The system we used for Task 6 (Automated Audio Captioning)of the Detection and Classification of Acoustic Scenes and Events(DCASE) 2020 Challenge combines three elements, namely, dataaugmentation, multi-task learning, and post-processing, for audiocaptioning. The system received the highest evaluation scores, butwhich of the individual elements most fully contributed to its perfor-mance has not yet been clarified. Here, to asses their contributions,we first conducted an element-wise ablation study on our systemto estimate to what extent each element is effective. We then con-ducted a detailed module-wise ablation study to further clarify thekey processing modules for improving accuracy. The results showthat data augmentation and post-processing significantly improvethe score in our system. In particular, mix-up data augmentationand beam search in post-processing improve SPIDEr by 0.8 and 1.6points, respectively.
In this work, we develop a novel fairness learning approach for multi-task regression models based on a biased training dataset, using a popular rank-based non-parametric independence test, i.e., Mann Whitney U statistic, for measuring the dependency between target variable and protected variables. To solve this learning problem efficiently, we first reformulate the problem as a new non-convex optimization problem, in which a non-convex constraint is defined based on group-wise ranking functions of individual objects. We then develop an efficient model-training algorithm based on the framework of non-convex alternating direction method of multipliers (NC-ADMM), in which one of the main challenges is to implement an efficient projection oracle to the preceding non-convex set defined based on ranking functions. Through the extensive experiments on both synthetic and real-world datasets, we validated the out-performance of our new approach against several state-of-the-art competitive methods on several popular metrics relevant to fairness learning.
Obtaining annotations for 3D medical images is expensive and time-consuming, despite its importance for automating segmentation tasks. Although multi-task learning is considered an effective method for training segmentation models using small amounts of annotated data, a systematic understanding of various subtasks is still lacking. In this study, we propose a multi-task segmentation model with a contrastive learning based subtask and compare its performance with other multi-task models, varying the number of labeled data for training. We further extend our model so that it can utilize unlabeled data through the regularization branch in a semi-supervised manner. We experimentally show that our proposed method outperforms other multi-task methods including the state-of-the-art fully supervised model when the amount of annotated data is limited.
Multi-task learning (MTL) is a subfield of machine learning in which multiple tasks are simultaneously learned by a shared model. Such approaches offer advantages like improved data efficiency, reduced overfitting through shared representations, and fast learning by leveraging auxiliary information. However, the simultaneous learning of multiple tasks presents new design and optimization challenges, and choosing which tasks should be learned jointly is in itself a non-trivial problem. In this survey, we give an overview of multi-task learning methods for deep neural networks, with the aim of summarizing both the well-established and most recent directions within the field. Our discussion is structured according to a partition of the existing deep MTL techniques into three groups: architectures, optimization methods, and task relationship learning. We also provide a summary of common multi-task benchmarks.
Aggressive language detection (ALD), detecting the abusive and offensive language in texts, is one of the crucial applications in NLP community. Most existing works treat ALD as regular classification with neural models, while ignoring the inherent conflicts of social media text that they are quite unnormalized and irregular. In this work, we target improving the ALD by jointly performing text normalization (TN), via an adversarial multi-task learning framework. The private encoders for ALD and TN focus on the task-specific features retrieving, respectively, and the shared encoder learns the underlying common features over two tasks. During adversarial training, a task discriminator distinguishes the separate learning of ALD or TN. Experimental results on four ALD datasets show that our model outperforms all baselines under differing settings by large margins, demonstrating the necessity of joint learning the TN with ALD. Further analysis is conducted for a better understanding of our method.
Multi-Task Learning (MTL) networks have emerged as a promising method for transferring learned knowledge across different tasks. However, MTL must deal with challenges such as: overfitting to low resource tasks, catastrophic forgetting, and negative task transfer, or learning interference. Often, in Natural Language Processing (NLP), a separate model per task is needed to obtain the best performance. However, many fine-tuning approaches are both parameter inefficient, i.e., potentially involving one new model per task, and highly susceptible to losing knowledge acquired during pretraining. We propose a novel Transformer architecture consisting of a new conditional attention mechanism as well as a set of task-conditioned modules that facilitate weight sharing. Through this construction (a hypernetwork adapter), we achieve more efficient parameter sharing and mitigate forgetting by keeping half of the weights of a pretrained model fixed. We also use a new multi-task data sampling strategy to mitigate the negative effects of data imbalance across tasks. Using this approach, we are able to surpass single task fine-tuning methods while being parameter and data efficient (using around 66% of the data for weight updates). Compared to other BERT Large methods on GLUE, our 8-task model surpasses other Adapter methods by 2.8% and our 24-task model outperforms by 0.7-1.0% models that use MTL and single task fine-tuning. We show that a larger variant of our single multi-task model approach performs competitively across 26 NLP tasks and yields state-of-the-art results on a number of test and development sets. Our code is publicly available at https://github.com/CAMTL/CA-MTL.
In recent years, several online platforms have seen a rapid increase in the number of review systems that request users to provide aspect-level feedback. Document-level Multi-aspect Sentiment Classification (DMSC), where the goal is to predict the ratings/sentiment from a review at an individual aspect level, has become a challenging and imminent problem. To tackle this challenge, we propose a deliberate self-attention-based deep neural network model, namely FEDAR, for the DMSC problem, which can achieve competitive performance while also being able to interpret the predictions made. FEDAR is equipped with a highway word embedding layer to transfer knowledge from pre-trained word embeddings, an RNN encoder layer with output features enriched by pooling and factorization techniques, and a deliberate self-attention layer. In addition, we also propose an Attention-driven Keywords Ranking (AKR) method, which can automatically discover aspect keywords and aspect-level opinion keywords from the review corpus based on the attention weights. These keywords are significant for rating predictions by FEDAR. Since crowdsourcing annotation can be an alternate way to recover missing ratings of reviews, we propose a LEcture-AuDience (LEAD) strategy to estimate model uncertainty in the context of multi-task learning, so that valuable human resources can focus on the most uncertain predictions. Our extensive set of experiments on five different open-domain DMSC datasets demonstrate the superiority of the proposed FEDAR and LEAD models. We further introduce two new DMSC datasets in the healthcare domain and benchmark different baseline models and our models on them. Attention weights visualization results and visualization of aspect and opinion keywords demonstrate the interpretability of our model and the effectiveness of our AKR method.
To minimize the accelerating amount of time invested in the biomedical literature search, numerous approaches for automated knowledge extraction have been proposed. Relation extraction is one such task where semantic relations between the entities are identified from the free text. In the biomedical domain, extraction of regulatory pathways, metabolic processes, adverse drug reaction or disease models necessitates knowledge from the individual relations, for example, physical or regulatory interactions between genes, proteins, drugs, chemical, disease or phenotype. In this paper, we study the relation extraction task from three major biomedical and clinical tasks, namely drug-drug interaction, protein-protein interaction, and medical concept relation extraction. Towards this, we model the relation extraction problem in multi-task learning (MTL) framework and introduce for the first time the concept of structured self-attentive network complemented with the adversarial learning approach for the prediction of relationships from the biomedical and clinical text. The fundamental notion of MTL is to simultaneously learn multiple problems together by utilizing the concepts of the shared representation. Additionally, we also generate the highly efficient single task model which exploits the shortest dependency path embedding learned over the attentive gated recurrent unit to compare our proposed MTL models. The framework we propose significantly improves overall the baselines (deep learning techniques) and single-task models for predicting the relationships, without compromising on the performance of all the tasks.
We consider retrofitting structure-aware Transformer-based language model for facilitating end tasks by proposing to exploit syntactic distance to encode both the phrasal constituency and dependency connection into the language model. A middle-layer structural learning strategy is leveraged for structure integration, accomplished with main semantic task training under multi-task learning scheme. Experimental results show that the retrofitted structure-aware Transformer language model achieves improved perplexity, meanwhile inducing accurate syntactic phrases. By performing structure-aware fine-tuning, our model achieves significant improvements for both semantic- and syntactic-dependent tasks.
The prompt and accurate detection of faults and abnormalities in electric transmission lines is a critical challenge in smart grid systems. Existing methods mostly rely on model-based approaches, which may not capture all the aspects of these complex temporal series. Recently, the availability of data sets collected using advanced metering devices, such as Micro-Phasor Measurement units ($\mu$ PMU), which provide measurements at microsecond timescale, boosted the development of data-driven methodologies. In this paper, we introduce a novel deep learning-based approach for fault detection and test it on a real data set, namely, the Kaggle platform for a partial discharge detection task. Our solution adopts a Long-Short Term Memory architecture with attention mechanism to extract time series features, and uses a 1D-Convolutional Neural Network structure to exploit frequency information of the signal for prediction. Additionally, we propose an unsupervised method to cluster signals based on their frequency components, and apply multi task learning on different clusters. The method we propose outperforms the winner solutions in the Kaggle competition and other state of the art methods in many performance metrics, and improves the interpretability of analysis.
We present a novel methodology to jointly perform multi-task learning and infer intrinsic relationship among tasks by an interpretable and sparse graph. Unlike existing multi-task learning methodologies, the graph structure is not assumed to be known a priori or estimated separately in a preprocessing step. Instead, our graph is learned simultaneously with model parameters of each task, thus it reflects the critical relationship among tasks in the specific prediction problem. We characterize graph structure with its weighted adjacency matrix and show that the overall objective can be optimized alternatively until convergence. We also show that our methodology can be simply extended to a nonlinear form by being embedded into a multi-head radial basis function network (RBFN). Extensive experiments, against six state-of-the-art methodologies, on both synthetic data and real-world applications suggest that our methodology is able to reduce generalization error, and, at the same time, reveal a sparse graph over tasks that is much easier to interpret.
In this work, we propose a deep learning-based approach for kin verification using a unified multi-task learning scheme where all kinship classes are jointly learned. This allows us to better utilize small training sets that are typical of kin verification. We introduce a novel approach for fusing the embeddings of kin images, to avoid overfitting, which is a common issue in training such networks. An adaptive sampling scheme is derived for the training set images to resolve the inherent imbalance in kin verification datasets. A thorough ablation study exemplifies the effectivity of our approach, which is experimentally shown to outperform contemporary state-of-the-art kin verification results when applied to the Families In the Wild, FG2018, and FG2020 datasets.
Accurate demand forecasting of different public transport modes(e.g., buses and light rails) is essential for public service operation.However, the development level of various modes often varies sig-nificantly, which makes it hard to predict the demand of the modeswith insufficient knowledge and sparse station distribution (i.e.,station-sparse mode). Intuitively, different public transit modes mayexhibit shared demand patterns temporally and spatially in a city.As such, we propose to enhance the demand prediction of station-sparse modes with the data from station-intensive mode and designaMemory-Augmented Multi-taskRecurrent Network (MATURE)to derive the transferable demand patterns from each mode andboost the prediction of station-sparse modes through adaptingthe relevant patterns from the station-intensive mode. Specifically,MATUREcomprises three components: 1) a memory-augmentedrecurrent network for strengthening the ability to capture the long-short term information and storing temporal knowledge of eachtransit mode; 2) a knowledge adaption module to adapt the rele-vant knowledge from a station-intensive source to station-sparsesources; 3) a multi-task learning framework to incorporate all theinformation and forecast the demand of multiple modes jointly.The experimental results on a real-world dataset covering four pub-lic transport modes demonstrate that our model can promote thedemand forecasting performance for the station-sparse modes.
Although deep neural networks have achieved tremendous success for question answering (QA), they are still suffering from heavy computational and energy cost for real product deployment. Further, existing QA systems are bottlenecked by the encoding time of real-time questions with neural networks, thus suffering from detectable latency in deployment for large-volume traffic. To reduce the computational cost and accelerate real-time question answering (RTQA) for practical usage, we propose to remove all the neural networks from online QA systems, and present Ocean-Q (an Ocean of Questions), which introduces a new question generation (QG) model to generate a large pool of QA pairs offline, then in real time matches an input question with the candidate QA pool to predict the answer without question encoding. Ocean-Q can be readily deployed in existing distributed database systems or search engine for large-scale query usage, and much greener with no additional cost for maintaining large neural networks. Experiments on SQuAD(-open) and HotpotQA benchmarks demonstrate that Ocean-Q is able to accelerate the fastest state-of-the-art RTQA system by 4X times, with only a 3+% accuracy drop.
Interpretable Multi-Task Learning can be expressed as learning a sparse graph of the task relationship based on the prediction performance of the learned models. Since many natural phenomenon exhibit sparse structures, enforcing sparsity on learned models reveals the underlying task relationship. Moreover, different sparsification degrees from a fully connected graph uncover various types of structures, like cliques, trees, lines, clusters or fully disconnected graphs. In this paper, we propose a bilevel formulation of multi-task learning that induces sparse graphs, thus, revealing the underlying task relationships, and an efficient method for its computation. We show empirically how the induced sparse graph improves the interpretability of the learned models and their relationship on synthetic and real data, without sacrificing generalization performance. Code at https://bit.ly/GraphGuidedMTL
General embeddings like word2vec, GloVe and ELMo have shown a lot of success in natural language tasks. The embeddings are typically extracted from models that are built on general tasks such as skip-gram models and natural language generation. In this paper, we extend the work from natural language understanding to multi-modal architectures that use audio, visual and textual information for machine learning tasks. The embeddings in our network are extracted using the encoder of a transformer model trained using multi-task training. We use person identification and automatic speech recognition as the tasks in our embedding generation framework. We tune and evaluate the embeddings on the downstream task of emotion recognition and demonstrate that on the CMU-MOSEI dataset, the embeddings can be used to improve over previous state of the art results.
Video-based person re-identification aims to match a specific pedestrian in surveillance videos across different time and locations. Human attributes and appearance are complementary to each other, both of them contribute to pedestrian matching. In this work, we propose a novel Temporal Attribute-Appearance Learning Network (TALNet) for video-based person re-identification. TALNet simultaneously exploits human attributes and appearance to learn comprehensive and effective pedestrian representations from videos. It explores hard visual attention and temporal-semantic context for attributes, and spatial-temporal dependencies among body parts for appearance, to boost the learning of them. Specifically, an attribute branch network is proposed with a spatial attention block and a temporal-semantic context block for learning robust attribute representation. The spatial attention block focuses the network on corresponding regions within video frames related to each attribute, the temporal-semantic context block learns both the temporal context for each attribute across video frames and the semantic context among attributes in each video frame. The appearance branch network is designed to learn effective appearance representation from both whole body and body parts with spatial-temporal dependencies among them. TALNet leverages the complementation between attribute and appearance representations, and jointly optimizes them by multi-task learning fashion. Moreover, we annotate ID-level attributes for each pedestrian in the two commonly used video datasets. Extensive experiments on these datasets, have verified the superiority of TALNet over state-of-the-art methods.
Recent outbreak of COVID-19 has led a rapid global spread around the world. Many countries have implemented timely intensive suppression to minimize the infections, but resulted in high case fatality rate (CFR) due to critical demand of health resources. Other country-based factors such as sociocultural issues, ageing population etc., has also influenced practical effectiveness of taking interventions to improve morality in early phase. To better understand the relationship of these factors across different countries with COVID-19 CFR is of primary importance to prepare for potentially second wave of COVID-19 infections. In the paper, we propose a novel regularized multi-task learning based factor analysis approach for quantifying country-based factors affecting CFR in early phase of COVID-19 epidemic. We formulate the prediction of CFR progression as a ML regression problem with observed CFR and other countries-based factors. In this formulation, all CFR related factors were categorized into 6 sectors with 27 indicators. We proposed a hybrid feature selection method combining filter, wrapper and tree-based models to calibrate initial factors for a preliminary feature interaction. Then we adopted two typical single task model (Ridge and Lasso regression) and one state-of-the-art MTFL method (fused sparse group lasso) in our formulation. The fused sparse group Lasso (FSGL) method allows the simultaneous selection of a common set of country-based factors for multiple time points of COVID-19 epidemic and also enables incorporating temporal smoothness of each factor over the whole early phase period. Finally, we proposed one novel temporal voting feature selection scheme to balance the weight instability of multiple factors in our MTFL model.
Sentiment analysis is a process widely used in opinion mining campaigns conducted today. This phenomenon presents applications in a variety of fields, especially in collecting information related to the attitude or satisfaction of users concerning a particular subject. However, the task of managing such a process becomes noticeably more difficult when it is applied in cultures that tend to combine two languages in order to express ideas and thoughts. By interleaving words from two languages, the user can express with ease, but at the cost of making the text far less intelligible for those who are not familiar with this technique, but also for standard opinion mining algorithms. In this paper, we describe the systems developed by our team for SemEval-2020 Task 9 that aims to cover two well-known code-mixed languages: Hindi-English and Spanish-English. We intend to solve this issue by introducing a solution that takes advantage of several neural network approaches, as well as pre-trained word embeddings. Our approach (multlingual BERT) achieves promising performance on the Hindi-English task, with an average F1-score of 0.6850, registered on the competition leaderboard, ranking our team 16th out of 62 participants. For the Spanish-English task, we obtained an average F1-score of 0.7064 ranking our team 17th out of 29 participants by using another multilingual Transformer-based model, XLM-RoBERTa.
Users from the online environment can create different ways of expressing their thoughts, opinions, or conception of amusement. Internet memes were created specifically for these situations. Their main purpose is to transmit ideas by using combinations of images and texts such that they will create a certain state for the receptor, depending on the message the meme has to send. These posts can be related to various situations or events, thus adding a funny side to any circumstance our world is situated in. In this paper, we describe the system developed by our team for SemEval-2020 Task 8: Memotion Analysis. More specifically, we introduce a novel system to analyze these posts, a multimodal multi-task learning architecture that combines ALBERT for text encoding with VGG-16 for image representation. In this manner, we show that the information behind them can be properly revealed. Our approach achieves good performance on each of the three subtasks of the current competition, ranking 11th for Subtask A (0.3453 macro F1-score), 1st for Subtask B (0.5183 macro F1-score), and 3rd for Subtask C (0.3171 macro F1-score) while exceeding the official baseline results by high margins.