Document Grounded Conversations is a task to generate dialogue responses when chatting about the content of a given document. Obviously, document knowledge plays a critical role in Document Grounded Conversations, while existing dialogue models do not exploit this kind of knowledge effectively enough. In this paper, we propose a novel Transformer-based architecture for multi-turn document grounded conversations. In particular, we devise an Incremental Transformer to encode multi-turn utterances along with knowledge in related documents. Motivated by the human cognitive process, we design a two-pass decoder (Deliberation Decoder) to improve context coherence and knowledge correctness. Our empirical study on a real-world Document Grounded Dataset proves that responses generated by our model significantly outperform competitive baselines on both context coherence and knowledge relevance.
Automatic instrument segmentation in video is an essentially fundamental yet challenging problem for robot-assisted minimally invasive surgery. In this paper, we propose a novel framework to leverage instrument motion information, by incorporating a derived temporal prior to an attention pyramid network for accurate segmentation. Our inferred prior can provide reliable indication of the instrument location and shape, which is propagated from the previous frame to the current frame according to inter-frame motion flow. This prior is injected to the middle of an encoder-decoder segmentation network as an initialization of a pyramid of attention modules, to explicitly guide segmentation output from coarse to fine. In this way, the temporal dynamics and the attention network can effectively complement and benefit each other. As additional usage, our temporal prior enables semi-supervised learning with periodically unlabeled video frames, simply by reverse execution. We extensively validate our method on the public 2017 MICCAI EndoVis Robotic Instrument Segmentation Challenge dataset with three different tasks. Our method consistently exceeds the state-of-the-art results across all three tasks by a large margin. Our semi-supervised variant also demonstrates a promising potential for reducing annotation cost in the clinical practice.
Despite the advancement in the technology of autonomous driving cars, the safety of a self-driving car is still a challenging problem that has not been well studied. Motion prediction is one of the core functions of an autonomous driving car. Previously, we propose a novel scheme called GRIP which is designed to predict trajectories for traffic agents around an autonomous car efficiently. GRIP uses a graph to represent the interactions of close objects, applies several graph convolutional blocks to extract features, and subsequently uses an encoder-decoder long short-term memory (LSTM) model to make predictions. Even though our experimental results show that GRIP improves the prediction accuracy of the state-of-the-art solution by 30%, GRIP still has some limitations. GRIP uses a fixed graph to describe the relationships between different traffic agents and hence may suffer some performance degradations when it is being used in urban traffic scenarios. Hence, in this paper, we describe an improved scheme called GRIP++ where we use both fixed and dynamic graphs for trajectory predictions of different types of traffic agents. Such an improvement can help autonomous driving cars avoid many traffic accidents. Our evaluations using a recently released urban traffic dataset, namely ApolloScape showed that GRIP++ achieves better prediction accuracy than state-of-the-art schemes. GRIP++ ranked #1 on the leaderboard of the ApolloScape trajectory competition in October 2019. In addition, GRIP++ runs 21.7 times faster than a state-of-the-art scheme, CS-LSTM.
Segmenting stroke lesions from T1-weighted MR images is of great value for large-scale stroke rehabilitation neuroimaging analyses. Nevertheless, there are great challenges with this task, such as large range of stroke lesion scales and the tissue intensity similarity. The famous encoder-decoder convolutional neural network, which although has made great achievements in medical image segmentation areas, may fail to address these challenges due to the insufficient uses of multi-scale features and context information. To address these challenges, this paper proposes a Cross-Level fusion and Context Inference Network (CLCI-Net) for the chronic stroke lesion segmentation from T1-weighted MR images. Specifically, a Cross-Level feature Fusion (CLF) strategy was developed to make full use of different scale features across different levels; Extending Atrous Spatial Pyramid Pooling (ASPP) with CLF, we have enriched multi-scale features to handle the different lesion sizes; In addition, convolutional long short-term memory (ConvLSTM) is employed to infer context information and thus capture fine structures to address the intensity similarity issue. The proposed approach was evaluated on an open-source dataset, the Anatomical Tracings of Lesions After Stroke (ATLAS) with the results showing that our network outperforms five state-of-the-art methods. We make our code and models available at https://github.com/YH0517/CLCI_Net.
Knowledge graph embedding methods often suffer from a limitation of memorizing valid triples to predict new ones for triple classification and search personalization problems. To this end, we introduce a novel embedding model, named R-MeN, that explores a relational memory network to encode potential dependencies in relationship triples. R-MeN considers each triple as a sequence of 3 input vectors that recurrently interact with a memory using a transformer self-attention mechanism. Thus R-MeN encodes new information from interactions between the memory and each input vector to return a corresponding vector. Consequently, R-MeN feeds these 3 returned vectors to a convolutional neural network-based decoder to produce a scalar score for the triple. Experimental results show that our proposed R-MeN obtains state-of-the-art results on SEARCH17 for the search personalization task, and on WN11 and FB13 for the triple classification task.
Monocular depth estimation is an essential task for scene understanding. The underlying structure of objects and stuff in a complex scene is critical to recovering accurate and visually-pleasing depth maps. Global structure conveys scene layouts, while local structure reflects shape details. Recently developed approaches based on convolutional neural networks (CNNs) significantly improve the performance of depth estimation. However, few of them take into account multi-scale structures in complex scenes. In this paper, we propose a Structure-Aware Residual Pyramid Network (SARPN) to exploit multi-scale structures for accurate depth prediction. We propose a Residual Pyramid Decoder (RPD) which expresses global scene structure in upper levels to represent layouts, and local structure in lower levels to present shape details. At each level, we propose Residual Refinement Modules (RRM) that predict residual maps to progressively add finer structures on the coarser structure predicted at the upper level. In order to fully exploit multi-scale image features, an Adaptive Dense Feature Fusion (ADFF) module, which adaptively fuses effective features from all scales for inferring structures of each scale, is introduced. Experiment results on the challenging NYU-Depth v2 dataset demonstrate that our proposed approach achieves state-of-the-art performance in both qualitative and quantitative evaluation. The code is available at https://github.com/Xt-Chen/SARPN.
Predicting distant future trajectories of agents in a dynamic scene is not an easy problem because the future trajectory of an agent is affected by not only his/her past trajectory but also the scene contexts. To tackle this problem, we propose a model based on recurrent neural networks (RNNs) and a novel method for training the model. The proposed model is based on an encoder-decoder architecture where the encoder encodes inputs (past trajectories and scene context information) while the decoder produces a trajectory from the context vector given by the encoder. We train the networks of the proposed model to produce a future trajectory, which is the closest to the true trajectory, while maximizing a reward from a reward function. The reward function is also trained at the same time to maximize the margin between the rewards from the ground-truth trajectory and its estimate. The reward function plays the role of a regularizer for the proposed model so the trained networks are able to better utilize the scene context information for the prediction task. We evaluated the proposed model on several public datasets. Experimental results show that the prediction performance of the proposed model is much improved by the regularization, which outperforms the-state-of-the-arts in terms of accuracy. The implementation codes are available at https://github.com/d1024choi/traj-pred-irl/.
Existing methods in the Visual Storytelling field often suffer from the problem of generating general descriptions, while the image contains a lot of meaningful contents remaining unnoticed. The failure of informative story generation can be concluded to the model's incompetence of capturing enough meaningful concepts. The categories of these concepts include entities, attributes, actions, and events, which are in some cases crucial to grounded storytelling. To solve this problem, we propose a method to mine the cross-modal rules to help the model infer these informative concepts given certain visual input. We first build the multimodal transactions by concatenating the CNN activations and the word indices. Then we use the association rule mining algorithm to mine the cross-modal rules, which will be used for the concept inference. With the help of the cross-modal rules, the generated stories are more grounded and informative. Besides, our proposed method holds the advantages of interpretation, expandability, and transferability, indicating potential for wider application. Finally, we leverage these concepts in our encoder-decoder framework with the attention mechanism. We conduct several experiments on the VIsual StoryTelling~(VIST) dataset, the results of which demonstrate the effectiveness of our approach in terms of both automatic metrics and human evaluation. Additional experiments are also conducted showing that our mined cross-modal rules as additional knowledge helps the model gain better performance when trained on a small dataset.
Head-driven phrase structure grammar (HPSG) enjoys a uniform formalism representing rich contextual syntactic and even semantic meanings. This paper makes the first attempt to formulate a simplified HPSG by integrating constituent and dependency formal representations into head-driven phrase structure. Then two parsing algorithms are respectively proposed for two converted tree representations, division span and joint span. As HPSG encodes both constituent and dependency structure information, the proposed HPSG parsers may be regarded as a sort of joint decoder for both types of structures and thus are evaluated in terms of extracted or converted constituent and dependency parsing trees. Our parser achieves new state-of-the-art performance for both parsing tasks on Penn Treebank (PTB) and Chinese Penn Treebank, verifying the effectiveness of joint learning constituent and dependency structures. In details, we report 96.33 F1 of constituent parsing and 97.20\% UAS of dependency parsing on PTB.
Research on style transfer and domain translation has clearly demonstrated the ability of deep learning-based algorithms to manipulate images in terms of artistic style. More recently, several attempts have been made to extend such approaches to music (both symbolic and audio) in order to enable transforming musical style in a similar manner. In this study, we focus on symbolic music with the goal of altering the 'style' of a piece while keeping its original 'content'. As opposed to the current methods, which are inherently restricted to be unsupervised due to the lack of 'aligned' data (i.e. the same musical piece played in multiple styles), we develop the first fully supervised algorithm for this task. At the core of our approach lies a synthetic data generation scheme which allows us to produce virtually unlimited amounts of aligned data, and hence avoid the above issue. In view of this data generation scheme, we propose an encoder-decoder model for translating symbolic music accompaniments between a number of different styles. Our experiments show that our models, although trained entirely on synthetic data, are capable of producing musically meaningful accompaniments even for real (non-synthetic) MIDI recordings.
Terahertz (THz) sensing is a promising imaging technology for a wide variety of different applications. Extracting the interpretable and physically meaningful parameters for such applications, however, requires solving an inverse problem in which a model function determined by these parameters needs to be fitted to the measured data. Since the underlying optimization problem is nonconvex and very costly to solve, we propose learning the prediction of suitable parameters from the measured data directly. More precisely, we develop a model-based autoencoder in which the encoder network predicts suitable parameters and the decoder is fixed to a physically meaningful model function, such that we can train the encoding network in an unsupervised way. We illustrate numerically that the resulting network is more than 140 times faster than classical optimization techniques while making predictions with only slightly higher objective values. Using such predictions as starting points of local optimization techniques allows us to converge to better local minima about twice as fast as optimization without the network-based initialization.
Facial makeup transfer is a widely-used technology that aims to transfer the makeup style from a reference face image to a non-makeup face. Existing literature leverage the adversarial loss so that the generated faces are of high quality and realistic as real ones, but are only able to produce fixed outputs. Inspired by recent advances in disentangled representation, in this paper we propose DMT (Disentangled Makeup Transfer), a unified generative adversarial network to achieve different scenarios of makeup transfer. Our model contains an identity encoder as well as a makeup encoder to disentangle the personal identity and the makeup style for arbitrary face images. Based on the outputs of the two encoders, a decoder is employed to reconstruct the original faces. We also apply a discriminator to distinguish real faces from fake ones. As a result, our model can not only transfer the makeup styles from one or more reference face images to a non-makeup face with controllable strength, but also produce various outputs with styles sampled from a prior distribution. Extensive experiments demonstrate that our model is superior to existing literature by generating high-quality results for different scenarios of makeup transfer.
The transformer is a state-of-the-art neural translation model that uses attention to iteratively refine lexical representations with information drawn from the surrounding context. Lexical features are fed into the first layer and propagated through a deep network of hidden layers. We argue that the need to represent and propagate lexical features in each layer limits the model's capacity for learning and representing other information relevant to the task. To alleviate this bottleneck, we introduce gated shortcut connections between the embedding layer and each subsequent layer within the encoder and decoder. This enables the model to access relevant lexical content dynamically, without expending limited resources on storing it within intermediate states. We show that the proposed modification yields consistent improvements over a baseline transformer on standard WMT translation tasks in 5 translation directions (0.9 BLEU on average) and reduces the amount of lexical information passed along the hidden layers. We furthermore evaluate different ways to integrate lexical connections into the transformer architecture and present ablation experiments exploring the effect of proposed shortcuts on model behavior.
Retinal vessel segmentation is a crucial step in diagnosing and screening various diseases, including diabetes, ophthalmologic diseases, and cardiovascular diseases. In this paper, we propose an effective and efficient method for vessel segmentation in color fundus images using encoder-decoder based octave convolution networks. Compared with other convolution networks utilizing standard convolution for feature extraction, the proposed method utilizes octave convolutions and octave transposed convolutions for learning multiple-spatial-frequency features, thus can better capture retinal vasculatures with varying sizes and shapes. To provide the network the capability of learning how to decode multifrequency features, we extend octave convolution and propose a new operation named octave transposed convolution. A novel architecture of convolutional neural network, named as Octave UNet integrating both octave convolutions and octave transposed convolutions is proposed based on the encoder-decoder architecture of UNet, which can generate high resolution vessel segmentation in one single forward feeding without post-processing steps. Comprehensive experimental results demonstrate that the proposed Octave UNet outperforms the baseline UNet achieving better or comparable performance to the state-of-the-art methods with fast processing speed. Specifically, the proposed method achieves 0.9664 / 0.9713 / 0.9759 / 0.9698 accuracy, 0.8374 / 0.8664 / 0.8670 / 0.8076 sensitivity, 0.9790 / 0.9798 / 0.9840 / 0.9831 specificity, 0.8127 / 0.8191 / 0.8313 / 0.7963 F1 score, and 0.9835 / 0.9875 / 0.9905 / 0.9845 Area Under Receiver Operating Characteristic curve, on DRIVE, STARE, CHASE_DB1, and HRF datasets, respectively.
Modern generative models are usually designed to match target distributions directly in the data space, where the intrinsic dimension of data can be much lower than the ambient dimension. We argue that this discrepancy may contribute to the difficulties in training generative models. We therefore propose to map both the generated and target distributions to a latent space using the encoder of a standard autoencoder, and train the generator (or decoder) to match the target distribution in the latent space. Specifically, we enforce the consistency in both the data space and the latent space with theoretically justified data and latent reconstruction losses. The resulting generative model, which we call a perceptual generative autoencoder (PGA), is then trained with a maximum likelihood or variational autoencoder (VAE) objective. With maximum likelihood, PGAs generalize the idea of reversible generative models to unrestricted neural network architectures and arbitrary number of latent dimensions. When combined with VAEs, PGAs substantially improve over the baseline VAEs in terms of sample quality. Compared to other autoencoder-based generative models using simple priors, PGAs achieve state-of-the-art FID scores on CIFAR-10 and CelebA.
Non-Autoregressive Transformer (NAT) aims to accelerate the Transformer model through discarding the autoregressive mechanism and generating target words independently, which fails to exploit the target sequential information. Over-translation and under-translation errors often occur for the above reason, especially in the long sentence translation scenario. In this paper, we propose two approaches to retrieve the target sequential information for NAT to enhance its translation ability while preserving the fast-decoding property. Firstly, we propose a sequence-level training method based on a novel reinforcement algorithm for NAT (Reinforce-NAT) to reduce the variance and stabilize the training procedure. Secondly, we propose an innovative Transformer decoder named FS-decoder to fuse the target sequential information into the top layer of the decoder. Experimental results on three translation tasks show that the Reinforce-NAT surpasses the baseline NAT system by a significant margin on BLEU without decelerating the decoding speed and the FS-decoder achieves comparable translation performance to the autoregressive Transformer with considerable speedup.
Latest development of neural models has connected the encoder and decoder through a self-attention mechanism. In particular, Transformer, which is solely based on self-attention, has led to breakthroughs in Natural Language Processing (NLP) tasks. However, the multi-head attention mechanism, as a key component of Transformer, limits the effective deployment of the model to a resource-limited setting. In this paper, based on the ideas of tensor decomposition and parameters sharing, we propose a novel self-attention model (namely Multi-linear attention) with Block-Term Tensor Decomposition (BTD). We test and verify the proposed attention method on three language modeling tasks (i.e., PTB, WikiText-103 and One-billion) and a neural machine translation task (i.e., WMT-2016 English-German). Multi-linear attention can not only largely compress the model parameters but also obtain performance improvements, compared with a number of language modeling approaches, such as Transformer, Transformer-XL, and Transformer with tensor train decomposition.
In this paper we explore techniques for generating new music using a Variational Autoencoder (VAE) neural network that was trained on a corpus of specific style. Instead of randomly sampling the latent states of the network to produce free improvisation, we generate new music by querying the network with musical input in a style different from the training corpus. This allows us to produce new musical output with longer-term structure that blends aspects of the query to the style of the network. In order to control the level of this blending we add a noisy channel between the VAE encoder and decoder using bit-allocation algorithm from communication rate-distortion theory. Our experiments provide new insight into relations between the representational and structural information of latent states and the query signal, suggesting their possible use for composition purposes.
Zero-shot learning (ZSL) and cold-start recommendation (CSR) are two challenging problems in computer vision and recommender system, respectively. In general, they are independently investigated in different communities. This paper, however, reveals that ZSL and CSR are two extensions of the same intension. Both of them, for instance, attempt to predict unseen classes and involve two spaces, one for direct feature representation and the other for supplementary description. Yet there is no existing approach which addresses CSR from the ZSL perspective. This work, for the first time, formulates CSR as a ZSL problem, and a tailor-made ZSL method is proposed to handle CSR. Specifically, we propose a Low-rank Linear Auto-Encoder (LLAE), which challenges three cruxes, i.e., domain shift, spurious correlations and computing efficiency, in this paper. LLAE consists of two parts, a low-rank encoder maps user behavior into user attributes and a symmetric decoder reconstructs user behavior from user attributes. Extensive experiments on both ZSL and CSR tasks verify that the proposed method is a win-win formulation, i.e., not only can CSR be handled by ZSL models with a significant performance improvement compared with several conventional state-of-the-art methods, but the consideration of CSR can benefit ZSL as well.
Generative Adversarial Networks (GANs) play an increasingly important role in machine learning. However, there is one fundamental issue hindering their practical applications: the absence of capability for encoding real-world samples. The conventional way of addressing this issue is to learn an encoder for GAN via Variational Auto-Encoder (VAE). In this paper, we show that the entanglement of the latent space for the VAE/GAN framework poses the main challenge for encoder learning. To address the entanglement issue and enable inference in GAN we propose a novel algorithm named Latently Invertible Autoencoder (LIA). The framework of LIA is that an invertible network and its inverse mapping are symmetrically embedded in the latent space of VAE. The decoder of LIA is first trained as a standard GAN with the invertible network and then the partial encoder is learned from a disentangled autoencoder by detaching the invertible network from LIA, thus avoiding the entanglement problem caused by the random latent space. Experiments conducted on the FFHQ face dataset and three LSUN datasets validate the effectiveness of LIA/GAN.