<p>We consider the Ising perceptron with gaussian disorder, which is equivalent to the discrete cube <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\{-1,+1\}^N\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mrow> <mo stretchy="false">{</mo> <mo>-</mo> <mn>1</mn> <mo>,</mo> <mo>+</mo> <mn>1</mn> <mo stretchy="false">}</mo> </mrow> <mi>N</mi> </msup> </math></EquationSource> </InlineEquation> intersected by <i>M</i> random half-spaces. The perceptron’s <i>capacity</i> is the largest integer <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(M_N\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>M</mi> <mi>N</mi> </msub> </math></EquationSource> </InlineEquation> for which the intersection is nonempty. It is conjectured by Krauth and Mézard (1989) that the (random) ratio <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(M_N/N\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msub> <mi>M</mi> <mi>N</mi> </msub> <mo stretchy="false">/</mo> <mi>N</mi> </mrow> </math></EquationSource> </InlineEquation> converges in probability to an explicit constant <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\alpha _\star \doteq 0.83\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msub> <mi>α</mi> <mo>⋆</mo> </msub> <mo>≐</mo> <mn>0.83</mn> </mrow> </math></EquationSource> </InlineEquation>. Kim and Roche (1998) proved the existence of a positive constant <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\gamma \)</EquationSource> <EquationSource Format="MATHML"><math> <mi>γ</mi> </math></EquationSource> </InlineEquation> such that <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(\gamma \leqslant M_N/N \leqslant 1-\gamma \)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>γ</mi> <mo>⩽</mo> <msub> <mi>M</mi> <mi>N</mi> </msub> <mo stretchy="false">/</mo> <mi>N</mi> <mo>⩽</mo> <mn>1</mn> <mo>-</mo> <mi>γ</mi> </mrow> </math></EquationSource> </InlineEquation> with high probability; see also Talagrand (1999). In this paper we show that the Krauth–Mézard conjecture <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(\alpha _\star \)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>α</mi> <mo>⋆</mo> </msub> </math></EquationSource> </InlineEquation> is a lower bound with positive probability, under the condition that an explicit univariate function <InlineEquation ID="IEq8"> <EquationSource Format="TEX">\(\mathscr {S}_\star (\lambda )\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msub> <mi mathvariant="script">S</mi> <mo>⋆</mo> </msub> <mrow> <mo stretchy="false">(</mo> <mi>λ</mi> <mo stretchy="false">)</mo> </mrow> </mrow> </math></EquationSource> </InlineEquation> is maximized at <InlineEquation ID="IEq9"> <EquationSource Format="TEX">\(\lambda =0\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>λ</mi> <mo>=</mo> <mn>0</mn> </mrow> </math></EquationSource> </InlineEquation>. Our proof is an application of the second moment method to a certain slice of perceptron configurations, as selected by the so-called TAP (Thouless, Anderson, and Palmer, 1977) or AMP (approximate message passing) iteration, whose scaling limit has been characterized by Bayati and Montanari (2011) and Bolthausen (2012).</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Capacity lower bound for the Ising perceptron

  • Jian Ding,
  • Nike Sun

摘要

We consider the Ising perceptron with gaussian disorder, which is equivalent to the discrete cube \(\{-1,+1\}^N\) { - 1 , + 1 } N intersected by M random half-spaces. The perceptron’s capacity is the largest integer \(M_N\) M N for which the intersection is nonempty. It is conjectured by Krauth and Mézard (1989) that the (random) ratio \(M_N/N\) M N / N converges in probability to an explicit constant \(\alpha _\star \doteq 0.83\) α 0.83 . Kim and Roche (1998) proved the existence of a positive constant \(\gamma \) γ such that \(\gamma \leqslant M_N/N \leqslant 1-\gamma \) γ M N / N 1 - γ with high probability; see also Talagrand (1999). In this paper we show that the Krauth–Mézard conjecture \(\alpha _\star \) α is a lower bound with positive probability, under the condition that an explicit univariate function \(\mathscr {S}_\star (\lambda )\) S ( λ ) is maximized at \(\lambda =0\) λ = 0 . Our proof is an application of the second moment method to a certain slice of perceptron configurations, as selected by the so-called TAP (Thouless, Anderson, and Palmer, 1977) or AMP (approximate message passing) iteration, whose scaling limit has been characterized by Bayati and Montanari (2011) and Bolthausen (2012).