Seismic full-waveform inversion (FWI) techniques aim to find a high-resolution subsurface geophysical model provided with waveform data. Some recent effort in data-driven FWI has shown some encouraging results in obtaining 2D velocity maps. However, due to high computational complexity and large memory consumption, the reconstruction of 3D high-resolution velocity maps via deep networks is still a great challenge. In this paper, we present InversionNet3D, an efficient and scalable encoder-decoder network for 3D FWI. The proposed method employs group convolution in the encoder to establish an effective hierarchy for learning information from multiple sources while cutting down unnecessary parameters and operations at the same time. The introduction of invertible layers further reduces the memory consumption of intermediate features during training and thus enables the development of deeper networks with more layers and higher capacity as required by different application scenarios. Experiments on the 3D Kimberlina dataset demonstrate that InversionNet3D achieves state-of-the-art reconstruction performance with lower computational cost and lower memory footprint compared to the baseline.
Text-based games simulate worlds and interact with players using natural language. Recent work has used them as a testbed for autonomous language-understanding agents, with the motivation being that understanding the meanings of words or semantics is a key component of how humans understand, reason, and act in these worlds. However, it remains unclear to what extent artificial agents utilize semantic understanding of the text. To this end, we perform experiments to systematically reduce the amount of semantic information available to a learning agent. Surprisingly, we find that an agent is capable of achieving high scores even in the complete absence of language semantics, indicating that the currently popular experimental setup and models may be poorly designed to understand and leverage game texts. To remedy this deficiency, we propose an inverse dynamics decoder to regularize the representation space and encourage exploration, which shows improved performance on several games including Zork I. We discuss the implications of our findings for designing future agents with stronger semantic understanding.
Electronic Health Records (EHRs) have become the primary form of medical data-keeping across the United States. Federal law restricts the sharing of any EHR data that contains protected health information (PHI). De-identification, the process of identifying and removing all PHI, is crucial for making EHR data publicly available for scientific research. This project explores several deep learning-based named entity recognition (NER) methods to determine which method(s) perform better on the de-identification task. We trained and tested our models on the i2b2 training dataset, and qualitatively assessed their performance using EHR data collected from a local hospital. We found that 1) BiLSTM-CRF represents the best-performing encoder/decoder combination, 2) character-embeddings and CRFs tend to improve precision at the price of recall, and 3) transformers alone under-perform as context encoders. Future work focused on structuring medical text may improve the extraction of semantic and syntactic information for the purposes of EHR de-identification.
Objective: Deep learning-based neural decoders have emerged as the prominent approach to enable dexterous and intuitive control of neuroprosthetic hands. Yet few studies have materialized the use of deep learning in clinical settings due to its high computational requirements. Methods: Recent advancements of edge computing devices bring the potential to alleviate this problem. Here we present the implementation of a neuroprosthetic hand with embedded deep learning-based control. The neural decoder is designed based on the recurrent neural network (RNN) architecture and deployed on the NVIDIA Jetson Nano - a compacted yet powerful edge computing platform for deep learning inference. This enables the implementation of the neuroprosthetic hand as a portable and self-contained unit with real-time control of individual finger movements. Results: The proposed system is evaluated on a transradial amputee using peripheral nerve signals (ENG) with implanted intrafascicular microelectrodes. The experiment results demonstrate the system's capabilities of providing robust, high-accuracy (95-99%) and low-latency (50-120 msec) control of individual finger movements in various laboratory and real-world environments. Conclusion: Modern edge computing platforms enable the effective use of deep learning-based neural decoders for neuroprosthesis control as an autonomous system. Significance: This work helps pioneer the deployment of deep neural networks in clinical applications underlying a new class of wearable biomedical devices with embedded artificial intelligence.
In this paper, we present DRANet, a network architecture that disentangles image representations and transfers the visual attributes in a latent space for unsupervised cross-domain adaptation. Unlike the existing domain adaptation methods that learn associated features sharing a domain, DRANet preserves the distinctiveness of each domain's characteristics. Our model encodes individual representations of content (scene structure) and style (artistic appearance) from both source and target images. Then, it adapts the domain by incorporating the transferred style factor into the content factor along with learnable weights specified for each domain. This learning framework allows bi-/multi-directional domain adaptation with a single encoder-decoder network and aligns their domain shift. Additionally, we propose a content-adaptive domain transfer module that helps retain scene structure while transferring style. Extensive experiments show our model successfully separates content-style factors and synthesizes visually pleasing domain-transferred images. The proposed method demonstrates state-of-the-art performance on standard digit classification tasks as well as semantic segmentation tasks.
We introduce dense vision transformers, an architecture that leverages vision transformers in place of convolutional networks as a backbone for dense prediction tasks. We assemble tokens from various stages of the vision transformer into image-like representations at various resolutions and progressively combine them into full-resolution predictions using a convolutional decoder. The transformer backbone processes representations at a constant and relatively high resolution and has a global receptive field at every stage. These properties allow the dense vision transformer to provide finer-grained and more globally coherent predictions when compared to fully-convolutional networks. Our experiments show that this architecture yields substantial improvements on dense prediction tasks, especially when a large amount of training data is available. For monocular depth estimation, we observe an improvement of up to 28% in relative performance when compared to a state-of-the-art fully-convolutional network. When applied to semantic segmentation, dense vision transformers set a new state of the art on ADE20K with 49.02% mIoU. We further show that the architecture can be fine-tuned on smaller datasets such as NYUv2, KITTI, and Pascal Context where it also sets the new state of the art. Our models are available at https://github.com/intel-isl/DPT.
This paper introduces the Descriptive Variational Autoencoder (DVAE), an unsupervised and end-to-end trainable neural network for predicting vehicle trajectories that provides partial interpretability. The novel approach is based on the architecture and objective of common variational autoencoders. By introducing expert knowledge within the decoder part of the autoencoder, the encoder learns to extract latent parameters that provide a graspable meaning in human terms. Such an interpretable latent space enables the validation by expert defined rule sets. The evaluation of the DVAE is performed using the publicly available highD dataset for highway traffic scenarios. In comparison to a conventional variational autoencoder with equivalent complexity, the proposed model provides a similar prediction accuracy but with the great advantage of having an interpretable latent space. For crucial decision making and assessing trustworthiness of a prediction this property is highly desirable.
Accurate retinal vessel segmentation is a challenging problem in color fundus image analysis. An automatic retinal vessel segmentation system can effectively facilitate clinical diagnosis and ophthalmological research. Technically, this problem suffers from various degrees of vessel thickness, perception of details, and contextual feature fusion. For addressing these challenges, a deep learning based method has been proposed and several customized modules have been integrated into the well-known encoder-decoder architecture U-net, which is mainly employed in medical image segmentation. Structurally, cascaded dilated convolutional modules have been integrated into the intermediate layers, for obtaining larger receptive field and generating denser encoded feature maps. Also, the advantages of the pyramid module with spatial continuity have been taken, for multi-thickness perception, detail refinement, and contextual feature fusion. Additionally, the effectiveness of different normalization approaches has been discussed in network training for different datasets with specific properties. Experimentally, sufficient comparative experiments have been enforced on three retinal vessel segmentation datasets, DRIVE, CHASEDB1, and the unhealthy dataset STARE. As a result, the proposed method outperforms the work of predecessors and achieves state-of-the-art performance in Sensitivity/Recall, F1-score and MCC.
We propose a hybrid architecture composed of a fully convolutional network (FCN) and a Dempster-Shafer layer for image semantic segmentation. In the so-called evidential FCN (E-FCN), an encoder-decoder architecture first extracts pixel-wise feature maps from an input image. A Dempster-Shafer layer then computes mass functions at each pixel location based on distances to prototypes. Finally, a utility layer performs semantic segmentation from mass functions and allows for imprecise classification of ambiguous pixels and outliers. We propose an end-to-end learning strategy for jointly updating the network parameters, which can make use of soft (imprecise) labels. Experiments using three databases (Pascal VOC 2011, MIT-scene Parsing and SIFT Flow) show that the proposed combination improves the accuracy and calibration of semantic segmentation by assigning confusing pixels to multi-class sets.
In recent years, how to strike a good trade-off between accuracy and inference speed has become the core issue for real-time semantic segmentation applications, which plays a vital role in real-world scenarios such as autonomous driving systems and drones. In this study, we devise a novel lightweight network using a multi-scale context fusion (MSCFNet) scheme, which explores an asymmetric encoder-decoder architecture to dispose this problem. More specifically, the encoder adopts some developed efficient asymmetric residual (EAR) modules, which are composed of factorization depth-wise convolution and dilation convolution. Meanwhile, instead of complicated computation, simple deconvolution is applied in the decoder to further reduce the amount of parameters while still maintaining high segmentation accuracy. Also, MSCFNet has branches with efficient attention modules from different stages of the network to well capture multi-scale contextual information. Then we combine them before the final classification to enhance the expression of the features and improve the segmentation efficiency. Comprehensive experiments on challenging datasets have demonstrated that the proposed MSCFNet, which contains only 1.15M parameters, achieves 71.9\% Mean IoU on the Cityscapes testing dataset and can run at over 50 FPS on a single Titan XP GPU configuration.
We propose a new end-to-end trainable approach for multi-instance pose estimation by combining a convolutional neural network with a transformer. We cast multi-instance pose estimation from images as a direct set prediction problem. Inspired by recent work on end-to-end trainable object detection with transformers, we use a transformer encoder-decoder architecture together with a bipartite matching scheme to directly regress the pose of all individuals in a given image. Our model, called POse Estimation Transformer (POET), is trained using a novel set-based global loss that consists of a keypoint loss, a keypoint visibility loss, a center loss and a class loss. POET reasons about the relations between detected humans and the full image context to directly predict the poses in parallel. We show that POET can achieve high accuracy on the challenging COCO keypoint detection task. To the best of our knowledge, this model is the first end-to-end trainable multi-instance human pose estimation method.
While convolutional neural networks have shown a tremendous impact on various computer vision tasks, they generally demonstrate limitations in explicitly modeling long-range dependencies due to the intrinsic locality of the convolution operation. Initially designed for natural language processing tasks, Transformers have emerged as alternative architectures with innate global self-attention mechanisms to capture long-range dependencies. In this paper, we propose TransDepth, an architecture that benefits from both convolutional neural networks and transformers. To avoid the network losing its ability to capture local-level details due to the adoption of transformers, we propose a novel decoder that employs attention mechanisms based on gates. Notably, this is the first paper that applies transformers to pixel-wise prediction problems involving continuous labels (i.e., monocular depth prediction and surface normal estimation). Extensive experiments demonstrate that the proposed TransDepth achieves state-of-the-art performance on three challenging datasets. Our code is available at: https://github.com/ygjwd12345/TransDepth.
User-intended visual content fills the hole regions of an input image in the image editing scenario. The coarse low-level inputs, which typically consist of sparse sketch lines and color dots, convey user intentions for content creation (\ie, free-form editing). While existing methods combine an input image and these low-level controls for CNN inputs, the corresponding feature representations are not sufficient to convey user intentions, leading to unfaithfully generated content. In this paper, we propose DeFLOCNet which relies on a deep encoder-decoder CNN to retain the guidance of these controls in the deep feature representations. In each skip-connection layer, we design a structure generation block. Instead of attaching low-level controls to an input image, we inject these controls directly into each structure generation block for sketch line refinement and color propagation in the CNN feature space. We then concatenate the modulated features with the original decoder features for structure generation. Meanwhile, DeFLOCNet involves another decoder branch for texture generation and detail enhancement. Both structures and textures are rendered in the decoder, leading to user-intended editing results. Experiments on benchmarks demonstrate that DeFLOCNet effectively transforms different user intentions to create visually pleasing content.
Semantic segmentation, which refers to pixel-wise classification of an image, is a fundamental topic in computer vision owing to its growing importance in robot vision and autonomous driving industries. It provides rich information about objects in the scene such as object boundary, category, and location. Recent methods for semantic segmentation often employ an encoder-decoder structure using deep convolutional neural networks. The encoder part extracts feature of the image using several filters and pooling operations, whereas the decoder part gradually recovers the low-resolution feature maps of the encoder into a full input resolution feature map for pixel-wise prediction. However, the encoder-decoder variants for semantic segmentation suffer from severe spatial information loss, caused by pooling operations or convolutions with stride, and does not consider the context in the scene. In this paper, we propose a dense upsampling convolution method based on guided filtering to effectively preserve the spatial information of the image in the network. We further propose a novel local context convolution method that not only covers larger-scale objects in the scene but covers them densely for precise object boundary delineation. Theoretical analyses and experimental results on several benchmark datasets verify the effectiveness of our method. Qualitatively, our approach delineates object boundaries at a level of accuracy that is beyond the current excellent methods. Quantitatively, we report a new record of 82.86% and 81.62% of pixel accuracy on ADE20K and Pascal-Context benchmark datasets, respectively. In comparison with the state-of-the-art methods, the proposed method offers promising improvements.
Code summaries are brief natural language descriptions of source code pieces. The main purpose of code summarization is to assist developers in understanding code and to reduce documentation workload. In this paper, we design a novel multi-task learning (MTL) approach for code summarization through mining the relationship between method code summaries and method names. More specifically, since a method's name can be considered as a shorter version of its code summary, we first introduce the tasks of generation and informativeness prediction of method names as two auxiliary training objectives for code summarization. A novel two-pass deliberation mechanism is then incorporated into our MTL architecture to generate more consistent intermediate states fed into a summary decoder, especially when informative method names do not exist. To evaluate our deliberation MTL approach, we carried out a large-scale experiment on two existing datasets for Java and Python. The experiment results show that our technique can be easily applied to many state-of-the-art neural models for code summarization and improve their performance. Meanwhile, our approach shows significant superiority when generating summaries for methods with non-informative names.
Non-autoregressive Transformer is a promising text generation model. However, current non-autoregressive models still fall behind their autoregressive counterparts in translation quality. We attribute this accuracy gap to the lack of dependency modeling among decoder inputs. In this paper, we propose CNAT, which learns implicitly categorical codes as latent variables into the non-autoregressive decoding. The interaction among these categorical codes remedies the missing dependencies and improves the model capacity. Experiment results show that our model achieves comparable or better performance in machine translation tasks, compared with several strong baselines.
Being one of the most popular generative framework, variational autoencoders(VAE) are known to suffer from a phenomenon termed posterior collapse, i.e. the latent variational distributions collapse to the prior, especially when a strong decoder network is used. In this work, we analyze the latent representation of collapsed VAEs, and proposed a novel model, neighbor embedding VAE(NE-VAE), which explicitly constraints the encoder to encode inputs close in the input space to be close in the latent space. We observed that for VAE variants that report similar ELBO, KL divergence or even mutual information scores may still behave quite differently in the latent organization. In our experiments, NE-VAE can produce qualitatively different latent representations with majority of the latent dimensions remained active, which may benefit downstream latent space optimization tasks. NE-VAE can prevent posterior collapse to a much greater extent than it's predecessors, and can be easily plugged into any autoencoder framework, without introducing addition model components and complex training routines.
Light fields are 4D scene representation typically structured as arrays of views, or several directional samples per pixel in a single view. This highly correlated structure is not very efficient to transmit and manipulate (especially for editing), though. To tackle these problems, we present a novel compact and editable light field representation, consisting of a set of visual channels (i.e. the central RGB view) and a complementary meta channel that encodes the residual geometric and appearance information. The visual channels in this representation can be edited using existing 2D image editing tools, before accurately reconstructing the whole edited light field back. We propose to learn this representation via an autoencoder framework, consisting of an encoder for learning the representation, and a decoder for reconstructing the light field. To handle the challenging occlusions and propagation of edits, we specifically designed an editing-aware decoding network and its associated training strategy, so that the edits to the visual channels can be consistently propagated to the whole light field upon reconstruction.Experimental results show that our proposed method outperforms related existing methods in reconstruction accuracy, and achieves visually pleasant performance in editing propagation.
Source code summarization of a subroutine is the task of writing a short, natural language description of that subroutine. The description usually serves in documentation aimed at programmers, where even brief phrase (e.g. "compresses data to a zip file") can help readers rapidly comprehend what a subroutine does without resorting to reading the code itself. Techniques based on neural networks (and encoder-decoder model designs in particular) have established themselves as the state-of-the-art. Yet a problem widely recognized with these models is that they assume the information needed to create a summary is present within the code being summarized itself - an assumption which is at odds with program comprehension literature. Thus a current research frontier lies in the question of encoding source code context into neural models of summarization. In this paper, we present a project-level encoder to improve models of code summarization. By project-level, we mean that we create a vectorized representation of selected code files in a software project, and use that representation to augment the encoder of state-of-the-art neural code summarization techniques. We demonstrate how our encoder improves several existing models, and provide guidelines for maximizing improvement while controlling time and resource costs in model size.
Remaining useful life (RUL) estimation is a crucial component in the implementation of intelligent predictive maintenance and health management. Deep neural network (DNN) approaches have been proven effective in RUL estimation due to their capacity in handling high-dimensional non-linear degradation features. However, the applications of DNN in practice face two challenges: (a) online update of lifetime information is often unavailable, and (b) uncertainties in predicted values may not be analytically quantified. This paper addresses these issues by developing a hybrid DNN-based prognostic approach, where a Wiener-based-degradation model is enhanced with adaptive drift to characterize the system degradation. An LSTM-CNN encoder-decoder is developed to predict future degradation trajectories by jointly learning noise coefficients as well as drift coefficients, and adaptive drift is updated via Bayesian inference. A computationally efficient algorithm is proposed for the calculation of RUL distributions. Numerical experiments are presented using turbofan engines degradation data to demonstrate the superior accuracy of RUL prediction of our proposed approach.