• The Korean Journal of Applied Statistics
• /
• v.37 no.5
• /
• pp.631-641
• /
• 2024

Multi-task learning is a popular approach in machine learning that aims to learn multiple related tasks simultaneously by sharing information across them. In this paper, we consider a hierarchical structure across multiple related tasks with a hierarchy of sub-tasks under the same main task, where representations used to solve the sub-tasks share more information through task-specific layers, globally shared layers, and locally shared layers. We thus propose the hierarchical multi-task learning with self-supervised auxiliary task (HiSS), which is a novel approach for hierarchical multi-task learning that incorporates self-supervised learning as an auxiliary task. The goal of the auxiliary task is to further extract latent information from the unlabeled data by predicting a cluster label directly derived from the data. The proposed approach is tested on the Hyodoll dataset, which consists of user information and activity logs of elderly individuals collected by AI companion robots, for predicting emergency calls based on the time of day and month. Our proposed algorithm is more efficient than other well-known machine learning algorithms as it requires only a single model regardless of the number of tasks, and demonstrates superior performance in classification tasks using various metrics. The source codes are available at: https://github.com/seunghan96/HiSS.

• ETRI Journal
• /
• v.45 no.1
• /
• pp.93-104
• /
• 2023

Coreference resolution is a task in discourse analysis that links several headwords used in any document object. We suggest pointer networks-based coreference resolution for Korean using multi-task learning (MTL) with an attention mechanism for a hierarchical structure. As Korean is a head-final language, the head can easily be found. Our model learns the distribution by referring to the same entity position and utilizes a pointer network to conduct coreference resolution depending on the input headword. As the input is a document, the input sequence is very long. Thus, the core idea is to learn the word- and sentence-level distributions in parallel with MTL, while using a shared representation to address the long sequence problem. The suggested technique is used to generate word representations for Korean based on contextual information using pre-trained language models for Korean. In the same experimental conditions, our model performed roughly 1.8% better on CoNLL F1 than previous research without hierarchical structure.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.2
• /
• pp.483-503
• /
• 2014

In this paper, human action recognition using pyramid histograms of oriented gradients and collaborative multi-task learning is proposed. First, we accumulate global activities and construct motion history image (MHI) for both RGB and depth channels respectively to encode the dynamics of one action in different modalities, and then different action descriptors are extracted from depth and RGB MHI to represent global textual and structural characteristics of these actions. Specially, average value in hierarchical block, GIST and pyramid histograms of oriented gradients descriptors are employed to represent human motion. To demonstrate the superiority of the proposed method, we evaluate them by KNN, SVM with linear and RBF kernels, SRC and CRC models on DHA dataset, the well-known dataset for human action recognition. Large scale experimental results show our descriptors are robust, stable and efficient, and outperform the state-of-the-art methods. In addition, we investigate the performance of our descriptors further by combining these descriptors on DHA dataset, and observe that the performances of combined descriptors are much better than just using only sole descriptor. With multimodal features, we also propose a collaborative multi-task learning method for model learning and inference based on transfer learning theory. The main contributions lie in four aspects: 1) the proposed encoding the scheme can filter the stationary part of human body and reduce noise interference; 2) different kind of features and models are assessed, and the neighbor gradients information and pyramid layers are very helpful for representing these actions; 3) The proposed model can fuse the features from different modalities regardless of the sensor types, the ranges of the value, and the dimensions of different features; 4) The latent common knowledge among different modalities can be discovered by transfer learning to boost the performance.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2004.05a
• /
• pp.106-109
• /
• 2004

Many reported methods assume that the faces in an image or an image sequence have been identified and localization. Face detection from image is a challenging task because of variability in scale, location, orientation and pose. In this paper, we present an efficient linear discriminant for multi-view face detection. Our approaches are based on linear discriminant. We define training data with fisher linear discriminant to efficient learning method. Face detection is considerably difficult because it will be influenced by poses of human face and changes in illumination. This idea can solve the multi-view and scale face detection problem poses. Quickly and efficiently, which fits for detecting face automatically. In this paper, we extract face using fisher linear discriminant that is hierarchical models invariant pose and background. We estimation the pose in detected face and eye detect. The purpose of this paper is to classify face and non-face and efficient fisher linear discriminant..

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.6
• /
• pp.783-788
• /
• 2004

Many reported methods assume that the faces in an image or an image sequence have been identified and localization. Face detection from image is a challenging task because of the variability in scale, location, orientation and pose. The difficulties in visual detection and recognition are caused by the variations in viewpoint, viewing distance, illumination. In this paper, we present an efficient linear discriminant for multi-view face detection and face location. We define the training data by using the Fisher`s linear discriminant in an efficient learning method. Face detection is very difficult because it is influenced by the poses of the human face and changes in illumination. This idea can solve the multi-view and scale face detection problems. In this paper, we extract the face using the Fisher`s linear discriminant that has hierarchical models invariant size and background. The purpose of this paper is to classify face and non-face for efficient Fisher`s linear discriminant.