Breaking down the structure of long texts into semantically coherent segments makes the texts more readable and supports downstream applications like summarization and retrieval. Starting from an apparent link between text coherence and segmentation, we introduce a novel supervised model for text segmentation with simple but explicit coherence modeling. Our model -- a neural architecture consisting of two hierarchically connected Transformer networks -- is a multi-task learning model that couples the sentence-level segmentation objective with the coherence objective that differentiates correct sequences of sentences from corrupt ones. The proposed model, dubbed Coherence-Aware Text Segmentation (CATS), yields state-of-the-art segmentation performance on a collection of benchmark datasets. Furthermore, by coupling CATS with cross-lingual word embeddings, we demonstrate its effectiveness in zero-shot language transfer: it can successfully segment texts in languages unseen in training.
In this paper, we propose a refined multi-stage multi-task training strategy to improve the performance of online attention-based encoder-decoder (AED) models. A three-stage training based on three levels of architectural granularity namely, character encoder, byte pair encoding (BPE) based encoder, and attention decoder, is proposed. Also, multi-task learning based on two-levels of linguistic granularity namely, character and BPE, is used. We explore different pre-training strategies for the encoders including transfer learning from a bidirectional encoder. Our encoder-decoder models with online attention show 35% and 10% relative improvement over their baselines for smaller and bigger models, respectively. Our models achieve a word error rate (WER) of 5.04% and 4.48% on the Librispeech test-clean data for the smaller and bigger models respectively after fusion with long short-term memory (LSTM) based external language model (LM).
Multi-task learning is a powerful method for solving multiple correlated tasks simultaneously. However, it is often impossible to find one single solution to optimize all the tasks, since different tasks might conflict with each other. Recently, a novel method is proposed to find one single Pareto optimal solution with good trade-off among different tasks by casting multi-task learning as multiobjective optimization. In this paper, we generalize this idea and propose a novel Pareto multi-task learning algorithm (Pareto MTL) to find a set of well-distributed Pareto solutions which can represent different trade-offs among different tasks. The proposed algorithm first formulates a multi-task learning problem as a multiobjective optimization problem, and then decomposes the multiobjective optimization problem into a set of constrained subproblems with different trade-off preferences. By solving these subproblems in parallel, Pareto MTL can find a set of well-representative Pareto optimal solutions with different trade-off among all tasks. Practitioners can easily select their preferred solution from these Pareto solutions, or use different trade-off solutions for different situations. Experimental results confirm that the proposed algorithm can generate well-representative solutions and outperform some state-of-the-art algorithms on many multi-task learning applications.
Automated Parking is a low speed manoeuvring scenario which is quite unstructured and complex, requiring full 360{\deg} near-field sensing around the vehicle. In this paper, we discuss the design and implementation of an automated parking system from the perspective of camera based deep learning algorithms. We provide a holistic overview of an industrial system covering the embedded system, use cases and the deep learning architecture. We demonstrate a real-time multi-task deep learning network called FisheyeMultiNet, which detects all the necessary objects for parking on a low-power embedded system. FisheyeMultiNet runs at 15 fps for 4 cameras and it has three tasks namely object detection, semantic segmentation and soiling detection. To encourage further research, we release a partial dataset of 5,000 images containing semantic segmentation and bounding box detection ground truth via WoodScape project \cite{yogamani2019woodscape}.
In this paper, we focus on graph representation learning of heterogeneous information network (HIN), in which various types of vertices are connected by various types of relations. Most of the existing methods conducted on HIN revise homogeneous graph embedding models via meta-paths to learn low-dimensional vector space of HIN. In this paper, we propose a novel Heterogeneous Graph Structural Attention Neural Network (HetSANN) to directly encode structural information of HIN without meta-path and achieve more informative representations. With this method, domain experts will not be needed to design meta-path schemes and the heterogeneous information can be processed automatically by our proposed model. Specifically, we implicitly represent heterogeneous information using the following two methods: 1) we model the transformation between heterogeneous vertices through a projection in low-dimensional entity spaces; 2) afterwards, we apply the graph neural network to aggregate multi-relational information of projected neighborhood by means of attention mechanism. We also present three extensions of HetSANN, i.e., voices-sharing product attention for the pairwise relationships in HIN, cycle-consistency loss to retain the transformation between heterogeneous entity spaces, and multi-task learning with full use of information. The experiments conducted on three public datasets demonstrate that our proposed models achieve significant and consistent improvements compared to state-of-the-art solutions.
Current face detectors utilize anchors to frame a multi-task learning problem which combines classification and bounding box regression. Effective anchor design and anchor matching strategy enable face detectors to localize faces under large pose and scale variations. However, we observe that more than 80% correctly predicted bounding boxes are regressed from the unmatched anchors (the IoUs between anchors and target faces are lower than a threshold) in the inference phase. It indicates that these unmatched anchors perform excellent regression ability, but the existing methods neglect to learn from them. In this paper, we propose an Online High-quality Anchor Mining Strategy (HAMBox), which explicitly helps outer faces compensate with high-quality anchors. Our proposed HAMBox method could be a general strategy for anchor-based single-stage face detection. Experiments on various datasets, including WIDER FACE, FDDB, AFW and PASCAL Face, demonstrate the superiority of the proposed method. Furthermore, our team win the championship on the Face Detection test track of WIDER Face and Pedestrian Challenge 2019. We will release the codes with PaddlePaddle.
Despite great success in human parsing, progress for parsing other deformable articulated objects, like animals, is still limited by the lack of labeled data. In this paper, we use synthetic images and ground truth generated from CAD animal models to address this challenge. To bridge the domain gap between real and synthetic images, we propose a novel consistency-constrained semi-supervised learning method (CC-SSL). Our method leverages both spatial and temporal consistencies, to bootstrap weak models trained on synthetic data with unlabeled real images. We demonstrate the effectiveness of our method on highly deformable animals, such as horses and tigers. Without using any real image label, our method allows for accurate keypoint prediction on real images. Moreover, we quantitatively show that models using synthetic data achieve better generalization performance than models trained on real images across different domains in the Visual Domain Adaptation Challenge dataset. Our synthetic dataset contains 10+ animals with diverse poses and rich ground truth, which enables us to use the multi-task learning strategy to further boost models' performance.
Recently, neuroimaging-based Alzheimer’s disease (AD) or mild cognitive impairment (MCI) diagnosis has attracted researchers in the field, due to the increasing prevalence of the diseases. Unfortunately, the unfavorable high-dimensional nature of neuroimaging data, but a limited small number of samples available, makes it challenging to build a robust computer-aided diagnosis system. Machine learning techniques have been considered as a useful tool in this respect and, among various methods, sparse regression has shown its validity in the literature. However, to our best knowledge, the existing sparse regression methods mostly try to select features based on the optimal regression coefficients in one step. We argue that since the training feature vectors are composed of both informative and uninformative or less informative features, the resulting optimal regression coefficients are inevidently affected by the uninformative or less informative features. To this end, we first propose a novel deep architecture to recursively discard uninformative features by performing sparse multi-task learning in a hierarchical fashion. We further hypothesize that the optimal regression coefficients reflect the relative importance of features in representing the target response variables. In this regard, we use the optimal regression coefficients learned in one hierarchy as feature weighting factors in the following hierarchy, and formulate a weighted sparse multi-task learning method. Lastly, we also take into account the distributional characteristics of samples per class and use clustering-induced subclass label vectors as target response values in our sparse regression model. In our experiments on the ADNI cohort, we performed both binary and multi-class classification tasks in AD/MCI diagnosis and showed the superiority of the proposed method by comparing with the state-of-the-art methods.
Aspect-based sentiment analysis (ABSA) task is a multi-grained task of natural language processing and consists of two subtasks: aspect term extraction (ATE) and aspect polarity classification (APC). Most of the existing work focuses on the subtask of aspect term polarity inferring and ignores the significance of aspect term extraction. Besides, the existing researches do not pay attention to the research of the Chinese-oriented ABSA task. Based on the local context focus (LCF) mechanism, this paper firstly proposes a multi-task learning model for Chinese-oriented aspect-based sentiment analysis, namely LCF-ATEPC. Compared with existing models, this model equips the capability of extracting aspect term and inferring aspect term polarity synchronously, moreover, this model is effective to analyze both Chinese and English comments simultaneously and the experiment on a multilingual mixed dataset proved its availability. By integrating the domain-adapted BERT model, the LCF-ATEPC model achieved the state-of-the-art performance of aspect term extraction and aspect polarity classification in four Chinese review datasets. Besides, the experimental results on the most commonly used SemEval-2014 task4 Restaurant and Laptop datasets outperform the state-of-the-art performance on the ATE and APC subtask.
Over the past few years many research efforts have been devoted to the field of affect analysis. Various approaches have been proposed for: i) discrete emotion recognition in terms of the primary facial expressions; ii) emotion analysis in terms of facial Action Units (AUs), assuming a fixed expression intensity; iii) dimensional emotion analysis, in terms of valence and arousal (VA). These approaches can only be effective, if they are developed using large, appropriately annotated databases, showing behaviors of people in-the-wild, i.e., in uncontrolled environments. Aff-Wild has been the first, large-scale, in-the-wild database (including around 1,200,000 frames of 300 videos), annotated in terms of VA. In the vast majority of existing emotion databases, their annotation is limited to either primary expressions, or valence-arousal, or action units. In this paper, we first annotate a part (around $234,000$ frames) of the Aff-Wild database in terms of $8$ AUs and another part (around $288,000$ frames) in terms of the $7$ basic emotion categories, so that parts of this database are annotated in terms of VA, as well as AUs, or primary expressions. Then, we set up and tackle multi-task learning for emotion recognition, as well as for facial image generation. Multi-task learning is performed using: i) a deep neural network with shared hidden layers, which learns emotional attributes by exploiting their inter-dependencies; ii) a discriminator of a generative adversarial network (GAN). On the other hand, image generation is implemented through the generator of the GAN. For these two tasks, we carefully design loss functions that fit the examined set-up. Experiments are presented which illustrate the good performance of the proposed approach when applied to the new annotated parts of the Aff-Wild database.
Speech-to-text translation (ST), which translates source language speech into target language text, has attracted intensive attention in recent years. Compared to the traditional pipeline system, the end-to-end ST model has potential benefits of lower latency, smaller model size, and less error propagation. However, it is notoriously difficult to implement such a model without transcriptions as intermediate. Existing works generally apply multi-task learning to improve translation quality by jointly training end-to-end ST along with automatic speech recognition (ASR). However, different tasks in this method cannot utilize information from each other, which limits the improvement. Other works propose a two-stage model where the second model can use the hidden state from the first one, but its cascade manner greatly affects the efficiency of training and inference process. In this paper, we propose a novel interactive attention mechanism which enables ASR and ST to perform synchronously and interactively in a single model. Specifically, the generation of transcriptions and translations not only relies on its previous outputs but also the outputs predicted in the other task. Experiments on TED speech translation corpora have shown that our proposed model can outperform strong baselines on the quality of speech translation and achieve better speech recognition performances as well.
Recently, multi-task networks have shown to both offer additional estimation capabilities, and, perhaps more importantly, increased performance over single-task networks on a "main/primary" task. However, balancing the optimization criteria of multi-task networks across different tasks is an area of active exploration. Here, we extend a previously proposed 3D attention-based network with four additional multi-task subnetworks for the detection of lung cancer and four auxiliary tasks (diagnosis of asthma, chronic bronchitis, chronic obstructive pulmonary disease, and emphysema). We introduce and evaluate a learning policy, Periodic Focusing Learning Policy (PFLP), that alternates the dominance of tasks throughout the training. To improve performance on the primary task, we propose an Internal-Transfer Weighting (ITW) strategy to suppress the loss functions on auxiliary tasks for the final stages of training. To evaluate this approach, we examined 3386 patients (single scan per patient) from the National Lung Screening Trial (NLST) and de-identified data from the Vanderbilt Lung Screening Program, with a 2517/277/592 (scans) split for training, validation, and testing. Baseline networks include a single-task strategy and a multi-task strategy without adaptive weights (PFLP/ITW), while primary experiments are multi-task trials with either PFLP or ITW or both. On the test set for lung cancer prediction, the baseline single-task network achieved prediction AUC of 0.8080 and the multi-task baseline failed to converge (AUC 0.6720). However, applying PFLP helped multi-task network clarify and achieved test set lung cancer prediction AUC of 0.8402. Furthermore, our ITW technique boosted the PFLP enabled multi-task network and achieved an AUC of 0.8462 (McNemar test, p < 0.01).
Deep neural networks are a promising approach towards multi-task learning because of their capability to leverage knowledge across domains and learn general purpose representations. Nevertheless, they can fail to live up to these promises as tasks often compete for a model's limited resources, potentially leading to lower overall performance. In this work we tackle the issue of interfering tasks through a comprehensive analysis of their training, derived from looking at the interaction between gradients within their shared parameters. Our empirical results show that well-performing models have low variance in the angles between task gradients and that popular regularization methods implicitly reduce this measure. Based on this observation, we propose a novel gradient regularization term that minimizes task interference by enforcing near orthogonal gradients. Updating the shared parameters using this property encourages task specific decoders to optimize different parts of the feature extractor, thus reducing competition. We evaluate our method with classification and regression tasks on the multiDigitMNIST, NYUv2 and SUN RGB-D datasets where we obtain competitive results.
Word embeddings are substantially successful in capturing semantic relations among words. However, these lexical semantics are difficult to be interpreted. Definition modeling provides a more intuitive way to evaluate embeddings by utilizing them to generate natural language definitions of corresponding words. This task is of great significance for practical application and in-depth understanding of word representations. We propose a novel framework for definition modeling, which can generate reasonable and understandable context-dependent definitions. Moreover, we introduce usage modeling and study whether it is possible to utilize embeddings to generate example sentences of words. These ways are a more direct and explicit expression of embedding's semantics for better interpretability. We extend the single task model to multi-task setting and investigate several joint multi-task models to combine usage modeling and definition modeling together. Experimental results on existing Oxford dataset and a new collected Oxford-2019 dataset show that our single-task model achieves the state-of-the-art result in definition modeling and the multi-task learning methods are helpful for two tasks to improve the performance.
Calcaneus is the largest tarsal bone to withstand the daily stresses of weight-bearing. The calcaneal fracture is the most common type in the tarsal bone fractures. After a fracture is suspected, plain radiographs should be taken first. Bohler's Angle (BA) and Critical Angle of Gissane (CAG), measured by four anatomic landmarks in lateral foot radiograph, can guide fracture diagnosis and facilitate operative recovery of the fractured calcaneus. This study aims to develop an analysis system that can automatically locate four anatomic landmarks, measure BA and CAG for fracture assessment, identify fractured calcaneus, and segment fractured regions. For landmark detection, we proposed a coarse-to-fine Rotation-Invariant Regression-Voting (RIRV) landmark detection method based on regressive Multi-Layer Perceptron (MLP) and Scale Invariant Feature Transform (SIFT) patch descriptor, which solves the problem of fickle rotation of calcaneus. By implementing a novel normalization approach, the RIRV method is explicitly rotation-invariance comparing with traditional regressive methods. For fracture identification and segmentation, a convolution neural network (CNN) based on U-Net with auxiliary classification head (U-Net-CH) is designed. The input ROIs of the CNN are normalized by detected landmarks to uniform view, orientation, and scale. The advantage of this approach is the multi-task learning that combines classification and segmentation. Our system can accurately measure BA and CAG with a mean angle error of 3.8 and 6.2 respectively. For fracture identification and fracture region segmentation, our system presents good performance with an F1-score of 96.55%, recall of 94.99%, and segmentation IoU-score of 0.586.
Spatiotemporal representations learned using 3D convolutional neural networks (CNN) are currently used in state-of-the-art approaches for action related tasks. However, 3D-CNN are notorious for being memory and compute resource intensive as compared with more simple 2D-CNN architectures. We propose to hallucinate spatiotemporal representations from a 3D-CNN teacher with a 2D-CNN student. By requiring the 2D-CNN to predict the future and intuit upcoming activity, it is encouraged to gain a deeper understanding of actions and how they evolve. The hallucination task is treated as an auxiliary task, which can be used with any other action related task in a multitask learning setting. Thorough experimental evaluation shows that the hallucination task indeed helps improve performance on action recognition, action quality assessment, and dynamic scene recognition tasks. From a practical standpoint, being able to hallucinate spatiotemporal representations without an actual 3D-CNN can enable deployment in resource-constrained scenarios, such as with limited computing power and/or lower bandwidth. Codebase is available here: https://github.com/ParitoshParmar/HalluciNet.
The quality control of fetal sonographic (FS) images is essential for the correct biometric measurements and fetal anomaly diagnosis. However, quality control requires professional sonographers to perform and is often labor-intensive. To solve this problem, we propose an automatic image quality assessment scheme based on multi-task learning to assist in FS image quality control. An essential criterion for FS image quality control is that all the essential anatomical structures in the section should appear full and remarkable with a clear boundary. Therefore, our scheme aims to identify those essential anatomical structures to judge whether an FS image is the standard image, which is achieved by three convolutional neural networks. The Feature Extraction Network aims to extract deep level features of FS images. Based on the extracted features, the Class Prediction Network determines whether the structure meets the standard and Region Proposal Network identifies its position. The scheme has been applied to three types of fetal sections, which are the head, abdominal, and heart. The experimental results show that our method can make a quality assessment of an FS image within less a second. Also, our method achieves competitive performance in both the detection and classification compared with state-of-the-art methods.
The release of large natural language inference (NLI) datasets like SNLI and MNLI have led to rapid development and improvement of completely neural systems for the task. Most recently, heavily pre-trained, Transformer-based models like BERT and MT-DNN have reached near-human performance on these datasets. However, these standard datasets have been shown to contain many annotation artifacts, allowing models to shortcut understanding using simple fallible heuristics, and still perform well on the test set. So it is no surprise that many adversarial (challenge) datasets have been created that cause models trained on standard datasets to fail dramatically. Although extra training on this data generally improves model performance on just that type of data, transferring that learning to unseen examples is still partial at best. This work evaluates the failures of state-of-the-art models on existing adversarial datasets that test different linguistic phenomena, and find that even though the models perform similarly on MNLI, they differ greatly in their robustness to these attacks. In particular, we find syntax-related attacks to be particularly effective across all models, so we provide a fine-grained analysis and comparison of model performance on those examples. We draw conclusions about the value of model size and multi-task learning (beyond comparing their standard test set performance), and provide suggestions for more effective training data.
A considerable part of the success experienced by Voice-controlled virtual assistants (VVA) is due to the emotional and personalized experience they deliver, with humor being a key component in providing an engaging interaction. In this paper we describe methods used to improve the joke skill of a VVA through personalization. The first method, based on traditional NLP techniques, is robust and scalable. The others combine self-attentional network and multi-task learning to obtain better results, at the cost of added complexity. A significant challenge facing these systems is the lack of explicit user feedback needed to provide labels for the models. Instead, we explore the use of two implicit feedback-based labelling strategies. All models were evaluated on real production data. Online results show that models trained on any of the considered labels outperform a heuristic method, presenting a positive real-world impact on user satisfaction. Offline results suggest that the deep-learning approaches can improve the joke experience with respect to the other considered methods.
In most machine learning algorithms, training data is assumed to be independent and identically distributed (iid). When it is not the case, the algorithm's performances are challenged, leading to the famous phenomenon of catastrophic forgetting. Algorithms dealing with it are gathered in the Continual Learning research field. In this paper, we study the regularization based approaches to continual learning and show that those approaches can not learn to discriminate classes from different tasks in an elemental continual benchmark: the class-incremental scenario. We make theoretical reasoning to prove this shortcoming and illustrate it with examples and experiments. Moreover, we show that it can have some important consequences on continual multi-tasks reinforcement learning or in pre-trained models used for continual learning. We believe that highlighting and understanding the shortcomings of regularization strategies will help us to use them more efficiently.