Learning Probabilistic Linear-Threshold Classifiers via Selective Sampling

Cesa-Bianchi, Nicolò; Conconi, Alex; Gentile, Claudio

doi:10.1007/978-3-540-45167-9_28

Cited by 22 publications

(20 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A standard tf-idf bag-of-words encoding was used to transform each news story into a normalized vector of real attributes (see Cesa-Bianchi et al, 2003, for details on preprocessing).…”

Section: Methodsmentioning

confidence: 99%

Tracking the Best Hyperplane with a Simple Budget Perceptron

Cesa-Bianchi

Gentile

2006

Learning Theory

Self Cite

View full text Add to dashboard Cite

Abstract. Shifting bounds for on-line classification algorithms ensure good performance on any sequence of examples that is well predicted by a sequence of changing classifiers. When proving shifting bounds for kernel-based classifiers, one also faces the problem of storing a number of support vectors that can grow unboundedly, unless an eviction policy is used to keep this number under control. In this paper, we show that shifting and on-line learning on a budget can be combined surprisingly well. First, we introduce and analyze a shifting Perceptron algorithm achieving the best known shifting bounds while using an unlimited budget. Second, we show that by applying to the Perceptron algorithm the simplest possible eviction policy, which discards a random support vector each time a new one comes in, we achieve a shifting bound close to the one we obtained with no budget restrictions. More importantly, we show that our randomized algorithm strikes the optimal trade-off U = Θ( √ B) between budget B and norm U of the largest classifier in the comparison sequence. Experiments are presented comparing several linear-threshold algorithms on chronologically-ordered textual datasets. These experiments support our theoretical findings in that they show to what extent randomized budget algorithms are more robust than deterministic ones when learning shifting target data streams.

show abstract

“…A standard tf-idf bag-of-words encoding was used to transform each news story into a normalized vector of real attributes (see Cesa-Bianchi et al, 2003, for details on preprocessing).…”

Section: Methodsmentioning

confidence: 99%

Tracking the Best Hyperplane with a Simple Budget Perceptron

Cesa-Bianchi

Gentile

2006

Learning Theory

Self Cite

View full text Add to dashboard Cite

show abstract

“…If there is disagreement between these experts (usually measured using metrics such as Kullback-Leibler divergence [59] or Jensen-Shannon divergence [270]) beyond a specified threshold, the example is queried and used to re-train the classifiers. [273], and demonstrated their practical applicability in optical character recognition. Dredze and Crammer [263] proposed an online active learner for natural language processing, where the distance of a point from the margin in a large-margin classifier is combined with parameter confidence.…”

Section: Online Active Learning: Related Workmentioning

confidence: 99%

Conformal predictions in multimedia pattern recognition

Balasubramanian¹

2011

SIGMultimedia Rec.

View full text Add to dashboard Cite

The fields of pattern recognition and machine learning are on a fundamental quest to design systems that can learn the way humans do. One important aspect of human intelligence that has so far not been given sufficient attention is the capability of humans to express when they are certain about a decision, or when they are not. like a bird', in his words). I cannot thank him enough for his strong belief in me over the years, for setting standards of excellence that will take me a lifetime to scale, for housing me in an environment suffused with bright minds and numerous opportunities for exposure and growth, and most importantly, for his never-failing support through every high and low of my PhD.

show abstract

“…Many criteria reflecting various measures of the expected error reduction have been proposed [5,14,18,16,8], with sometimes encouraging results in, for instance, pharmaceutical industry [25]. Specific algorithms and methods have been developed for active learning in linear and kernel spaces, either heuristically [20]or with theoretical foundations [3,8,2].…”

Section: State Of the Artmentioning

confidence: 99%

Complexity Bounds for Batch Active Learning in Classification

Rolet

Teytaud

2010

Machine Learning and Knowledge Discovery in Databases

View full text Add to dashboard Cite

Abstract. Active learning [1] is a branch of Machine Learning in which the learning algorithm, instead of being directly provided with pairs of problem instances and their solutions (their labels), is allowed to choose, from a set of unlabeled data, which instances to query. It is suited to settings where labeling instances is costly. This paper analyzes the speed-up of batch (parallel) active learning compared to sequential active learning (where instances are chosen 1 by 1): how faster can an algorithm become if it can query λ instances at once?There are two main contributions: proving lower and upper bounds on the possible gain, and illustrating them by experimenting on usual active learning algorithms. Roughly speaking, the speed-up is asymptotically logarithmic in the batch size λ (i.e. when λ → ∞). However, for some classes of functions with finite VC-dimension V , a linear speed-up can be achieved until a batch size of V . Practically speaking, this means that parallelizing computations on an expensive-to-label problem which is suited to active learning is very beneficial until V simultaneous queries, and less interesting (yet still bringing improvement) afterwards.

show abstract

Learning Probabilistic Linear-Threshold Classifiers via Selective Sampling

Cited by 22 publications

References 18 publications

Tracking the Best Hyperplane with a Simple Budget Perceptron

Tracking the Best Hyperplane with a Simple Budget Perceptron

Conformal predictions in multimedia pattern recognition

Complexity Bounds for Batch Active Learning in Classification

Contact Info

Product

Resources

About