Analogical Dissimilarity: Definition, Algorithms and Two Experiments in Machine Learning

Miclet, Laurent; Bayoudh, Sabri; Delhay, Arnaud

doi:10.1613/jair.2519

Cited by 94 publications

(108 citation statements)

References 26 publications

Supporting

Mentioning

104

Contrasting

Order By: Relevance

“…This approach provides remarkable results and, in several cases, outperforms the best known algorithms [4]. Another equivalent approach [6] does not use a dissimilarity measure but just applies the previous continuity principle, adding flexibility by allowing to have some components where analogy does not hold.…”

Section: Analogical Classification: the Standard Viewmentioning

confidence: 98%

“…Such a modeling has been only recently developed in algebraic or logical settings [2,8,5,6]. Then analogical proportions turn to be a powerful tool in classification tasks [4].…”

Section: Introductionmentioning

confidence: 99%

“…This is done on the basis of triples of examples a, b and c of the training set that are such that the analogical proportion a : b :: c : d holds on vector components for all, or at least on a large number of, the attributes describing the items. This approach has been tested on benchmarks [4] where results competitive with the ones of classical machine learning methods have been obtained. These good results have remained largely unexplained since it looks unclear why this analogical proportion-based approach may be so effective, beyond the general merits of analogy.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Analogical Classification: A Rule-Based View

Bounhas¹,

Prade

Richard

2014

Information Processing and Management of Uncertainty in Knowledge-Based Systems

View full text Add to dashboard Cite

OATAO is an open access repository that collects the work of Toulouse researchers and makes it freely available over the web where possible. This is an author-deposited version published in : http://oatao.univ-toulouse.fr/ Eprints ID : 15181The contribution was presented at IPMU 2014:http://www.ipmu2014.univ-montp2.fr/ Abstract. Analogical proportion-based classification methods have been introduced a few years ago. They look in the training set for suitable triples of examples that are in an analogical proportion with the item to be classified, on a maximal set of attributes. This can be viewed as a lazy classification technique since, like k-nn algorithms, there is no static model built from the set of examples. The amazing results (at least in terms of accuracy) that have been obtained from such techniques are not easy to justify from a theoretical viewpoint. In this paper, we show that there exists an alternative method to build analogical proportionbased learners by statically building a set of inference rules during a preliminary training step. This gives birth to a new classification algorithm that deals with pairs rather than with triples of examples. Experiments on classical benchmarks of the UC Irvine repository are reported, showing that we get comparable results.

show abstract

Section: Analogical Classification: the Standard Viewmentioning

confidence: 98%

“…Such a modeling has been only recently developed in algebraic or logical settings [2,8,5,6]. Then analogical proportions turn to be a powerful tool in classification tasks [4].…”

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Analogical Classification: A Rule-Based View

Bounhas¹,

Prade

Richard

2014

Information Processing and Management of Uncertainty in Knowledge-Based Systems

View full text Add to dashboard Cite

show abstract

“…The corresponding hexagon is pictured in Figure 3. This hexagon has a nice interpretation in terms of analogical dissimilarity in the sense of [14]. Indeed the analogical dissimilarity of the 6 valuation patterns in L is 0, since they correspond to the 6 cases where A holds true; the analogical dissimilarity of the 8 valuation patterns in K is 1, since in each case it is enough to switch one bit for getting a pattern for which proportion A is true; the analogical dissimilarity of the 2 patterns in J is 2 since one needs to change 2 bits to get a pattern where proportion A is true.…”

Section: Structures Of Opposition Among Proportionsmentioning

confidence: 99%

Picking the one that does not fit - A matter of logical proportions

Prade¹,

Richard²

2013

Proceedings of the 8th Conference of the European Society for Fuzzy Logic and Technology

View full text Add to dashboard Cite

Quiz or tests about reasoning capabilities often pertain to the perception of similarity and dissimilarity between situations. Thus, one may be asked to complete a series of entities A, B, C by an appropriate X, or to pick the one that does not fit in a list. It has been shown that the first problem can receive a solution by solving analogical proportion equations between the representations of the entities in a logical setting, where we assume that X should be such that A is to B as C is to X. In this paper, we focus on the second problem, and we show that it can be properly handled by means of heterogeneous proportions that are the logical dual of the homogeneous proportions involved in the modeling of analogical proportions and related proportions. Thus, the formal setting of logical proportions, to which homogeneous and heterogeneous proportions belong, provides an appropriate framework for handling the two problems in a coherent way. As it already exists for homogeneous proportions, a particular multiple-valued logic extension of heterogeneous proportions is discussed (indeed being an intruder in a group may be a matter of degree).

show abstract

“…It is possible to augment this form of inference with other models of plausible reasoning, such as reasoning based on analogical (and other logical) proportions [9,12]. Moreover, as in [2], we could take into account externally obtained similarity degrees, using rules such as those in Section 4.…”

Section: Related Workmentioning

confidence: 99%