Real world datasets often contain entries with missing elements e.g. in a medical dataset, a patient is unlikely to have taken all possible diagnostic tests. Variational Autoencoders (VAEs) are popular generative models often used for unsupervised learning. Despite their widespread use it is unclear how best to apply VAEs to datasets with missing data. We develop a novel latent variable model of a corruption process which generates missing data, and derive a corresponding tractable evidence lower bound (ELBO). Our model is straightforward to implement, can handle both missing completely at random (MCAR) and missing not at random (MNAR) data, scales to high dimensional inputs and gives both the VAE encoder and decoder principled access to indicator variables for whether a data element is missing or not. On the MNIST and SVHN datasets we demonstrate improved marginal log-likelihood of observed data and better missing data imputation, compared to existing approaches.
The framework of variational autoencoders (VAEs) provides a principled method for jointly learning latent-variable models and corresponding inference models. However, the main drawback of this approach is the blurriness of the generated images. Some studies link this effect to the objective function, namely, the (negative) log-likelihood. Here, we propose to enhance VAEs by adding a random variable that is a downscaled version of the original image and still use the log-likelihood function as the learning objective. Further, by providing the downscaled image as an input to the decoder, it can be used in a manner similar to the super-resolution. We present empirically that the proposed approach performs comparably to VAEs in terms of the negative log-likelihood, but it obtains a better FID score in data synthesis.
Beyond the common difficulties faced in the natural image captioning, medical report generation specifically requires the model to describe a medical image with a fine-grained and semantic-coherence paragraph that should satisfy both medical commonsense and logic. Previous works generally extract the global image features and attempt to generate a paragraph that is similar to referenced reports; however, this approach has two limitations. Firstly, the regions of primary interest to radiologists are usually located in a small area of the global image, meaning that the remainder parts of the image could be considered as irrelevant noise in the training procedure. Secondly, there are many similar sentences used in each medical report to describe the normal regions of the image, which causes serious data bias. This deviation is likely to teach models to generate these inessential sentences on a regular basis. To address these problems, we propose an Auxiliary Signal-Guided Knowledge Encoder-Decoder (ASGK) to mimic radiologists' working patterns. In more detail, ASGK integrates internal visual feature fusion and external medical linguistic information to guide medical knowledge transfer and learning. The core structure of ASGK consists of a medical graph encoder and a natural language decoder, inspired by advanced Generative Pre-Training (GPT). Experiments on the CX-CHR dataset and our COVID-19 CT Report dataset demonstrate that our proposed ASGK is able to generate a robust and accurate report, and moreover outperforms state-of-the-art methods on both medical terminology classification and paragraph generation metrics.
With the success of language pretraining, it is highly desirable to develop more efficient architectures of good scalability that can exploit the abundant unlabeled data at a lower cost. To improve the efficiency, we examine the much-overlooked redundancy in maintaining a full-length token-level presentation, especially for tasks that only require a single-vector presentation of the sequence. With this intuition, we propose Funnel-Transformer which gradually compresses the sequence of hidden states to a shorter one and hence reduces the computation cost. More importantly, by re-investing the saved FLOPs from length reduction in constructing a deeper or wider model, we further improve the model capacity. In addition, to perform token-level predictions as required by common pretraining objectives, Funnel-Transformer is able to recover a deep representation for each token from the reduced hidden sequence via a decoder. Empirically, with comparable or fewer FLOPs, Funnel-Transformer outperforms the standard Transformer on a wide variety of sequence-level prediction tasks, including text classification, language understanding, and reading comprehension. The code and pretrained checkpoints are available at https://github.com/laiguokun/Funnel-Transformer.
The image relighting task of transferring illumination conditions between two images offers an interesting and difficult challenge with potential applications in photography, cinematography and computer graphics. In this report we present methods that we tried to achieve that goal. Our models are trained on a rendered dataset of artificial locations with varied scene content, light source location and color temperature. With this dataset, we used a network with illumination estimation component aiming to infer and replace light conditions in the latent space representation of the concerned scenes.
Acquiring accurate summarization and sentiment from user reviews is an essential component of modern e-commerce platforms. Review summarization aims at generating a concise summary that describes the key opinions and sentiment of a review, while sentiment classification aims to predict a sentiment label indicating the sentiment attitude of a review. To effectively leverage the shared sentiment information in both review summarization and sentiment classification tasks, we propose a novel dual-view model that jointly improves the performance of these two tasks. In our model, an encoder first learns a context representation for the review, then a summary decoder generates a review summary word by word. After that, a source-view sentiment classifier uses the encoded context representation to predict a sentiment label for the review, while a summary-view sentiment classifier uses the decoder hidden states to predict a sentiment label for the generated summary. During training, we introduce an inconsistency loss to penalize the disagreement between these two classifiers. It helps the decoder to generate a summary to have a consistent sentiment tendency with the review and also helps the two sentiment classifiers learn from each other. Experiment results on four real-world datasets from different domains demonstrate the effectiveness of our model.
In this paper we propose a method based on deep learning that detects multiple people from a single overhead depth image with high reliability. Our neural network, called DPDnet, is based on two fully-convolutional encoder-decoder neural blocks based on residual layers. The Main Block takes a depth image as input and generates a pixel-wise confidence map, where each detected person in the image is represented by a Gaussian-like distribution. The refinement block combines the depth image and the output from the main block, to refine the confidence map. Both blocks are simultaneously trained end-to-end using depth images and head position labels. The experimental work shows that DPDNet outperforms state-of-the-art methods, with accuracies greater than 99% in three different publicly available datasets, without retraining not fine-tuning. In addition, the computational complexity of our proposal is independent of the number of people in the scene and runs in real time using conventional GPUs.
Recently, deep learning has become much more popular in computer vision area. The Convolution Neural Network (CNN) has brought a breakthrough in images segmentation areas, especially, for medical images. In this regard, U-Net is the predominant approach to medical image segmentation task. The U-Net not only performs well in segmenting multimodal medical images generally, but also in some tough cases of them. However, we found that the classical U-Net architecture has limitation in several aspects. Therefore, we applied modifications: 1) designed efficient CNN architecture to replace encoder and decoder, 2) applied residual module to replace skip connection between encoder and decoder to improve based on the-state-of-the-art U-Net model. Following these modifications, we designed a novel architecture--DC-UNet, as a potential successor to the U-Net architecture. We created a new effective CNN architecture and build the DC-UNet based on this CNN. We have evaluated our model on three datasets with tough cases and have obtained a relative improvement in performance of 2.90%, 1.49% and 11.42% respectively compared with classical U-Net. In addition, we used the Tanimoto similarity to replace the Jaccard similarity for gray-to-gray image comparisons.
Recently, there has been a strong push to transition from hybrid models to end-to-end (E2E) models for automatic speech recognition. Currently, there are three promising E2E methods: recurrent neural network transducer (RNN-T), RNN attention-based encoder-decoder (AED), and Transformer-AED. In this study, we conduct an empirical comparison of RNN-T, RNN-AED, and Transformer-AED models, in both non-streaming and streaming modes. We use 65 thousand hours of Microsoft anonymized training data to train these models. As E2E models are more data hungry, it is better to compare their effectiveness with large amount of training data. To the best of our knowledge, no such comprehensive study has been conducted yet. We show that although AED models are stronger than RNN-T in the non-streaming mode, RNN-T is very competitive in streaming mode if its encoder can be properly initialized. Among all three E2E models, transformer-AED achieved the best accuracy in both streaming and non-streaming mode. We show that both streaming RNN-T and transformer-AED models can obtain better accuracy than a highly-optimized hybrid model.
We present a `CLAssifier-DECoder' architecture (\emph{ClaDec}) which facilitates the comprehension of the output of an arbitrary layer in a neural network (NN). It uses a decoder to transform the non-interpretable representation of the given layer to a representation that is more similar to the domain a human is familiar with. In an image recognition problem, one can recognize what information is represented by a layer by contrasting reconstructed images of \emph{ClaDec} with those of a conventional auto-encoder(AE) serving as reference. We also extend \emph{ClaDec} to allow the trade-off between human interpretability and fidelity. We evaluate our approach for image classification using Convolutional NNs. We show that reconstructed visualizations using encodings from a classifier capture more relevant information for classification than conventional AEs. Relevant code is available at \url{https://github.com/JohnTailor/ClaDec}
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
We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression procedure or anchor generation that explicitly encode our prior knowledge about the task. The main ingredients of the new framework, called DEtection TRansformer or DETR, are a set-based global loss that forces unique predictions via bipartite matching, and a transformer encoder-decoder architecture. Given a fixed small set of learned object queries, DETR reasons about the relations of the objects and the global image context to directly output the final set of predictions in parallel. The new model is conceptually simple and does not require a specialized library, unlike many other modern detectors. DETR demonstrates accuracy and run-time performance on par with the well-established and highly-optimized Faster RCNN baseline on the challenging COCO object detection dataset. Moreover, DETR can be easily generalized to produce panoptic segmentation in a unified manner. We show that it significantly outperforms competitive baselines. Training code and pretrained models are available at https://github.com/facebookresearch/detr.
Studies on grammatical error correction (GEC) have reported the effectiveness of pretraining a Seq2Seq model with a large amount of pseudodata. However, this approach requires time-consuming pretraining for GEC because of the size of the pseudodata. In this study, we explore the utility of bidirectional and auto-regressive transformers (BART) as a generic pretrained encoder-decoder model for GEC. With the use of this generic pretrained model for GEC, the time-consuming pretraining can be eliminated. We find that monolingual and multilingual BART models achieve high performance in GEC, with one of the results being comparable to the current strong results in English GEC. Our implementations are publicly available at GitHub (https://github.com/Katsumata420/generic-pretrained-GEC).
In this paper, we explore prosody transfer for audiobook generation under rather realistic condition where training DB is plain audio mostly from multiple ordinary people and reference audio given during inference is from professional and richer in prosody than training DB. To be specific, we explore transferring Korean dialects and emotive speech even though training set is mostly composed of standard and neutral Korean. We found that under this setting, original global style token method generates undesirable glitches in pitch, energy and pause length. To deal with this issue, we propose two models, hard and soft pitchtron and release the toolkit and corpus that we have developed. Hard pitchtron uses pitch as input to the decoder while soft pitchtron uses pitch as input to the prosody encoder. We verify the effectiveness of proposed models with objective and subjective tests. AXY score over GST is 2.01 and 1.14 for hard pitchtron and soft pitchtron respectively.
Encoder-decoder models provide a generic architecture for sequence-to-sequence tasks such as speech recognition and translation. While offline systems are often evaluated on quality metrics like word error rates (WER) and BLEU, latency is also a crucial factor in many practical use-cases. We propose three latency reduction techniques for chunk-based incremental inference and evaluate their efficiency in terms of accuracy-latency trade-off. On the 300-hour How2 dataset, we reduce latency by 83% to 0.8 second by sacrificing 1% WER (6% rel.) compared to offline transcription. Although our experiments use the Transformer, the hypothesis selection strategies are applicable to other encoder-decoder models. To avoid expensive re-computation, we use a unidirectionally-attending encoder. After an adaptation procedure to partial sequences, the unidirectional model performs on-par with the original model. We further show that our approach is also applicable to low-latency speech translation. On How2 English-Portuguese speech translation, we reduce latency to 0.7 second (-84% rel.) while incurring a loss of 2.4 BLEU points (5% rel.) compared to the offline system.
Scene text recognition is a hot research topic in computer vision. Recently, many recognition methods based on the encoder-decoder framework have been proposed, and they can handle scene texts of perspective distortion and curve shape. Nevertheless, they still face lots of challenges like image blur, uneven illumination, and incomplete characters. We argue that most encoder-decoder methods are based on local visual features without explicit global semantic information. In this work, we propose a semantics enhanced encoder-decoder framework to robustly recognize low-quality scene texts. The semantic information is used both in the encoder module for supervision and in the decoder module for initializing. In particular, the state-of-the art ASTER method is integrated into the proposed framework as an exemplar. Extensive experiments demonstrate that the proposed framework is more robust for low-quality text images, and achieves state-of-the-art results on several benchmark datasets.
Artificial intelligence can now provide more solutions for different problems, especially in the medical field. One of those problems the lack of answers to any given medical/health-related question. The Internet is full of forums that allow people to ask some specific questions and get great answers for them. Nevertheless, browsing these questions in order to locate one similar to your own, also finding a satisfactory answer is a difficult and time-consuming task. This research will introduce a solution to this problem by automating the process of generating qualified answers to these questions and creating a kind of digital doctor. Furthermore, this research will train an end-to-end model using the framework of RNN and the encoder-decoder to generate sensible and useful answers to a small set of medical/health issues. The proposed model was trained and evaluated using data from various online services, such as WebMD, HealthTap, eHealthForums, and iCliniq.
Fueled by recent advances in machine learning, there has been tremendous progress in the field of semantic segmentation for the medical image computing community. However, developed algorithms are often optimized and validated by hand based on one task only. In combination with small datasets, interpreting the generalizability of the results is often difficult. The Medical Segmentation Decathlon challenge addresses this problem, and aims to facilitate development of generalizable 3D semantic segmentation algorithms that require no manual parametrization. Such an algorithm was developed and is presented in this paper. It consists of a 3D convolutional neural network with encoder-decoder architecture employing residual-connections, skip-connections and multi-level generation of predictions. It works on anisotropic voxel-geometries and has anisotropic depth, i.e., the number of downsampling steps is a task-specific parameter. These depths are automatically inferred for each task prior to training. By combining this flexible architecture with on-the-fly data augmentation and little-to-no pre-- or postprocessing, promising results could be achieved. The code developed for this challenge will be available online after the final deadline at: https://github.com/ORippler/MSD_2018
End-to-end speaker diarization for an unknown number of speakers is addressed in this paper. Recently proposed end-to-end speaker diarization outperformed conventional clustering-based speaker diarization, but it has one drawback: it is less flexible in terms of the number of speakers. This paper proposes a method for encoder-decoder based attractor calculation (EDA), which first generates a flexible number of attractors from a speech embedding sequence. Then, the generated multiple attractors are multiplied by the speech embedding sequence to produce the same number of speaker activities. The speech embedding sequence is extracted using the conventional self-attentive end-to-end neural speaker diarization (SA-EEND) network. In a two-speaker condition, our method achieved a 2.69 % diarization error rate (DER) on simulated mixtures and a 8.07 % DER on the two-speaker subset of CALLHOME, while vanilla SA-EEND attained 4.56 % and 9.54 %, respectively. In unknown numbers of speakers conditions, our method attained a 15.29 % DER on CALLHOME, while the x-vector-based clustering method achieved a 19.43 % DER.
Following the rationale of end-to-end modeling, CTC, RNN-T or encoder-decoder-attention models for automatic speech recognition (ASR) use graphemes or grapheme-based subword units based on e.g. byte-pair encoding (BPE). The mapping from pronunciation to spelling is learned completely from data. In contrast to this, classical approaches to ASR employ secondary knowledge sources in the form of phoneme lists to define phonetic output labels and pronunciation lexica. In this work, we do a systematic comparison between grapheme- and phoneme-based output labels for an encoder-decoder-attention ASR model. We investigate the use of single phonemes as well as BPE-based phoneme groups as output labels of our model. To preserve a simplified and efficient decoder design, we also extend the phoneme set by auxiliary units to be able to distinguish homophones. Experiments performed on the Switchboard 300h and LibriSpeech benchmarks show that phoneme-based modeling is competitive to grapheme-based encoder-decoder-attention modeling.