Incremental Concept Learning for Bounded Data Mining

Journal of Computer and System Sciences

Ong

et al. 2014

Self Cite

Abstract. Automatic classes are classes of languages for which a finite automaton can decide whether a given element is in a set given by its index. The present work studies the learnability of automatic families by automatic learners which, in each round, output a hypothesis and update a long term memory, depending on the input datum, via an automatic function, that is, via a function whose graph is recognised by a finite automaton. Many variants of automatic learners are investigated: where the long term memory is restricted to be the just prior hypothesis whenever this exists, cannot be of size larger than the size of the longest example or has to consist of a constant number of examples seen so far. Furthermore, learnability is also studied with respect to queries which reveal information about past data or past computation history; the number of queries per round is bounded by a constant. These models are generalisations of the model of feedback queries, given by Lange, Wiehagen and Zeugmann.

Section: Learning With Feedback and Memory Limitationsmentioning

confidence: 99%

Automatic learners with feedback queries

Case

Journal of Computer and System Sciences

Ong

et al. 2014

Self Cite

“…Both these concepts were reformalized (the former named n-feedback learning, and the latter named n-memory learning) and thoroughly studied and discussed in [CJLZ99]. Motivation for these sorts of learnability models, as discussed in [CJLZ99], comes from the rapidly developing field of knowledge discovery in databases, which includes, in particular, data mining, knowledge extraction, information discovery, data pattern processing, information harvesting, etc. Many of these tasks represent interactive incremental iterative processes (cf., e.g., [BA96] and [FPSS96]), working on huge data sets, finding regularities, and verifying them on small samples of the overall data.…”

Section: Introductionmentioning

confidence: 99%

“…Many of these tasks represent interactive incremental iterative processes (cf., e.g., [BA96] and [FPSS96]), working on huge data sets, finding regularities, and verifying them on small samples of the overall data. While the authors in [CJLZ99] explore the aforementioned formalizations of "looking back" at small (uniformly limited by some upper bound n) portions of input data in the context of regular iterative learning, we, in this research, allow the learner to test with the teacher if conjectures do not contain data in excess of the target language. Our learners may also be allowed to memorize some bounds derived from the input data seen so far -in the form of the maximal element or the length of input seen so far (the latter type of additional information for iterative learners was first considered in [CM08]).…”

Section: Introductionmentioning

confidence: 99%

Iterative Learning from Texts and Counterexamples Using Additional Information

Lecture Notes in Computer Science

Kinber

2009

Self Cite

Abstract. A variant of iterative learning in the limit (cf. [LZ96]) is studied when a learner gets negative examples refuting conjectures containing data in excess of the target language and uses additional information of the following four types: a) memorizing up to n input elements seen so far; b) up to n feedback memberships queries (testing if an item is a member of the input seen so far); c) the number of input elements seen so far; d) the maximal element of the input seen so far. We explore how additional information available to such learners (defined and studied in [JK07]) may help. In particular, we show that adding the maximal element or the number of elements seen so far helps such learners to infer any indexed class of languages class-preservingly (using a descriptive numbering defining the class) -as it is proved in [JK07], this is not possible without using additional information. We also study how, in the given context, different types of additional information fare against each other, and establish hierarchies of learners memorizing n + 1 versus n input elements seen and n + 1 versus n feedback membership queries.

“…This has motivated the analysis of so-called incremental learning, cf. [4,5,[7][8][9]12], where in each step of the learning process, the learner has access only to a limited number of examples. Thus, in each step, its hypothesis can be built upon these examples and its former hypothesis, only.…”

Section: Introductionmentioning

confidence: 99%

Towards a Better Understanding of Incremental Learning

Lecture Notes in Computer Science

Lange

Zilles

2006

Self Cite

Abstract. The present study aims at insights into the nature of incremental learning in the context of Gold's model of identification in the limit. With a focus on natural requirements such as consistency and conservativeness, incremental learning is analysed both for learning from positive examples and for learning from positive and negative examples. The results obtained illustrate in which way different consistency and conservativeness demands can affect the capabilities of incremental learners. These results may serve as a first step towards characterising the structure of typical classes learnable incrementally and thus towards elaborating uniform incremental learning methods.