Scalable Multitask Representation Learning for Scene Classification

Abstract

We propose a multitask learning approach to jointly train a low-dimensional representation and the corresponding classifiers which scales to high-dimensional image descriptors, such as the Fisher Vector, and consistently outperforms the current state of the art on the SUN397 scene classification benchmark with varying amounts of training data.

Multi-Task Learning with Low Rank Attribute Embedding for Person Re-Identification

“…MTL has been shown successful in discovering latent relationships among tasks, which cannot be found by learning each task independently. It has been widely applied to machine learning [2,44] and computer vision [45,24]. In addition, MTL is particularly suitable for the situation in which only a limited amount of training data is available for each task.…”

Multi-task proximal support vector machine

“…Recently, Maksim et al proposed a novel multi-task learning method to learn a low-dimensional representation jointly with corresponding classifiers. This scalable multi-task representation learning method is suitable for high-dimensional features [36]. Recent works point out that we need to consider whether all the tasks are related and share a common set of features.…”

Recovery of simultaneous low rank and two-way sparse coefficient matrices, a nonconvex approach

“…We apply GDT algorithm to the problem of multi-task learning, which has been successfully applied in a wide range of application areas, ranging from neuroscience [112], natural language understanding [41], speech recognition [100], computer vision [101], and genetics [128] to remote sensing [123], image classification [74], spam filtering [119], web search [33], disease prediction [135], and eQTL mapping [69]. By transferring information between related tasks it is hoped that samples will be better utilized, leading to improved generalization performance.…”