Hierarchical Multi-task Learning Research Articles

Gated hierarchical multi-task learning network for judicial decision prediction

Judicial Decision Prediction (JDP) aims to predict legal judgments given the fact description of a criminal case. It consists of multiple subtasks, e.g., law article prediction, charge prediction, and term of penalty prediction. Generally, a fact description contains in-depth semantic information. Besides, there exist complex dependencies among subtasks. For instance, law article prediction could guide charge prediction and term of penalty prediction. Nonetheless, the majority of previous approaches usually capture in-depth semantic information of fact description inadequately or neglect the dependencies among subtasks. In this paper, we propose a novel gated hierarchical multi-task learning network, named GHE-DAP, to jointly model multiple subtasks in JDP. Specifically, GHE-DAP combines a Gated Hierarchical Encoder (GHE) to extract in-depth semantic information of fact description from multiple perspectives, and a Dependencies Auto-learning Predictor (DAP) to learn the dependencies among subtasks dynamically. We evaluate our model on several representative subtasks, and the experimental results demonstrate that our model outperforms state-of-art baselines consistently and significantly for JDP.

Hierarchical multi-task learning with CRF for implicit discourse relation recognition

Implicit discourse relation recognition (IDRR) remains an ongoing challenge. Recently, various neural network models have been proposed for this task, and have achieved promising results. However, almost all of them predict multi-level discourse senses separately, which not only ignores the semantic hierarchy of and mapping relationships between senses, but also may result in inconsistent predictions at different levels. In this paper, we propose a hierarchical multi-task neural network with a conditional random field layer (HierMTN-CRF) for multi-level IDRR. Specifically, a HierMTN component is designed to jointly model multi-level sense classifications, with these senses as supervision signals at different feature layers. Consequently, the hierarchical semantics of senses are explicitly encoded into features at different layers. To further exploit the mapping relationships between adjacent-level senses, a CRF layer is introduced to perform collective sense predictions. In this way, our model infers a sequence of multi-level senses rather than separate sense predictions in previous models. In addition, our model can be easily constructed based on existing IDRR models. Experimental results and in-depth analyses on the benchmark PDTB data set show that our model achieves significantly better and more consistent results over several competitive baselines on multi-level IDRR, without additional time overhead.

A Novel Orthogonality Loss for Deep Hierarchical Multi-Task Learning

In this paper, a novel loss function is proposed to measure the correlation among different learning tasks and select useful feature components for each classification task. Firstly, the knowledge map we proposed is used for organizing the affiliation relationship between objects in natural world. Secondly, a novel loss function-orthogonality loss is proposed to make the deep features more discriminative by removing useless feature components. Furthermore, in order to prevent the extracted feature maps from being too divergent and causing over-fitting which will reduce network performance, this paper also added the orthogonal distribution regularization term to constrain the distribution of network parameters. Finally, the proposed orthogonality loss is applied in a multi-task network structure to learn more discriminative deep feature, and also to evaluate the validity of the proposed loss function.The results show that compared with the traditional deep convolutional neural network and a multi-task network without orthogonality loss, the multi -task based orthogonality loss is significantly better than the other two types of networks on image classification.

Aggregating diverse deep attention networks for large-scale plant species identification

In this paper, a novel fusion method is proposed to deal with large-scale plant species identification by aggregating diverse outputs from multiple deep networks, where each deep network focus on one subset of the whole plant species. Firstly, a fixed plant taxonomy is constructed for organizing large number of fine-grained plant species hierarchically and it is further used as a guideline to help generating diverse but overlapped task groups. Secondly, an attention-based deep hierarchical multi-task learning (AHMTL) algorithm is proposed to recognize fine-grained plant species belonging to the same task group effectively by learning more discriminative deep features and classifiers jointly. Finally, we fuse all outputs from multiple deep networks to obtain the final high-level feature representation and give the prediction probability for each plant species. The experimental results have proved the effectiveness of our proposed method on large-scale plant species identification.

Camera Identification Based on Domain Knowledge-Driven Deep Multi-Task Learning

Camera identification has recently attracted considerable attention in the image forensic field of research. Several algorithms have been established based on the hand-crafted features and deep learning, through analysis of the traces achieved by the digital image acquisition process. Although these approaches have led to a breakthrough in the image forensics, some important problems still remain unsolved. For instance, extracting the hand-crafted features with human efforts is a difficult and time-consuming process, while data-driven deep learning methods tend to learn features that represent image contents rather than cameras’ characteristics. To fully take advantages of both hand-crafted and data-driven technologies, we propose a domain knowledge-driven method, which consists of one pre-processing module, one feature extractor, and one hierarchical multi-task learning procedure. The pre-processing module can introduce the domain knowledge to the subsequent deep learning network. Moreover, for device-level identification, hierarchical multi-task learning can provide more supervise information from the brand and model. The proposed framework is evaluated on three different tasks, i.e., the brand, model, and device-level identification using original and manipulated images. Our classification results demonstrate that the proposed method is effective and robust. To evaluate the robustness of the proposed method, we create a new database for the cell-phone identification and evaluate the proposed method. It is found that the accuracy of the cell-phone device identification can reach 84.3%, which is much higher than that of the camera identification. Moreover, the t-distributed stochastic neighbor embedding visualization results confirm that the features of different cell-phone devices are visually more separable than cameras.

Spatial Projection of Multiple Climate Variables Using Hierarchical Multitask Learning

Future projection of climate is typically obtained by combining outputs from multiple Earth System Models (ESMs) for several climate variables such as temperature and precipitation. While IPCC has traditionally used a simple model output average, recent work has illustrated potential advantages of using a multitask learning (MTL) framework for projections of individual climate variables. In this paper we introduce a framework for hierarchical multitask learning (HMTL) with two levels of tasks such that each super-task, i.e., task at the top level, is itself a multitask learning problem over sub-tasks. For climate projections, each super-task focuses on projections of specific climate variables spatially using an MTL formulation. For the proposed HMTL approach, a group lasso regularization is added to couple parameters across the super-tasks, which in the climate context helps exploit relationships among the behavior of different climate variables at a given spatial location. We show that some recent works on MTL based on learning task dependency structures can be viewed as special cases of HMTL. Experiments on synthetic and real climate data show that HMTL produces better results than decoupled MTL methods applied separately on the super-tasks and HMTL significantly outperforms baselines for climate projection.

Hierarchical Learning of Tree Classifiers for Large-Scale Plant Species Identification.

In this paper, a hierarchical multi-task structural learning algorithm is developed to support large-scale plant species identification, where a visual tree is constructed for organizing large numbers of plant species in a coarse-to-fine fashion and determining the inter-related learning tasks automatically. For a given parent node on the visual tree, it contains a set of sibling coarse-grained categories of plant species or sibling fine-grained plant species, and a multi-task structural learning algorithm is developed to train their inter-related classifiers jointly for enhancing their discrimination power. The inter-level relationship constraint, e.g., a plant image must first be assigned to a parent node (high-level non-leaf node) correctly if it can further be assigned to the most relevant child node (low-level non-leaf node or leaf node) on the visual tree, is formally defined and leveraged to learn more discriminative tree classifiers over the visual tree. Our experimental results have demonstrated the effectiveness of our hierarchical multi-task structural learning algorithm on training more discriminative tree classifiers for large-scale plant species identification.

A sparse nonparametric hierarchical Bayesian approach towards inductive transfer for preference modeling

In this paper, we present a novel methodology for preference learning based on the concept of inductive transfer. Specifically, we introduce a nonparametric hierarchical Bayesian multitask learning approach, based on the notion that human subjects may cluster together forming groups of individuals with similar preference rationale (but not identical preferences). Our approach is facilitated by the utilization of a Dirichlet process prior, which allows for the automatic inference of the most appropriate number of subject groups (clusters), as well as the employment of the automatic relevance determination (ARD) mechanism, giving rise to a sparse nature for our model, which significantly enhances its computational efficiency. We explore the efficacy of our novel approach by applying it to both a synthetic experiment and a real-world music recommendation application. As we show, our approach offers a significant enhancement in the effectiveness of knowledge transfer in statistical preference learning applications, being capable of correctly inferring the actual number of human subject groups in a modeled dataset, and limiting knowledge transfer only to subjects belonging to the same group (wherein knowledge transferability is more likely).