A probabilistic framework for learning task relationships in multi-task learning

Yeung, Dit–Yan; Zhang, Yu

doi:10.14711/thesis-b1155168

Cited by 114 publications

(227 citation statements)

References 121 publications

Supporting

Mentioning

224

Contrasting

Order By: Relevance

“…In the current transfer counting method, we imposed an assumption that the source and target data sharing a similar manifold representation. Future work will explore ways to relax this assumption through automatic estimation of source-target relevance [37].…”

Section: Resultsmentioning

confidence: 99%

From Semi-supervised to Transfer Counting of Crowds

Loy

Gong

Xiang

2013

2013 IEEE International Conference on Computer Vision

122

View full text Add to dashboard Cite

Section: Resultsmentioning

confidence: 99%

From Semi-supervised to Transfer Counting of Crowds

Loy

Gong

Xiang

2013

2013 IEEE International Conference on Computer Vision

122

View full text Add to dashboard Cite

“…Thus, it is crucial how to capture task relatedness and incorporate it into an MTL framework. Although, many different MTL methods [7,12,18,15,28,1] have been proposed, which differ in how the relatedness across multiple tasks is modeled, they all utilize the parameter or structure sharing strategy to capture the task relatedness.…”

Section: Introductionmentioning

confidence: 99%

“…For example, some methods [3,27,28,29], use a regularized probabilistic setting, where sharing among tasks is done based on a common prior. These approaches are usually computationally expensive.…”

Section: Introductionmentioning

confidence: 99%

Multi-Task Learning with Group-Specific Feature Space Sharing

Yousefi

Georgiopoulos

Anagnostopoulos

2015

Machine Learning and Knowledge Discovery in Databases

View full text Add to dashboard Cite

When faced with learning a set of inter-related tasks from a limited amount of usable data, learning each task independently may lead to poor generalization performance. Multi-Task Learning (MTL) exploits the latent relations between tasks and overcomes data scarcity limitations by co-learning all these tasks simultaneously to offer improved performance. We propose a novel Multi-Task Multiple Kernel Learning framework based on Support Vector Machines for binary classification tasks. By considering pair-wise task affinity in terms of similarity between a pair's respective feature spaces, the new framework, compared to other similar MTL approaches, offers a high degree of flexibility in determining how similar feature spaces should be, as well as which pairs of tasks should share a common feature space in order to benefit overall performance. The associated optimization problem is solved via a block coordinate descent, which employs a consensus-form Alternating Direction Method of Multipliers algorithm to optimize the Multiple Kernel Learning weights and, hence, to determine task affinities. Empirical evaluation on seven data sets exhibits a statistically significant improvement of our framework's results compared to the ones of several other Clustered Multi-Task Learning methods.

show abstract

“…The multi-task Gaussian process (GP) model [10] and its extension [11] are recently proposed methods that adopt this approach under the Bayesian framework. Moreover, Zhang and Yeung proposed a method in [12] to learn task relationships under the regularization framework for classification and regression problems, and then extended it for feature selection problems in [13].…”

Section: Introductionmentioning

confidence: 99%

“…Our point of departure is a regularized method in [12] which learns the task relationships in the form of a task covariance matrix under a regularization framework and is related to maximum a posteriori (MAP) estimation of the weight-space interpretation of the multi-task GP model [10] presented in [11]. We then extend the formulation for boosting to give a method called multi-task boosting (MTBoost).…”

Section: Introductionmentioning

confidence: 99%

Multi-Task Boosting by Exploiting Task Relationships

Zhang

Yeung

2012

Machine Learning and Knowledge Discovery in Databases

Self Cite

View full text Add to dashboard Cite

Abstract.Multi-task learning aims at improving the performance of one learning task with the help of other related tasks. It is particularly useful when each task has very limited labeled data. A central issue in multi-task learning is to learn and exploit the relationships between tasks. In this paper, we generalize boosting to the multi-task learning setting and propose a method called multi-task boosting (MTBoost). Different tasks in MTBoost share the same base learners but with different weights which are related to the estimated task relationships in each iteration. In MTBoost, unlike ordinary boosting methods, the base learners, weights and task covariances are learned together in an integrated fashion using an alternating optimization procedure. We conduct theoretical analysis on the convergence of MTBoost and also empirical analysis comparing it with several related methods.

show abstract

A probabilistic framework for learning task relationships in multi-task learning

Abstract: Chapter 1 Introduction 1.1 Multi-Task Learning 1.2 Motivation 1.3 Main Contributions 1.4 Thesis Outline 1.5 Notations Chapter 2 Background 2.1 A Brief Survey of Multi-Task Learning 2.1.

Cited by 114 publications

References 121 publications

From Semi-supervised to Transfer Counting of Crowds

From Semi-supervised to Transfer Counting of Crowds

Multi-Task Learning with Group-Specific Feature Space Sharing

Multi-Task Boosting by Exploiting Task Relationships

Contact Info

Product

Resources

About