Fast Multi-Instance Multi-Label Learning

Abstract

In many real-world tasks, particularly those involving data objects with complicated semantics such as images and texts, one object can be represented by multiple instances and simultaneously be associated with multiple labels. Such tasks can be formulated as multi-instance multi-label learning (MIML) problems, and have been extensively studied during the past few years. Existing MIML approaches have been found useful in many applications; however, most of them can only handle moderate-sized data. To efficiently handle large data sets, in this paper we propose the MIMLfast approach, which first constructs a low-dimensional subspace shared by all labels, and then trains label specific linear models to optimize approximated ranking loss via stochastic gradient descent. Although the MIML problem is complicated, MIMLfast is able to achieve excellent performance by exploiting label relations with shared space and discovering sub-concepts for complicated labels. Experiments show that the performance of MIMLfast is highly competitive to state-of-the-art techniques, whereas its time cost is much less. Moreover, our approach is able to identify the most representative instance for each label, and thus providing a chance to understand the relation between input patterns and output label semantics.

Weakly Supervised Person Re-Identification

“…In Table 3, we report the comparison of our method to existing state-of-the-art MIML learning methods Deep-MIML [10] and MIMLfast [16]. The DeepMIML [10] method is an end-to-end deep neural network that integrates the instance representation learning process into the MIML learning.…”

“…We consider the whole weakly supervised person re-id problem as a multi-instance multi-label learning (MIML) problem and develop a Cross-View MIML (CV-MIML) method. Compared to existing MIML algorithms [3,2,16,15,26,46,10], our CV-MIML is able to exploit similar instances within a bag for intra-bag alignment and mine potential matched instances between bags that are captured across camera views through embedding distribution prototype into MIML, which is called the cross-view bag alignment in our modeling. Finally, we embed this CV-MIML method into a deep neural network to form an endto-end deep cross-view multi-label multi-instance learning (Deep CV-MIML) model.…”

Multi-View Multi-Instance Multi-Label Learning Based on Collaborative Matrix Factorization

“…It can make predictions at the instancelevel and automatically aggregate the predictions to the bag-level. (Huang, Gao, and Zhou 2018)). M3Lcmf is also robust to a wide range of input parameters.…”

Multi‐instance multi‐label learning for surgical image annotation

“…For example, multiple segments can be extracted from an image, where each segment is represented by an instance, providing a set of multiple instances associated with multiple labels. One approach to MLL is to formulate the task as multi‐instance multilabel learning (MIML), where the image is viewed as a bag of instances corresponding to features extracted from local segments . An important advantage of MIML is that it does not require manual segmentation of regions that relate to the labels of interest.…”

“…One approach to MLL is to formulate the task as multi-instance multilabel learning (MIML), where the image is viewed as a bag of instances corresponding to features extracted from local segments. [14][15][16] An important advantage of MIML is that it does not require manual segmentation of regions that relate to the labels of interest. The regions may be rectangular patches or local regions resulted from image segmentation.…”