No label defined (Q4956017)

Let \(\mathcal A =\langle A, \vee ,\wedge ,\neg , \mathbf 0_{\mathcal A}, \mathbf 1_{\mathcal A}\rangle \) be a finite Boolean algebra, let \(B = \{x_1,x_2,\dots ,x_n\}\) be the set of all atoms of \(\mathcal A\). A possibility measure on \(\mathcal A\) is a mapping \(\pi \:A\to \langle 0,1\rangle \) such that \(\pi (\mathbf 0_{\mathcal A}) =0, \pi (\mathbf 1_{\mathcal A})=1\), and \(\pi (a \vee b)=\max (\pi (a),\pi (b))\) for all \(a,b,\in A\). A probability measure on \(\mathcal A\) is a mapping \(p\:A\to \langle 0,1\rangle \) such that \(p(\mathbf 0_{\mathcal A})= 0\), \(p(\mathbf 1_{\mathcal A})=1\), and \(\pi (a \vee b) + \pi (a \wedge b) =\pi (a)+ \pi (b)\) for all \(a,b,\in A\). Given a probability measure \(p\) on \(\mathcal A\), its entropy \(E(p)\) is defined by \(E(p) = - \sum ^n_{i-1}p(x_i)\)log \(p(x_i)\) and the Maximal Entropy (ME) principle chooses, from a nonempty set of probability measures on \(\mathcal A\), such a \(p\) for which \(E(p)\) takes its maximal value. Let \(\pi \) be a possibility measure on \(\mathcal A\), let \(\pi _i=\pi (x_i)\), let \(v_1, v_2,\dots , v_n\) be a nonincreasing sequence of nonnegative real numbers such that \(v_1=1\), \(v_2\neq 0\), and \(\sum ^n_{j\geq 2}v_j\leq 1\). The specificity measure \(Sp(\pi)\) is then defined by \(Sp(\pi) = v_1\pi _1 - \sum ^n_{j\geq 2} v_i\pi _i\). The minimal specificity (mS) principle then chooses, given a nonempty set of possibility measures on \(\mathcal A\), this one with the minimal value of \(Sp(\pi)\). The author considers two mappings: \(T_{\text{prob}}\) which, given a probability distribution \(p\), outputs the possibility distribution \(\pi (x_i) =\sum ^n_{j\geq i}p(x_i)\), and the mapping \(T_{\text{poss}}\) which, given a possibility distribution \(\pi (x_i)\), outputs the probability distribution \(p(x_i)=\sum ^n_{i=1}\) (card\((A_i))^{-1}(\pi (x_i)-\pi (x_{i+1})m_{A_i}(x_ i)\), where \(A_i=\{x\in B\:\pi (x)\geq \pi _i\}\), and \(m_{A_i}(x_i)\) is the membership degree of \(x_i\) in the fuzzy set \(A^0\) defined by its cuts \(A_i\), \(i=1,2,\dots ,n\). The author proves a number of statements showing, very roughly speaking, that applying the mapping \(T_{\text{prob}}\) to a set of probability measures, that one satisfying the ME-principle transforms into the possibility distribution which satisfies the mS-principle among the possibility measures obtained in this way. The specificity measure in question is defined by particular and rather naturally chosen weights \(v_1,\dots ,v_n\). Vice versa, under some more regularity conditions, application of the mS-principle to a set possibility measures converts into an application of the ME-principle to the resulting probability measures when \(T_{\text{poss}}\) is applied. For more details, the reader should consult the paper itself.