|
|
|
|
|
|
|
|
|
Sparse data problem
| Глоссарий по искусственному интеллекту |
A common problem for statistical and other empirical learning techniques is the sparse data problem. if the learning is limited to simple frequency counts of "events" in the training sets, the model will always behave poorly on new input that introduces related but "new" events. the more detailed the original model becomes, the more data required to learn (estimate) the various parts of the model; therefore, the amount of data supporting any one part becomes increasingly sparser. practically speaking, this simply means that the input data can rarely cover all possible cases. one approach for dealing with this problem is use of theoretical knowledge to add structure to the empirical model or the use of smoothing techniques to "fill in" the missing data.
|
|
Statistical, английский
Supporting, английский
- Поддержка: обеспечение
- Монтаж и крепление
Theoretical, английский
- Теоретический
- Теоретический theor/y теория; учение; принцип; гипотеза ~ of buckling теория продольного изгиба ~ of elasticity теория упругости ~ of elastic stability теория устойчивости упругих систем ~ies of failure теории [гипотезы] прочности ~ of plastic behavior теория пластичности ~ of plasticity теория пластичности ~ of plates теория тонких пластин ~ of shallow shells теория пологих оболочек ~ of shells теория оболочек ~ of stability теория устойчивости ~ of structures теория сооружений, строительная механика ~ of thin shells теория тонких оболочек ~ of torsion теория кручения
|
Upward closure, английский
A collection of sets is upwardly closed with respect to a property if, when a set p has a property, all sets containing p have that property. an example would be a dependency in a market basket analysis. once a set of items are known to be dependent, any larger set containing those items is also known to be dependent. see also: dependence rule, downward closure. ur-resolution ur-resolution is an inference rule that can be used in automated reasoning systems. it focuses on two or more clauses, requiring that one of the clauses (the nucleus) contain at least two literals and the remaining (the satellites) contain exactly one literal. briefly, a conclusion is yielded if a unifier (substitution of terms) for variables can be found that, when applied, makes identical (except for being opposite in sign) pairs of literals, one literal from the nucleus with one from a satellite. the conclusion is yielded by, ignoring the paired literals, applying the unifier simultaneously to the nucleus and the satellites and taking the union of the resulting literals. see also: hyperresolution.
Market basket data, английский
A common problem in data mining deals with a type of data known as market basket data. this is data where each record consists of a list of items that were purchased simultaneously or otherwise naturally grouped. the typical example would be purchase records from a retail store although the techniques could be applied in many other diverse problems (e.g., medical symptoms). this label is commonly used when the purpose of the analysis is to discover association rules or dependency rules. the former are simply rules that state if item a is in the "basket," then item b is also in the basket, with two measures, the support (the percentage of cases that have both a & b) and the confidence (the percentage of cases with b among those that have a). the latter are a generalization of association rules that are covered elsewhere. see also: association rules, dependence rule.
|
|
|
|
|
|
|
