Neural Architecture Search (NAS) refers to automatically design the architecture. We propose an hourglass-inspired approach (HourNAS) for this problem that is motivated by the fact that the effects of the architecture often proceed from the vital few blocks. Acting like the narrow neck of an hourglass, vital blocks in the guaranteed path from the input to the output of a deep neural network restrict the information flow and influence the network accuracy. The other blocks occupy the major volume of the network and determine the overall network complexity, corresponding to the bulbs of an hourglass. To achieve an extremely fast NAS while preserving the high accuracy, we propose to identify the vital blocks and make them the priority in the architecture search. The search space of those non-vital blocks is further shrunk to only cover the candidates that are affordable under the computational resource constraints. Experimental results on the ImageNet show that only using 3 hours (0.1 days) with one GPU, our HourNAS can search an architecture that achieves a 77.0% Top-1 accuracy, which outperforms the state-of-the-art methods.
Towards 3D object tracking in point clouds, a novel point-to-box network termed P2B is proposed in an end-to-end learning manner. Our main idea is to first localize potential target centers in 3D search area embedded with target information. Then point-driven 3D target proposal and verification are executed jointly. In this way, the time-consuming 3D exhaustive search can be avoided. Specifically, we first sample seeds from the point clouds in template and search area respectively. Then, we execute permutation-invariant feature augmentation to embed target clues from template into search area seeds and represent them with target-specific features. Consequently, the augmented search area seeds regress the potential target centers via Hough voting. The centers are further strengthened with seed-wise targetness scores. Finally, each center clusters its neighbors to leverage the ensemble power for joint 3D target proposal and verification. We apply PointNet++ as our backbone and experiments on KITTI tracking dataset demonstrate P2B's superiority (~10%'s improvement over state-of-the-art). Note that P2B can run with 40FPS on a single NVIDIA 1080Ti GPU. Our code and model are available at https://github.com/HaozheQi/P2B.
GPUs are currently the platform of choice for training neural networks. However, training a deep neural network (DNN) is a time-consuming process even on GPUs because of the massive number of parameters that have to be learned. As a result, accelerating DNN training has been an area of significant research in the last couple of years. While earlier networks such as AlexNet had a linear dependency between layers and operations, state-of-the-art networks such as ResNet, PathNet, and GoogleNet have a non-linear structure that exhibits a higher level of inter-operation parallelism. However, popular deep learning (DL) frameworks such as TensorFlow and PyTorch launch the majority of neural network operations, especially convolutions, serially on GPUs and do not exploit this inter-op parallelism. In this brief announcement, we make a case for the need and potential benefit of exploiting this rich parallelism in state-of-the-art non-linear networks for reducing the training time. We identify the challenges and limitations in enabling concurrent layer execution on GPU backends (such as cuDNN) of DL frameworks and propose potential solutions.
Deep Neural Networks (DNNs) are well-known to be vulnerable to Adversarial Examples (AEs). A large amount of efforts have been spent to launch and heat the arms race between the attackers and defenders. Recently, advanced gradient-based attack techniques were proposed (e.g., BPDA and EOT), which have defeated a considerable number of existing defense methods. Up to today, there are still no satisfactory solutions that can effectively and efficiently defend against those attacks. In this paper, we make a steady step towards mitigating those advanced gradient-based attacks with two major contributions. First, we perform an in-depth analysis about the root causes of those attacks, and propose four properties that can break the fundamental assumptions of those attacks. Second, we identify a set of operations that can meet those properties. By integrating these operations, we design two preprocessing functions that can invalidate these powerful attacks. Extensive evaluations indicate that our solutions can effectively mitigate all existing standard and advanced attack techniques, and beat 11 state-of-the-art defense solutions published in top-tier conferences over the past 2 years. The defender can employ our solutions to constrain the attack success rate below 7% for the strongest attacks even the adversary has spent dozens of GPU hours.
Transformers are ubiquitous in Natural Language Processing (NLP) tasks, but they are difficult to be deployed on hardware due to the intensive computation. To enable low-latency inference on resource-constrained hardware platforms, we propose to design Hardware-Aware Transformers (HAT) with neural architecture search. We first construct a large design space with $\textit{arbitrary encoder-decoder attention}$ and $\textit{heterogeneous layers}$. Then we train a $\textit{SuperTransformer}$ that covers all candidates in the design space, and efficiently produces many $\textit{SubTransformers}$ with weight sharing. Finally, we perform an evolutionary search with a hardware latency constraint to find a specialized $\textit{SubTransformer}$ dedicated to run fast on the target hardware. Extensive experiments on four machine translation tasks demonstrate that HAT can discover efficient models for different hardware (CPU, GPU, IoT device). When running WMT'14 translation task on Raspberry Pi-4, HAT can achieve $\textbf{3}\times$ speedup, $\textbf{3.7}\times$ smaller size over baseline Transformer; $\textbf{2.7}\times$ speedup, $\textbf{3.6}\times$ smaller size over Evolved Transformer with $\textbf{12,041}\times$ less search cost and no performance loss. HAT code is https://github.com/mit-han-lab/hardware-aware-transformers.git
The renaissance of neural architecture search (NAS) has seen classical methods such as genetic algorithms (GA) and genetic programming (GP) being exploited for convolutional neural network (CNN) architectures. While recent work have achieved promising performance on visual perception tasks, the direct encoding scheme of both GA and GP has functional complexity deficiency and does not scale well on large architectures like CNN. To address this, we present a new generative encoding scheme -- $symbolic\ linear\ generative\ encoding$ (SLGE) -- simple, yet powerful scheme which embeds local graph transformations in chromosomes of linear fixed-length string to develop CNN architectures of variant shapes and sizes via evolutionary process of gene expression programming. In experiments, the effectiveness of SLGE is shown in discovering architectures that improve the performance of the state-of-the-art handcrafted CNN architectures on CIFAR-10 and CIFAR-100 image classification tasks; and achieves a competitive classification error rate with the existing NAS methods using less GPU resources.
In recent years, statistical parametric speech synthesis (SPSS) systems have been widely utilized in many interactive speech-based systems (e.g.~Amazon's Alexa, Bose's headphones). To select a suitable SPSS system, both speech quality and performance efficiency (e.g.~decoding time) must be taken into account. In the paper, we compared four popular Vietnamese SPSS techniques using: 1) hidden Markov models (HMM), 2) deep neural networks (DNN), 3) generative adversarial networks (GAN), and 4) end-to-end (E2E) architectures, which consists of Tacontron~2 and WaveGlow vocoder in terms of speech quality and performance efficiency. We showed that the E2E systems accomplished the best quality, but required the power of GPU to achieve real-time performance. We also showed that the HMM-based system had inferior speech quality, but it was the most efficient system. Surprisingly, the E2E systems were more efficient than the DNN and GAN in inference on GPU. Surprisingly, the GAN-based system did not outperform the DNN in term of quality.
Nowadays, large and complex deep learning (DL) models are increasingly trained in a distributed manner across multiple worker machines, in which extensive communications between workers pose serious scaling problems. In this article, we present a quantitative survey of communication optimization techniques for data parallel distributed DL. We first identify the major communication challenges and classify the existing solutions into three levels, namely the learning algorithm, the system architecture, and the network infrastructure. We present the state-of-the-art communication optimization techniques and conduct a comparative study of seven common lossless distributed DL methods on a 32-GPU cluster with 100Gbps InfiniBand (IB). We show that (1) the DL models with low model intensity (such as BERT and BERT-Large) are difficult to scale out even with the best available lossless algorithm over 100Gbps IB; (2) the system architecture and scheduling algorithms have a critical impact on the scaling property. We conclude the article with discussions on the open issues for further investigations.
Knee points, characterised as their smallest trade-off loss at all objectives, are attractive to decision makers in multi-criterion decision-making. In contrast, other Pareto-optimal solutions are less attractive since a small improvement on one objective can lead to a significant degradation on at least one of the other objectives. In this paper, we propose a simple and effective knee point identification method based on trade-off utility, dubbed KPITU, to help decision makers identify knee points from a given set of trade-off solutions. The basic idea of KPITU is to sequentially validate whether a solution is a knee point or not by comparing its trade-off utility with others within its neighbourhood. In particular, a solution is a knee point if and only if it has the best trade-off utility among its neighbours. Moreover, we implement a GPU version of KPITU that carries out the knee point identification in a parallel manner. This GPU version reduces the worst-case complexity from quadratic to linear. To validate the effectiveness of KPITU, we compare its performance with five state-of-the-art knee point identification methods on 134 test problem instances. Empirical results fully demonstrate the outstanding performance of KPITU especially on problems with many local knee points. At the end, we further validate the usefulness of KPITU for guiding EMO algorithms to search for knee points on the fly during the evolutionary process.
Vertebral body compression fractures are reliable early signs of osteoporosis. Though these fractures are visible on Computed Tomography (CT) images, they are frequently missed by radiologists in clinical settings. Prior research on automatic methods of vertebral fracture classification proves its reliable quality; however, existing methods provide hard-to-interpret outputs and sometimes fail to process cases with severe abnormalities such as highly pathological vertebrae or scoliosis. We propose a new two-step algorithm to localize the vertebral column in 3D CT images and then to simultaneously detect individual vertebrae and quantify fractures in 2D. We train neural networks for both steps using a simple 6-keypoints based annotation scheme, which corresponds precisely to current medical recommendation. Our algorithm has no exclusion criteria, processes 3D CT in 2 seconds on a single GPU, and provides an intuitive and verifiable output. The method approaches expert-level performance and demonstrates state-of-the-art results in vertebrae 3D localization (the average error is 1 mm), vertebrae 2D detection (precision is 0.99, recall is 1), and fracture identification (ROC AUC at the patient level is 0.93).
Neural machine translation (NMT) is nowadays commonly applied at the subword level, using byte-pair encoding. A promising alternative approach focuses on character-level translation, which simplifies processing pipelines in NMT considerably. This approach, however, must consider relatively longer sequences, rendering the training process prohibitively expensive. In this paper, we discuss a novel, Transformer-based approach, that we compare, both in speed and in quality to the Transformer at subword and character levels, as well as previously developed character-level models. We evaluate our models on 4 language pairs from WMT'15: DE-EN, CS-EN, FI-EN and RU-EN. The proposed novel architecture can be trained on a single GPU and is 34% percent faster than the character-level Transformer; still, the obtained results are at least on par with it. In addition, our proposed model outperforms the subword-level model in FI-EN and shows close results in CS-EN. To stimulate further research in this area and close the gap with subword-level NMT, we make all our code and models publicly available.
Nowadays, vision-based computing tasks play an important role in various real-world applications. However, many vision computing tasks, e.g. semantic segmentation, are usually computationally expensive, posing a challenge to the computing systems that are resource-constrained but require fast response speed. Therefore, it is valuable to develop accurate and real-time vision processing models that only require limited computational resources. To this end, we propose the Spatial-detail Guided Context Propagation Network (SGCPNet) for achieving real-time semantic segmentation. In SGCPNet, we propose the strategy of spatial-detail guided context propagation. It uses the spatial details of shallow layers to guide the propagation of the low-resolution global contexts, in which the lost spatial information can be effectively reconstructed. In this way, the need for maintaining high-resolution features along the network is freed, therefore largely improving the model efficiency. On the other hand, due to the effective reconstruction of spatial details, the segmentation accuracy can be still preserved. In the experiments, we validate the effectiveness and efficiency of the proposed SGCPNet model. On the Citysacpes dataset, for example, our SGCPNet achieves 69.5% mIoU segmentation accuracy, while its speed reaches 178.5 FPS on 768x1536 images on a GeForce GTX 1080 Ti GPU card. In addition, SGCPNet is very lightweight and only contains 0.61 M parameters.
SymJAX is a symbolic programming version of JAX simplifying graph input/output/updates and providing additional functionalities for general machine learning and deep learning applications. From an user perspective SymJAX provides a la Theano experience with fast graph optimization/compilation and broad hardware support, along with Lasagne-like deep learning functionalities.
We present PyChain, a fully parallelized PyTorch implementation of end-to-end lattice-free maximum mutual information (LF-MMI) training for the so-called \emph{chain models} in the Kaldi automatic speech recognition (ASR) toolkit. Unlike other PyTorch and Kaldi based ASR toolkits, PyChain is designed to be as flexible and light-weight as possible so that it can be easily plugged into new ASR projects, or other existing PyTorch-based ASR tools, as exemplified respectively by a new project PyChain-example, and Espresso, an existing end-to-end ASR toolkit. PyChain's efficiency and flexibility is demonstrated through such novel features as full GPU training on numerator/denominator graphs, and support for unequal length sequences. Experiments on the WSJ dataset show that with simple neural networks and commonly used machine learning techniques, PyChain can achieve competitive results that are comparable to Kaldi and better than other end-to-end ASR systems.
Recently data-driven image inpainting methods have made inspiring progress, impacting fundamental image editing tasks such as object removal and damaged image repairing. These methods are more effective than classic approaches, however, due to memory limitations they can only handle low-resolution inputs, typically smaller than 1K. Meanwhile, the resolution of photos captured with mobile devices increases up to 8K. Naive up-sampling of the low-resolution inpainted result can merely yield a large yet blurry result. Whereas, adding a high-frequency residual image onto the large blurry image can generate a sharp result, rich in details and textures. Motivated by this, we propose a Contextual Residual Aggregation (CRA) mechanism that can produce high-frequency residuals for missing contents by weighted aggregating residuals from contextual patches, thus only requiring a low-resolution prediction from the network. Since convolutional layers of the neural network only need to operate on low-resolution inputs and outputs, the cost of memory and computing power is thus well suppressed. Moreover, the need for high-resolution training datasets is alleviated. In our experiments, we train the proposed model on small images with resolutions 512x512 and perform inference on high-resolution images, achieving compelling inpainting quality. Our model can inpaint images as large as 8K with considerable hole sizes, which is intractable with previous learning-based approaches. We further elaborate on the light-weight design of the network architecture, achieving real-time performance on 2K images on a GTX 1080 Ti GPU. Codes are available at: Atlas200dk/sample-imageinpainting-HiFill.
We introduce BlaBla, an open-source Python library for extracting linguistic features with proven clinical relevance to neurological and psychiatric diseases across many languages. BlaBla is a unifying framework for accelerating and simplifying clinical linguistic research. The library is built on state-of-the-art NLP frameworks and supports multithreaded/GPU-enabled feature extraction via both native Python calls and a command line interface. We describe BlaBla's architecture and clinical validation of its features across 12 diseases. We further demonstrate the application of BlaBla to a task visualizing and classifying language disorders in three languages on real clinical data from the AphasiaBank dataset. We make the codebase freely available to researchers with the hope of providing a consistent, well-validated foundation for the next generation of clinical linguistic research.
Industrial Control Systems depend heavily on security and monitoring protocols. Several tools are available for this purpose, which scout vulnerabilities and take screenshots from various control panels for later analysis. However, they do not adequately classify images into specific control groups, which can difficult operations performed by manual operators. In order to solve this problem, we use transfer learning with five CNN architectures, pre-trained on Imagenet, to determine which one best classifies screenshots obtained from Industrial Controls Systems. Using 337 manually labeled images, we train these architectures and study their performance both in accuracy and CPU and GPU time. We find out that MobilenetV1 is the best architecture based on its 97,95% of F1-Score, and its speed on CPU with 0.47 seconds per image. In systems where time is critical and GPU is available, VGG16 is preferable because it takes 0.04 seconds to process images, but dropping performance to 87,67%.
In this paper, we design a simple yet powerful deep network architecture, U$^2$-Net, for salient object detection (SOD). The architecture of our U$^2$-Net is a two-level nested U-structure. The design has the following advantages: (1) it is able to capture more contextual information from different scales thanks to the mixture of receptive fields of different sizes in our proposed ReSidual U-blocks (RSU), (2) it increases the depth of the whole architecture without significantly increasing the computational cost because of the pooling operations used in these RSU blocks. This architecture enables us to train a deep network from scratch without using backbones from image classification tasks. We instantiate two models of the proposed architecture, U$^2$-Net (176.3 MB, 30 FPS on GTX 1080Ti GPU) and U$^2$-Net$^{\dagger}$ (4.7 MB, 40 FPS), to facilitate the usage in different environments. Both models achieve competitive performance on six SOD datasets. The code is available: https://github.com/NathanUA/U-2-Net.
GPU-based algorithms have greatly accelerated many machine learning methods; however, GPU memory is typically smaller than main memory, limiting the size of training data. In this paper, we describe an out-of-core GPU gradient boosting algorithm implemented in the XGBoost library. We show that much larger datasets can fit on a given GPU, without degrading model accuracy or training time. To the best of our knowledge, this is the first out-of-core GPU implementation of gradient boosting. Similar approaches can be applied to other machine learning algorithms
Vehicle tracking is an integral part of intelligent traffic management systems. Previous implementations of vehicle tracking used Global Positioning System(GPS) based systems that gave location of the vehicle of an individual on their smartphones.The proposed system uses a novel approach to vehicle tracking using Vehicle License plate detection and recognition (VLPR) technique, which can be integrated on a large scale with traffic management systems. Initial methods of implementing VLPR used simple image processing techniques which were quite experimental and heuristic. With the onset of Deep learning and Computer Vision, one can create robust VLPR systems that can produce results close to human efficiency. Previous implementations, based on deep learning, made use of object detection and support vector machines for detection and a heuristic image processing based approach for recognition. The proposed system makes use of scene text detection model architecture for License plate detection and for recognition it uses the Optical character recognition engine (OCR) Tesseract. The proposed system obtained extraordinary results when it was tested on a highway video using NVIDIA Ge-force RTX 2080ti GPU, results were obtained at a speed of 30 frames per second with accuracy close to human.