Active learning using a low-rank classifier

Babaee, Mohammadreza; Tsoukalas, Stefanos; Datcu, Mihai

doi:10.1109/iraniancee.2015.7146279

Cited by 2 publications

(1 citation statement)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“… Multi-class active learning studies. Active Learning (AL) in general as we mentioned before works to minimize the cost of obtaining labels by enhancing the quality of the labelled set [5], or enhance the process of annotating the labels [6].AL implemented was in multiple frame works, each frame-work have its own mechanism such as the stream-based sampling where the obtaining an unlabelled instance is free , so it can first be sampled from the actual unlabelled set and then the learner can decided whether or not to request its label [7], ranked [8] or the pool-based sampling [9]. Pool based sampling works by dividing the dataset into two sets where the smallest set is the training set and the larger set is the pool and the query queried in a greedy fashion.…”

Section: Related Workmentioning

confidence: 99%

Untitled

2019

International Journal of Advanced Research in Computer and Comm

View full text Add to dashboard Cite

The main goal of Active Learning (AL), is to empower the learning process by reducing the cost of obtaining the labels with a limited training samples and by selecting the most informative samples from the unlabelled set. AL was implemented to solve a wide range of problems in all fields [1]. In this paper, we will go forward to utilize a multiclass active learning model over an ecological zones dataset in order to classify the dune-beach interface to a divided sub-environments category. Pool, ranked & stream-based sampling were used as an active learning frames where three query strategies were tested with each frame to achieve the best performance. The performance metrics values would be illustrated in three different comparative statements, in term of different framework implemented by different query strategies, in term of incremental learning process behaviour in a pool-based sampling frame work by selecting 20 different query quires from the unlabelled set to present the incremental learning process behaviour with each query strategy (random, entropy & margin) and finally in term of incremental learning process behaviour above a three different AL frame works and three different query strategies in term of accuracy metric.

show abstract