Enabling learning from large datasets: applying active learning to mobile robotics

Abstract

Absrmct-Autonomous navigation in outdoor, om-road environments requires solving complex classification problems. Obstacle detection, road following and terrain classification are examples of tasks which have been successfully approached using supervised machine learning techniques for classification. Large amounts of training data are usually necessary in order to achieve satisfactory generalization. In such .cases, manually labeling data becomes an expensive and tedious process.This paper describes a method for reducing the amount of data that needs to he presented to a human trainer. The algorithm relies on kernel density estimation in order to identify "interesting" scenes in a dataset. Our method does not q u i r e any interaction with a human expert for selecting the images; and only minimal amounts of tuning are necessary.We dimonstrate ils effectiveness in several experiments using data collected with two diNerent vehicles. We first show that our method automatically selects those scenes from a large dataset that a person would consider "important" for classification tasks. Secondly, we show that hy labeling only few of the images selected hy our method, we obtain classification performance that is comparable to the one reached after labeling hundreds of images from the same dataset. 0-7803-8232-3/04/$17.00 WOO4 IEEE I _ on Research and dwelopmenr in informarion retrieval. Springer-Verlag New York. Inc., t994, pp. 3-12, H. Seung, M. Opper. and H. Sampolinsky, "Query by comminee:' in Pmceedings ofthe Fifth IVorkrhop on Compuroional Leoming Theocv, San Mateo, CA, 1992, pp. 287-294. C. M. Bishop, Neuml Ncworkr for pan ern^ Recoxnirion. Oxford University kiss: 1997. T. Hastie. R. libshirani. and J. Friedman. 7'he Elemenrr 2 Smrirrical ~enming,' ser. springer Series in Statistics. A. Gray and A. Moon, "Rapid evaluation of multiple density models," in Anificiol Intelligence m d Sforisrim, 2003. -. "'N-Body' pmbleb in staristical learning:' in Advances in Neural Infomurrion Pmcesring Sprema, T. K. Leen and T. 0. Dienerich, EA% M m Press. 2001. A. Moore, "Very fast EM-based mixture model clusrering using multiresolution KI-ma," in Adwnccs in Neuml Informarion Pmcessing

Active learning from demonstration for robust autonomous navigation

“…Novelty detection in and of itself is a rich field of research [8], and has seen many applications to mobile robotics [9], [10]. [11], [12] specifically propose density estimation in the context of active learning for robotic perception systems. Specifically, a large dataset consisting of a robot's recorded perceptual history is analyzed to identify the most unlikely single percepts (given the entire history).…”

“…While [11] uses non-parametric density estimation, parametric approaches are generally much faster at test time, allowing for easier application in an online setting (to allow for query filtering as well as a pool approach). Therefore, parametric density estimation is preferable in this context.…”

Batch Mode Active Learning for Multimedia Pattern Recognition

“…The method balanced exploration and exploitation and used probabilistic active learning algorithm to predict the policies that would produce higher expected gains. Dima et al [199] described an active learning algorithm based on kernel density estimation to identify exemplar images in a dataset.…”

Active Learning For Outdoor Obstacle Detection

“…To address this important problem, we have developed the Unlabeled Data Filtering (UDF) algorithm [24]. The intuition behind UDF is simple: we are interested in reducing the size of an unlabeled data set by discarding redundant examples, while keeping most of the rare patterns.…”

“…In [24] we argued that aggregating interest measures over all the patches within an image is undesirable because high interest patches can be overwhelmed by large amounts of uninteresting patches. The other extreme of scoring an image only based on its most interesting patch is also ineffective, because the most unusual patterns in an image will often be outliers corresponding to various artifacts.…”