Metadata-driven Task Relation Discovery for Multi-task Learning

Abstract

Task Relation Discovery (TRD), i.e., reveal the relation of tasks, has notable value: it is the key concept underlying Multi-task Learning (MTL) and provides a principled way for identifying redundancies across tasks. However, task relation is usually specifically determined by data scientist resulting in the additional human effort for TRD, while transfer based on brute-force methods or mere training samples may cause negative effects which degrade the learning performance. To avoid negative transfer in an automatic manner, our idea is to leverage commonly available context attributes in nowadays systems, i.e., the metadata. In this paper, we, for the first time, introduce metadata into TRD for MTL and propose a novel Metadata Clustering method, which jointly uses historical samples and additional metadata to automatically exploit the true relatedness. It also avoids the negative transfer by identifying reusable samples between related tasks. Experimental results on five real-world datasets demonstrate that the proposed method is effective for MTL with TRD, and particularly useful in complicated systems with diverse metadata but insufficient data samples. In general, this study helps in automatic relation discovery among partially related tasks and sheds new light on the development of TRD in MTL through the use of metadata as apriori information.

Multi-task Learning for Concurrent Prediction of Thermal Comfort, Sensation, and Preference

“…The advantage that MTL offers is best summarized by Caruana et al, "MTL improves generalization by leveraging the domain-specific information contained in the training signals of related tasks" [28]. The paradigm of Multi-task learning (MTL) has recently been applied to the domain of thermal comfort, primarily to solve the challenges of energy efficiency of buildings and HVAC control [26], [27], [29], [30]. In [27], authors employ multi-task learning to propose a portable building management solution for better HVAC control.…”

“…The task-definition is based on publicly available building metadata such as the Brick database [31] and the solution is validated on the ASHRAE RP884 database [32]. Using metadata for task-identification is suitable to avoid the problem of negative transfer i.e., incorrect task construction and learning unrelated tasks [29]. However, the use of metadata in MTL is also challenging due to the problems of inaccurate representation generation, need for domain expertise in creating metadata, variation in the context and types of the metadata itself, and improper integration with the MTL system [27], [29].…”

“…However, the current studies that leverage MTL studies are aimed at optimizing building and HVAC efficiency, and rely on sensor data and metadata. Further, none of the existing MTL studies have been conducted in naturally ventilated built environments (e.g., classrooms) or train their models using subjective survey and questionnaire data [26], [27], [29], [30].…”

Learning Multi-Tasks with Inconsistent Labels by using Auxiliary Big Task

“…Typically, [38] embeds prior knowledge (i.e., pathological images with different magnification belong to the same subclass) into the feature extraction process among different tasks to verify the relationship between tasks and pathological image categories for fine-grained classification and pathophysiological image classification. [39] uses a kind of meta data (i.e., contextual attributes) as a priori knowledge to capture the relationship between different tasks for multiple tasks clustering. [40] uses the same subclass of the gland area as the prior information in the convolutional neural network to guide the network inference for pathological colon image analysis.…”

Learning Twofold Heterogeneous Multi-Task by Sharing Similar Convolution Kernel Pairs

“…2) parameter matrix factorization based: e.g., [38] uses the matrix tri-factorization to process collective matrices associated with different tasks and performs joint learning to predict two types of drug-disease associations. 3) equal prior shared assumptions based: e.g., [39] uses a kind of meta data (i.e., contextual attributes) as a priori information to capture the relationships between different tasks for multiple tasks clustering. The above methods mainly utilize model parameters to associate different tasks.…”