Colorectal polyps are important precursors to colon cancer, the third most common cause of cancer mortality for both men and women. It is a disease where early detection is of crucial importance. Colonoscopy is commonly used for early detection of cancer and precancerous pathology. It is a demanding procedure requiring significant amount of time from specialized physicians and nurses, in addition to a significant miss-rates of polyps by specialists. Automated polyp detection in colonoscopy videos has been demonstrated to be a promising way to handle this problem. {However, polyps detection is a challenging problem due to the availability of limited amount of training data and large appearance variations of polyps. To handle this problem, we propose a novel deep learning method Y-Net that consists of two encoder networks with a decoder network. Our proposed Y-Net method} relies on efficient use of pre-trained and un-trained models with novel sum-skip-concatenation operations. Each of the encoders are trained with encoder specific learning rate along the decoder. Compared with the previous methods employing hand-crafted features or 2-D/3-D convolutional neural network, our approach outperforms state-of-the-art methods for polyp detection with 7.3% F1-score and 13% recall improvement.
Describing visual data into natural language is a very challenging task, at the intersection of computer vision, natural language processing and machine learning. Language goes well beyond the description of physical objects and their interactions and can convey the same abstract idea in many ways. It is both about content at the highest semantic level as well as about fluent form. Here we propose an approach to describe videos in natural language by reaching a consensus among multiple encoder-decoder networks. Finding such a consensual linguistic description, which shares common properties with a larger group, has a better chance to convey the correct meaning. We propose and train several network architectures and use different types of image, audio and video features. Each model produces its own description of the input video and the best one is chosen through an efficient, two-phase consensus process. We demonstrate the strength of our approach by obtaining state of the art results on the challenging MSR-VTT dataset.
Generative models with an encoding component such as autoencoders currently receive great interest. However, training of autoencoders is typically complicated by the need to train a separate encoder and decoder model that have to be enforced to be reciprocal to each other. To overcome this problem, by-design reversible neural networks (RevNets) had been previously used as generative models either directly optimizing the likelihood of the data under the model or using an adversarial approach on the generated data. Here, we instead investigate their performance using an adversary on the latent space in the adversarial autoencoder framework. We investigate the generative performance of RevNets on the CelebA dataset, showing that generative RevNets can generate coherent faces with similar quality as Variational Autoencoders. This first attempt to use RevNets inside the adversarial autoencoder framework slightly underperformed relative to recent advanced generative models using an autoencoder component on CelebA, but this gap may diminish with further optimization of the training setup of generative RevNets. In addition to the experiments on CelebA, we show a proof-of-principle experiment on the MNIST dataset suggesting that adversary-free trained RevNets can discover meaningful latent dimensions without pre-specifying the number of dimensions of the latent sampling distribution. In summary, this study shows that RevNets can be employed in different generative training settings. Source code for this study is at https://github.com/robintibor/generative-reversible
The Variational Autoencoder (VAE) has proven to be an effective model for producing semantically meaningful latent representations for natural data. However, it has thus far seen limited application to sequential data, and, as we demonstrate, existing recurrent VAE models have difficulty modeling sequences with long-term structure. To address this issue, we propose the use of a hierarchical decoder, which first outputs embeddings for subsequences of the input and then uses these embeddings to generate each subsequence independently. This structure encourages the model to utilize its latent code, thereby avoiding the "posterior collapse" problem, which remains an issue for recurrent VAEs. We apply this architecture to modeling sequences of musical notes and find that it exhibits dramatically better sampling, interpolation, and reconstruction performance than a "flat" baseline model. An implementation of our "MusicVAE" is available online at http://g.co/magenta/musicvae-code.
Semantic segmentation is in-demand in satellite imagery processing. Because of the complex environment, automatic categorization and segmentation of land cover is a challenging problem. Solving it can help to overcome many obstacles in urban planning, environmental engineering or natural landscape monitoring. In this paper, we propose an approach for automatic multi-class land segmentation based on a fully convolutional neural network of feature pyramid network (FPN) family. This network is consisted of pre-trained on ImageNet Resnet50 encoder and neatly developed decoder. Based on validation results, leaderboard score and our own experience this network shows reliable results for the DEEPGLOBE - CVPR 2018 land cover classification sub-challenge. Moreover, this network moderately uses memory that allows using GTX 1080 or 1080 TI video cards to perform whole training and makes pretty fast predictions.
We show that it is possible to learn meaningful representations of surgical motion, without supervision, by learning to predict the future. An architecture that combines an RNN encoder-decoder and mixture density networks (MDNs) is developed to model the conditional distribution over future motion given past motion. We show that the learned encodings naturally cluster according to high-level activities, and we demonstrate the usefulness of these learned encodings in the context of information retrieval, where a database of surgical motion is searched for suturing activity using a motion-based query. Future prediction with MDNs is found to significantly outperform simpler baselines as well as the best previously-published result for this task, advancing state-of-the-art performance from an F1 score of 0.60 +- 0.14 to 0.77 +- 0.05.
In this paper, we propose a novel deep neural network architecture, Speech2Vec, for learning fixed-length vector representations of audio segments excised from a speech corpus, where the vectors contain semantic information pertaining to the underlying spoken words, and are close to other vectors in the embedding space if their corresponding underlying spoken words are semantically similar. The proposed model can be viewed as a speech version of Word2Vec. Its design is based on a RNN Encoder-Decoder framework, and borrows the methodology of skipgrams or continuous bag-of-words for training. Learning word embeddings directly from speech enables Speech2Vec to make use of the semantic information carried by speech that does not exist in plain text. The learned word embeddings are evaluated and analyzed on 13 widely used word similarity benchmarks, and outperform word embeddings learned by Word2Vec from the transcriptions.
Machine translation is going through a radical revolution, driven by the explosive development of deep learning techniques using Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). In this paper, we consider a special case in machine translation problems, targeting to convert natural language into Structured Query Language (SQL) for data retrieval over relational database. Although generic CNN and RNN learn the grammar structure of SQL when trained with sufficient samples, the accuracy and training efficiency of the model could be dramatically improved, when the translation model is deeply integrated with the grammar rules of SQL. We present a new encoder-decoder framework, with a suite of new approaches, including new semantic features fed into the encoder, grammar-aware states injected into the memory of decoder, as well as recursive state management for sub-queries. These techniques help the neural network better focus on understanding semantics of operations in natural language and save the efforts on SQL grammar learning. The empirical evaluation on real world database and queries show that our approach outperform state-of-the-art solution by a significant margin.
Object occlusion boundary detection is a fundamental and crucial research problem in computer vision. This is challenging to solve as encountering the extreme boundary/non-boundary class imbalance during training an object occlusion boundary detector. In this paper, we propose to address this class imbalance by up-weighting the loss contribution of false negative and false positive examples with our novel Attention Loss function. We also propose a unified end-to-end multi-task deep object occlusion boundary detection network (DOOBNet) by sharing convolutional features to simultaneously predict object boundary and occlusion orientation. DOOBNet adopts an encoder-decoder structure with skip connection in order to automatically learn multi-scale and multi-level features. We significantly surpass the state-of-the-art on the PIOD dataset (ODS F-score of .702) and the BSDS ownership dataset (ODS F-score of .555), as well as improving the detecting speed to as 0.037s per image on the PIOD dataset.
We address the problem of simultaneous translation by modifying the Neural MT decoder to operate with dynamically built encoder and attention. We propose a tunable agent which decides the best segmentation strategy for a user-defined BLEU loss and Average Proportion (AP) constraint. Our agent outperforms previously proposed Wait-if-diff and Wait-if-worse agents (Cho and Esipova, 2016) on BLEU with a lower latency. Secondly we proposed data-driven changes to Neural MT training to better match the incremental decoding framework.
In neural abstractive summarization, the conventional sequence-to-sequence (seq2seq) model often suffers from repetition and semantic irrelevance. To tackle the problem, we propose a global encoding framework, which controls the information flow from the encoder to the decoder based on the global information of the source context. It consists of a convolutional gated unit to perform global encoding to improve the representations of the source-side information. Evaluations on the LCSTS and the English Gigaword both demonstrate that our model outperforms the baseline models, and the analysis shows that our model is capable of reducing repetition.
We describe initial work on an extension of the Kaldi toolkit that supports weighted finite-state transducer (WFST) decoding on Graphics Processing Units (GPUs). We implement token recombination as an atomic GPU operation in order to fully parallelize the Viterbi beam search, and propose a dynamic load balancing strategy for more efficient token passing scheduling among GPU threads. We also redesign the exact lattice generation and lattice pruning algorithms for better utilization of the GPUs. Experiments on the Switchboard corpus show that the proposed method achieves identical 1-best results and lattice quality in recognition and confidence measure tasks, while running 3 to 15 times faster than the single process Kaldi decoder. The above results are reported on different GPU architectures. Additionally we obtain a 46-fold speedup with sequence parallelism and multi-process service (MPS) in GPU.
Neural language models (NLMs) have recently gained a renewed interest by achieving state-of-the-art performance across many natural language processing (NLP) tasks. However, NLMs are very computationally demanding largely due to the computational cost of the softmax layer over a large vocabulary. We observe that, in decoding of many NLP tasks, only the probabilities of the top-K hypotheses need to be calculated preciously and K is often much smaller than the vocabulary size. This paper proposes a novel softmax layer approximation algorithm, called Fast Graph Decoder (FGD), which quickly identifies, for a given context, a set of K words that are most likely to occur according to a NLM. We demonstrate that FGD reduces the decoding time by an order of magnitude while attaining close to the full softmax baseline accuracy on neural machine translation and language modeling tasks. We also prove the theoretical guarantee on the softmax approximation quality.
We propose an adversarial learning approach for generating multi-turn dialogue responses. Our proposed framework, hredGAN, is based on conditional generative adversarial networks (GANs). The GAN's generator is a modified hierarchical recurrent encoder-decoder network (HRED) and the discriminator is a word-level bidirectional RNN that shares context and word embeddings with the generator. During inference, noise samples conditioned on the dialogue history are used to perturb the generator's latent space to generate several possible responses. The final response is the one ranked best by the discriminator. The hredGAN shows improved performance over existing methods: (1) it generalizes better than networks trained using only the log-likelihood criterion, and (2) it generates longer, more informative and more diverse responses with high utterance and topic relevance even with limited training data. This improvement is demonstrated on the Movie triples and Ubuntu dialogue datasets using both automatic and human evaluations.
We present a generative framework for generalized zero-shot learning where the training and test classes are not necessarily disjoint. Built upon a variational autoencoder based architecture, consisting of a probabilistic encoder and a probabilistic conditional decoder, our model can generate novel exemplars from seen/unseen classes, given their respective class attributes. These exemplars can subsequently be used to train any off-the-shelf classification model. One of the key aspects of our encoder-decoder architecture is a feedback-driven mechanism in which a discriminator (a multivariate regressor) learns to map the generated exemplars to the corresponding class attribute vectors, leading to an improved generator. Our model's ability to generate and leverage examples from unseen classes to train the classification model naturally helps to mitigate the bias towards predicting seen classes in generalized zero-shot learning settings. Through a comprehensive set of experiments, we show that our model outperforms several state-of-the-art methods, on several benchmark datasets, for both standard as well as generalized zero-shot learning.
Q-Ensembles are a model-free approach where input images are fed into different Q-networks and exploration is driven by the assumption that uncertainty is proportional to the variance of the output Q-values obtained. They have been shown to perform relatively well compared to other exploration strategies. Further, model-based approaches, such as encoder-decoder models have been used successfully for next frame prediction given previous frames. This paper proposes to integrate the model-free Q-ensembles and model-based approaches with the hope of compounding the benefits of both and achieving superior exploration as a result. Results show that a model-based trajectory memory approach when combined with Q-ensembles produces superior performance when compared to only using Q-ensembles.
Image style transfer has drawn broad attention in recent years. However, most existing methods aim to explicitly model the transformation between different styles, and the learned model is thus not generalizable to new styles. We here propose a unified style transfer framework for both character typeface transfer and neural style transfer tasks leveraging style and content separation. A key merit of such framework is its generalizability to new styles and contents. The overall framework consists of style encoder, content encoder, mixer and decoder. The style encoder and content encoder are used to extract the style and content representations from the corresponding reference images. The mixer integrates the above two representations and feeds it into the decoder to generate images with the target style and content. During training, the encoder networks learn to extract styles and contents from limited size of style/content reference images. This learning framework allows simultaneous style transfer among multiple styles and can be deemed as a special `multi-task' learning scenario. The encoders are expected to capture the underlying features for different styles and contents which is generalizable to new styles and contents. Under this framework, we design two individual networks for character typeface transfer and neural style transfer, respectively. For character typeface transfer, to separate the style features and content features, we leverage the conditional dependence of styles and contents given an image. For neural style transfer, we leverage the statistical information of feature maps in certain layers to represent style. Extensive experimental results have demonstrated the effectiveness and robustness of the proposed methods.
We introduce a novel self-supervised learning method based on adversarial training. Our objective is to train a discriminator network to distinguish real images from images with synthetic artifacts, and then to extract features from its intermediate layers that can be transferred to other data domains and tasks. To generate images with artifacts, we pre-train a high-capacity autoencoder and then we use a damage and repair strategy: First, we freeze the autoencoder and damage the output of the encoder by randomly dropping its entries. Second, we augment the decoder with a repair network, and train it in an adversarial manner against the discriminator. The repair network helps generate more realistic images by inpainting the dropped feature entries. To make the discriminator focus on the artifacts, we also make it predict what entries in the feature were dropped. We demonstrate experimentally that features learned by creating and spotting artifacts achieve state of the art performance in several benchmarks.
Encoder-decoder based Sequence to Sequence learning (S2S) has made remarkable progress in recent years. Different network architectures have been used in the encoder/decoder. Among them, Convolutional Neural Networks (CNN) and Self Attention Networks (SAN) are the prominent ones. The two architectures achieve similar performances but use very different ways to encode and decode context: CNN use convolutional layers to focus on the local connectivity of the sequence, while SAN uses self-attention layers to focus on global semantics. In this work we propose Double Path Networks for Sequence to Sequence learning (DPN-S2S), which leverage the advantages of both models by using double path information fusion. During the encoding step, we develop a double path architecture to maintain the information coming from different paths with convolutional layers and self-attention layers separately. To effectively use the encoded context, we develop a cross attention module with gating and use it to automatically pick up the information needed during the decoding step. By deeply integrating the two paths with cross attention, both types of information are combined and well exploited. Experiments show that our proposed method can significantly improve the performance of sequence to sequence learning over state-of-the-art systems.
Much of recent research has been devoted to video prediction and generation, yet most of the previous works have demonstrated only limited success in generating videos on short-term horizons. The hierarchical video prediction method by Villegas et al. (2017) is an example of a state-of-the-art method for long-term video prediction, but their method is limited because it requires ground truth annotation of high-level structures (e.g., human joint landmarks) at training time. Our network encodes the input frame, predicts a high-level encoding into the future, and then a decoder with access to the first frame produces the predicted image from the predicted encoding. The decoder also produces a mask that outlines the predicted foreground object (e.g., person) as a by-product. Unlike Villegas et al. (2017), we develop a novel training method that jointly trains the encoder, the predictor, and the decoder together without highlevel supervision; we further improve upon this by using an adversarial loss in the feature space to train the predictor. Our method can predict about 20 seconds into the future and provides better results compared to Denton and Fergus (2018) and Finn et al. (2016) on the Human 3.6M dataset.