposure. Several hyper-parameters are tuned to achieve better performance. We compare our proposed method with those of a conventional image processing-based object-detection method, you only look once (YOLOv3), and mask region-based convolutional neural network (Mask R-CNN). Results: In our experiment, we achieve 86.91 mAP on the test dataset and a processing speed of 50.3 fps. In comparison with YOLOv3, we achieve an increase of 16.66 mAP point, 3.26 x faster on testing, and 1.4 x faster on training with a small training dataset, which contains 40 video frames. The weights file size was also reduced significantly, with 16.94 x smaller than that of YOLOv3. Moreover, it requires 1.3 x less graphical processing unit (GPU) memory than YOLOv3. Conclusions: This study proposes DeepSperm, which is a simple, effective, and efficient architecture with its hyper-parameters and configuration to detect bull sperm cells robustly in real time. In our experiment, we surpass the state of the art in terms of accuracy, speed, and resource needs.
How to discover and evaluate the true strength of models quickly and accurately is one of the key challenges in Neural Architecture Search (NAS). To cope with this problem, we propose an Architecture-Driven Weight Prediction (ADWP) approach for neural architecture search (NAS). In our approach, we first design an architecture-intensive search space and then train a HyperNetwork by inputting stochastic encoding architecture parameters. In the trained HyperNetwork, weights of convolution kernels can be well predicted for neural architectures in the search space. Consequently, the target architectures can be evaluated efficiently without any finetuning, thus enabling us to search fortheoptimalarchitectureinthespaceofgeneralnetworks (macro-search). Through real experiments, we evaluate the performance of the models discovered by the proposed AD-WPNAS and results show that one search procedure can be completed in 4.0 GPU hours on CIFAR-10. Moreover, the discovered model obtains a test error of 2.41% with only 1.52M parameters which is superior to the best existing models.
We present PrecisionBatching, a quantized inference algorithm for speeding up neural network execution on traditional hardware platforms at low bitwidths without the need for retraining or recalibration. PrecisionBatching decomposes a neural network into individual bitlayers and accumulates them using fast 1-bit operations while maintaining activations in full precision. PrecisionBatching not only facilitates quantized inference at low bitwidths (< 8 bits) without the need for retraining/recalibration, but also 1) enables traditional hardware platforms the ability to realize inference speedups at a finer granularity of quantization (e.g: 1-16 bit execution) and 2) allows accuracy and speedup tradeoffs at runtime by exposing the number of bitlayers to accumulate as a tunable parameter. Across a variety of applications (MNIST, language modeling, natural language inference) and neural network architectures (fully connected, RNN, LSTM), PrecisionBatching yields end-to-end speedups of over 8x on a GPU within a < 1% error margin of the full precision baseline, outperforming traditional 8-bit quantized inference by over 1.5x-2x at the same error tolerance.
An estimated 60% of smartphones sold in 2018 were equipped with multiple rear cameras, enabling a wide variety of 3D-enabled applications such as 3D Photos. The success of 3D Photo platforms (Facebook 3D Photo, Holopix, etc) depend on a steady influx of user generated content. These platforms must provide simple image manipulation tools to facilitate content creation, akin to traditional photo platforms. Artistic neural style transfer, propelled by recent advancements in GPU technology, is one such tool for enhancing traditional photos. However, naively extrapolating single-view neural style transfer to the multi-view scenario produces visually inconsistent results and is prohibitively slow on mobile devices. We present a GPU-accelerated multi-view style transfer pipeline which enforces style consistency between views with on-demand performance on mobile platforms. Our pipeline is modular and creates high quality depth and parallax effects from a stereoscopic image pair.
In this paper, we build autoencoder based pipelines for extreme end-to-end image compression based on Ball\'e's approach, which is the state-of-the-art open source implementation in image compression using deep learning. We deepened the network by adding one more hidden layer before each strided convolutional layer with exactly the same number of down-samplings and up-samplings. Our approach outperformed Ball\'e's approach, and achieved around 4.0% reduction in bits per pixel (bpp), 0.03% increase in multi-scale structural similarity (MS-SSIM), and only 0.47% decrease in peak signal-to-noise ratio (PSNR), It also outperforms all traditional image compression methods including JPEG2000 and HEIC by at least 20% in terms of compression efficiency at similar reconstruction image quality. Regarding encoding and decoding time, our approach takes similar amount of time compared with traditional methods with the support of GPU, which means it's almost ready for industrial applications.
Recurrent neural networks (RNNs) based automatic speech recognition has nowadays become prevalent on mobile devices such as smart phones. However, previous RNN compression techniques either suffer from hardware performance overhead due to irregularity or significant accuracy loss due to the preserved regularity for hardware friendliness. In this work, we propose RTMobile that leverages both a novel block-based pruning approach and compiler optimizations to accelerate RNN inference on mobile devices. Our proposed RTMobile is the first work that can achieve real-time RNN inference on mobile platforms. Experimental results demonstrate that RTMobile can significantly outperform existing RNN hardware acceleration methods in terms of inference accuracy and time. Compared with prior work on FPGA, RTMobile using Adreno 640 embedded GPU on GRU can improve the energy-efficiency by about 40$\times$ while maintaining the same inference time.
In cloud and edge computing models, it is important that compute devices at the edge be as power efficient as possible. Long short-term memory (LSTM) neural networks have been widely used for natural language processing, time series prediction and many other sequential data tasks. Thus, for these applications there is increasing need for low-power accelerators for LSTM model inference at the edge. In order to reduce power dissipation due to data transfers within inference devices, there has been significant interest in accelerating vector-matrix multiplication (VMM) operations using non-volatile memory (NVM) weight arrays. In NVM array-based hardware, reduced bit-widths also significantly increases the power efficiency. In this paper, we focus on the application of quantization-aware training algorithm to LSTM models, and the benefits these models bring in terms of resilience against both quantization error and analog device noise. We have shown that only 4-bit NVM weights and 4-bit ADC/DACs are needed to produce equivalent LSTM network performance as floating-point baseline. Reasonable levels of ADC quantization noise and weight noise can be naturally tolerated within our NVMbased quantized LSTM network. Benchmark analysis of our proposed LSTM accelerator for inference has shown at least 2.4x better computing efficiency and 40x higher area efficiency than traditional digital approaches (GPU, FPGA, and ASIC). Some other novel approaches based on NVM promise to deliver higher computing efficiency (up to 4.7x) but require larger arrays with potential higher error rates.
Energy efficiency of hardware accelerators of deep neural networks (DNN) can be improved by introducing approximate arithmetic circuits. In order to quantify the error introduced by using these circuits and avoid the expensive hardware prototyping, a software emulator of the DNN accelerator is usually executed on CPU or GPU. However, this emulation is typically two or three orders of magnitude slower than a software DNN implementation running on CPU or GPU and operating with standard floating point arithmetic instructions and common DNN libraries. The reason is that there is no hardware support for approximate arithmetic operations on common CPUs and GPUs and these operations have to be expensively emulated. In order to address this issue, we propose an efficient emulation method for approximate circuits utilized in a given DNN accelerator which is emulated on GPU. All relevant approximate circuits are implemented as look-up tables and accessed through a texture memory mechanism of CUDA capable GPUs. We exploit the fact that the texture memory is optimized for irregular read-only access and in some GPU architectures is even implemented as a dedicated cache. This technique allowed us to reduce the inference time of the emulated DNN accelerator approximately 200 times with respect to an optimized CPU version on complex DNNs such as ResNet. The proposed approach extends the TensorFlow library and is available online at https://github.com/ehw-fit/tf-approximate.
Distributed Deep Learning (DDL) has rapidly grown its popularity since it helps boost the training performance on high-performance GPU clusters. Efficient job scheduling is indispensable to maximize the overall performance of the cluster when training multiple jobs simultaneously. However, existing schedulers do not consider the communication contention of multiple communication tasks from different distributed training jobs, which could deteriorate the system performance and prolong the job completion time. In this paper, we first establish a new DDL job scheduling framework which organizes DDL jobs as Directed Acyclic Graphs (DAGs) and considers communication contention between nodes. We then propose an efficient algorithm, LWF-$\kappa$, to balance the GPU utilization and consolidate the allocated GPUs for each job. When scheduling those communication tasks, we observe that neither avoiding all the contention nor blindly accepting them is optimal to minimize the job completion time. We thus propose a provable algorithm, AdaDUAL, to efficiently schedule those communication tasks. Based on AdaDUAL, we finally propose Ada-SRSF for the DDL job scheduling problem. Simulations on a 64-GPU cluster connected with 10 Gbps Ethernet show that LWF-$\kappa$ achieves up to $1.59\times$ improvement over the classical first-fit algorithms. More importantly, Ada-SRSF reduces the average job completion time by $20.1\%$ and $36.7\%$, as compared to the SRSF(1) scheme (avoiding all the contention) and the SRSF(2) scheme (blindly accepting all of two-way communication contention) respectively.
A fundamental component of neural network verification is the computation of bounds on the values their outputs can take. Previous methods have either used off-the-shelf solvers, discarding the problem structure, or relaxed the problem even further, making the bounds unnecessarily loose. We propose a novel approach based on Lagrangian Decomposition. Our formulation admits an efficient supergradient ascent algorithm, as well as an improved proximal algorithm. Both the algorithms offer three advantages: (i) they yield bounds that are provably at least as tight as previous dual algorithms relying on Lagrangian relaxations; (ii) they are based on operations analogous to forward/backward pass of neural networks layers and are therefore easily parallelizable, amenable to GPU implementation and able to take advantage of the convolutional structure of problems; and (iii) they allow for anytime stopping while still providing valid bounds. Empirically, we show that we obtain bounds comparable with off-the-shelf solvers in a fraction of their running time, and obtain tighter bounds in the same time as previous dual algorithms. This results in an overall speed-up when employing the bounds for formal verification. Code for our algorithms is available at https://github.com/oval-group/decomposition-plnn-bounds.
Neural architecture search (NAS) relies on a good controller to generate better architectures or predict the accuracy of given architectures. However, training the controller requires both abundant and high-quality pairs of architectures and their accuracy, while it is costly to evaluate an architecture and obtain its accuracy. In this paper, we propose SemiNAS, a semi-supervised NAS approach that leverages numerous unlabeled architectures (without evaluation and thus nearly no cost). Specifically, SemiNAS 1) trains an initial accuracy predictor with a small set of architecture-accuracy data pairs; 2) uses the trained accuracy predictor to predict the accuracy of large amount of architectures (without evaluation); and 3) adds the generated data pairs to the original data to further improve the predictor. The trained accuracy predictor can be applied to various NAS algorithms by predicting the accuracy of candidate architectures for them. SemiNAS has two advantages: 1) It reduces the computational cost under the same accuracy guarantee. On NASBench-101 benchmark dataset, it achieves comparable accuracy with gradient-based method while using only 1/7 architecture-accuracy pairs. 2) It achieves higher accuracy under the same computational cost. It achieves 94.02% test accuracy on NASBench-101, outperforming all the baselines when using the same number of architectures. On ImageNet, it achieves 23.5% top-1 error rate (under 600M FLOPS constraint) using 4 GPU-days for search. We further apply it to LJSpeech text to speech task and it achieves 97% intelligibility rate in the low-resource setting and 15% test error rate in the robustness setting, with 9%, 7% improvements over the baseline respectively.
If NAS methods are solutions, what is the problem? Most existing NAS methods require two-stage parameter optimization. However, performance of the same architecture in the two stages correlates poorly. In this work, we propose a new problem definition for NAS, task-specific end-to-end, based on this observation. We argue that given a computer vision task for which a NAS method is expected, this definition can reduce the vaguely-defined NAS evaluation to i) accuracy of this task and ii) the total computation consumed to finally obtain a model with satisfying accuracy. Seeing that most existing methods do not solve this problem directly, we propose DSNAS, an efficient differentiable NAS framework that simultaneously optimizes architecture and parameters with a low-biased Monte Carlo estimate. Child networks derived from DSNAS can be deployed directly without parameter retraining. Comparing with two-stage methods, DSNAS successfully discovers networks with comparable accuracy (74.4%) on ImageNet in 420 GPU hours, reducing the total time by more than 34%. Our implementation is available at https://github.com/SNAS-Series/SNAS-Series.
In this work we present a new framework for neural networks compression with fine-tuning, which we called Neural Network Compression Framework (NNCF). It leverages recent advances of various network compression methods and implements some of them, such as sparsity, quantization, and binarization. These methods allow getting more hardware-friendly models which can be efficiently run on general-purpose hardware computation units (CPU, GPU) or special Deep Learning accelerators. We show that the developed methods can be successfully applied to a wide range of models to accelerate the inference time while keeping the original accuracy. The framework can be used within the training samples, which are supplied with it, or as a standalone package that can be seamlessly integrated into the existing training code with minimal adaptations. Currently, a PyTorch version of NNCF is available as a part of OpenVINO Training Extensions at https://github.com/openvinotoolkit/nncf.
The research interest in specialized hardware accelerators for deep neural networks (DNN) spikes recently owing to their superior performance and efficiency. However, today's DNN accelerators primarily focus on accelerating specific "kernels" such as convolution and matrix multiplication, which are vital but only part of an end-to-end DNN-enabled application. Meaningful speedups over the entire application often require supporting computations that are, while massively parallel, ill-suited to DNN accelerators. Integrating a general-purpose processor such as a CPU or a GPU incurs significant data movement overhead and leads to resource under-utilization on the DNN accelerators. We propose Simultaneous Multi-mode Architecture (SMA), a novel architecture design and execution model that offers general-purpose programmability on DNN accelerators in order to accelerate end-to-end applications. The key to SMA is the temporal integration of the systolic execution model with the GPU-like SIMD execution model. The SMA exploits the common components shared between the systolic-array accelerator and the GPU, and provides lightweight reconfiguration capability to switch between the two modes in-situ. The SMA achieves up to 63% performance improvement while consuming 23% less energy than the baseline Volta architecture with TensorCore.
Creating a database of 21cm brightness temperature signals from the Epoch of Reionisation (EoR) for an array of reionisation histories is a complex and computationally expensive task, given the range of astrophysical processes involved and the possibly high-dimensional parameter space that is to be probed. We utilise a specific type of neural network, a Progressively Growing Generative Adversarial Network (PGGAN), to produce realistic tomography images of the 21cm brightness temperature during the EoR, covering a continuous three-dimensional parameter space that models varying X-ray emissivity, Lyman band emissivity, and ratio between hard and soft X-rays. The GPU-trained network generates new samples at a resolution of $\sim 3'$ in a second (on a laptop CPU), and the resulting global 21cm signal, power spectrum, and pixel distribution function agree well with those of the training data, taken from the 21SSD catalogue \citep{Semelin2017}. Finally, we showcase how a trained PGGAN can be leveraged for the converse task of inferring parameters from 21cm tomography samples via Approximate Bayesian Computation.