Dense 3D object reconstruction from a single image has recently witnessed remarkable advances, but supervising neural networks with ground-truth 3D shapes is impractical due to the laborious process of creating paired image-shape datasets. Recent efforts have turned to learning 3D reconstruction without 3D supervision from RGB images with annotated 2D silhouettes, dramatically reducing the cost and effort of annotation. These techniques, however, remain impractical as they still require multi-view annotations of the same object instance during training. As a result, most experimental efforts to date have been limited to synthetic datasets. In this paper, we address this issue and propose SDF-SRN, an approach that requires only a single view of objects at training time, offering greater utility for real-world scenarios. SDF-SRN learns implicit 3D shape representations to handle arbitrary shape topologies that may exist in the datasets. To this end, we derive a novel differentiable rendering formulation for learning signed distance functions (SDF) from 2D silhouettes. Our method outperforms the state of the art under challenging single-view supervision settings on both synthetic and real-world datasets.

Distributed training techniques have been widely deployed in large-scale deep neural networks (DNNs) training on dense-GPU clusters. However, on public cloud clusters, due to the moderate inter-connection bandwidth between instances, traditional state-of-the-art distributed training systems cannot scale well in training large-scale models. In this paper, we propose a new computing and communication efficient top-k sparsification communication library for distributed training. To further improve the system scalability, we optimize I/O by proposing a simple yet efficient multi-level data caching mechanism and optimize the update operation by introducing a novel parallel tensor operator. Experimental results on a 16-node Tencent Cloud cluster (each node with 8 Nvidia Tesla V100 GPUs) show that our system achieves 25%-40% faster than existing state-of-the-art systems on CNNs and Transformer. We finally break the record on DAWNBench on training ResNet-50 to 93% top-5 accuracy on ImageNet.

Building models for realistic natural language tasks requires dealing with long texts and accounting for complicated structural dependencies. Neural-symbolic representations have emerged as a way to combine the reasoning capabilities of symbolic methods, with the expressiveness of neural networks. However, most of the existing frameworks for combining neural and symbolic representations have been designed for classic relational learning tasks that work over a universe of symbolic entities and relations. In this paper, we present DRaiL, an open-source declarative framework for specifying deep relational models, designed to support a variety of NLP scenarios. Our framework supports easy integration with expressive language encoders, and provides an interface to study the interactions between representation, inference and learning.

Graph convolution is a fundamental building block for many deep neural networks on graph-structured data. In this paper, we introduce a simple, yet very effective graph convolutional network with skip connections for semi-supervised anomaly detection. The proposed multi-layer network architecture is theoretically motivated by the concept of implicit fairing in geometry processing, and comprises a graph convolution module for aggregating information from immediate node neighbors and a skip connection module for combining layer-wise neighborhood representations. In addition to capturing information from distant graph nodes through skip connections between the network's layers, our approach exploits both the graph structure and node features for learning discriminative node representations. The effectiveness of our model is demonstrated through extensive experiments on five benchmark datasets, achieving better or comparable anomaly detection results against strong baseline methods.

Popular task-oriented dialog data sets such as MultiWOZ (Budzianowski et al. 2018) are created by providing crowd-sourced workers a goal instruction, expressed in natural language, that describes the task to be accomplished. Crowd-sourced workers play the role of a user and an agent to generate dialogs to accomplish tasks involving booking restaurant tables, making train reservations, calling a taxi etc. However, creating large crowd-sourced datasets can be time consuming and expensive. To reduce the cost associated with generating such dialog datasets, recent work has explored methods to automatically create larger datasets from small samples.In this paper, we present a data creation strategy that uses the pre-trained language model, GPT2 (Radford et al. 2018), to simulate the interaction between crowd-sourced workers by creating a user bot and an agent bot. We train the simulators using a smaller percentage of actual crowd-generated conversations and their corresponding goal instructions. We demonstrate that by using the simulated data, we achieve significant improvements in both low-resource setting as well as in over-all task performance. To the best of our knowledge we are the first to present a model for generating entire conversations by simulating the crowd-sourced data collection process

State-of-the-art conversational agents have advanced significantly in conjunction with the use of large transformer-based language models. However, even with these advancements, conversational agents still lack the ability to produce responses that are informative and coherent with the local context. In this work, we propose a dialog framework that incorporates both local knowledge as well as users' past dialogues to generate high quality conversations. We introduce an approach to build a dataset based on Reddit conversations, where outbound URL links are widely available in the conversations and the hyperlinked documents can be naturally included as local external knowledge. Using our framework and dataset, we demonstrate that incorporating local knowledge can largely improve informativeness, coherency and realisticness measures using human evaluations. In particular, our approach consistently outperforms the state-of-the-art conversational model on the Reddit dataset across all three measures. We also find that scaling the size of our models from 117M to 8.3B parameters yields consistent improvement of validation perplexity as well as human evaluated metrics. Our model with 8.3B parameters can generate human-like responses as rated by various human evaluations in a single-turn dialog setting.

Document image enhancement and binarization methods are often used to improve the accuracy and efficiency of document image analysis tasks such as text recognition. Traditional non-machine-learning methods are constructed on low-level features in an unsupervised manner but have difficulty with binarization on documents with severely degraded backgrounds. Convolutional neural network-based methods focus only on grayscale images and on local textual features. In this paper, we propose a two-stage color document image enhancement and binarization method using generative adversarial neural networks. In the first stage, four color-independent adversarial networks are trained to extract color foreground information from an input image for document image enhancement. In the second stage, two independent adversarial networks with global and local features are trained for image binarization of documents of variable size. For the adversarial neural networks, we formulate loss functions between a discriminator and generators having an encoder-decoder structure. Experimental results show that the proposed method achieves better performance than many classical and state-of-the-art algorithms over the Document Image Binarization Contest (DIBCO) datasets, the LRDE Document Binarization Dataset (LRDE DBD), and our shipping label image dataset.

We consider the fundamental problem of how to automatically construct summary statistics for implicit generative models where the evaluation of likelihood function is intractable but sampling / simulating data from the model is possible. The idea is to frame the task of constructing sufficient statistics as learning mutual information maximizing representation of the data. This representation is computed by a deep neural network trained by a joint statistic-posterior learning strategy. We apply our approach to both traditional approximate Bayesian computation (ABC) and recent neural likelihood approaches, boosting their performance on a range of tasks.

Medical head CT-scan imaging has been successfully combined with deep learning for medical diagnostics of head diseases and lesions[1]. State of the art classification models and algorithms for this task usually are based on 3d convolution layers for volumetric data on a supervised learning setting (1 input volume, 1 prediction per patient) or 2d convolution layers in a supervised setting (1 input image, 1 prediction per image). However a very common scenario in developing countries is to have the volume metadata lost due multiple reasons for example formatting conversion in images (for example .dicom to jpg), in this scenario the doctor analyses the collection of images and then emits a single diagnostic for the patient (with possibly an unfixed and variable number of images per patient) , this prevents it from being possible to use state of the art 3d models, but also is not possible to convert it to a supervised problem in a (1 image,1 diagnostic) setting because different angles or positions of the images for a single patient may not contain the disease or lesion. In this study we propose a solution for this scenario by combining 2d convolutional[2] models with sequence models which generate a prediction only after all images have been processed by the model for a given patient \(i\), this creates a multi-image to single-diagnostic setting \(y^i=f(x_1,x_2,..,x_n)\) where \(n\) may be different between patients. The experimental results demonstrate that it is possible to get a multi-image to single diagnostic model which mimics human doctor diagnostic process: evaluate the collection of patient images and then use important information in memory to decide a single diagnostic for the patient.

Neural machine translation has achieved remarkable empirical performance over standard benchmark datasets, yet recent evidence suggests that the models can still fail easily dealing with substandard inputs such as misspelled words, To overcome this issue, we introduce a new encoding heuristic of the input symbols for character-level NLP models: it encodes the shape of each character through the images depicting the letters when printed. We name this new strategy visual embedding and it is expected to improve the robustness of NLP models because humans also process the corpus visually through printed letters, instead of machinery one-hot vectors. Empirically, our method improves models' robustness against substandard inputs, even in the test scenario where the models are tested with the noises that are beyond what is available during the training phase.