betaFeedback

WorksPhysics and Astronomy

The Constructive Cavity Method for the Ising Perceptron

Abstract

We apply the Constructive Cavity Method (CCM) to the Ising perceptron. The process starts from positive soft channels admitting an auxiliary representation U(x)=∫Retxρ(dt),ρ≥0, U(x)=\int_{\mathbb{R}} e^{tx}ρ(\mathrm{d}t),\qquad ρ\ge0, with an auxiliary measure having a finite exponential quadratic moment. This rewrites the perceptron as a dense, bipartite Gaussian model with an Ising layer and a continuous auxiliary layer. The CCM blanket geometry then gives two constructive variance profiles, QℓQ_\ell on the Ising side and ΓℓΓ_\ell on the auxiliary side. The bipartite shell scale is Qℓ+1Γℓ+1−QℓΓℓ, Q_{\ell+1}Γ_{\ell+1}-Q_\ellΓ_\ell, and Gaussian core comparison removes the core contribution ΔQℓΔΓℓΔQ_\ellΔΓ_\ell under blanket refinement. The scalar cavity functional has three terms, Aperc(ν)=E[log⁡⟨2cosh⁡x⟩ν+αlog⁡⟨U(s)⟩ν−log⁡⟨ey⟩ν], \mathcal{A}_{\mathrm{perc}}(ν)=\mathbb{E}\left[\log\langle2\cosh x\rangle_ν+α\log\langle U(s)\rangle_ν-\log\langle e^y\rangle_ν\right], where xx is the new-spin field, ss is the new-constraint field, and yy is the cavity correction. CCM identifies the constructive cavity variables with the Aizenman–Sims–Starr random overlap structure (ROSt) variables. Parisi's recursive averaging then gives GpercCCM(ζ,Q,Γ)=X0(0)+αY0(0)−12∑ℓ=0Rζℓ(Qℓ+1Γℓ+1−QℓΓℓ). \mathcal{G}_{\mathrm{perc}}^{\mathrm{CCM}}(ζ,Q,Γ)=X_0(0)+αY_0(0)-\frac{1}{2}\sum_{\ell=0}^{R}ζ_\ell\left(Q_{\ell+1}Γ_{\ell+1}-Q_\ellΓ_\ell\right). The replica-symmetric specialization gives Gardner's formula for the generalized Ising perceptron. Specializing the constraint terminal to the hard channel gives the Krauth–Mézard functional for the binary perceptron.

Paper details

vlnc:W_wby4hsm3
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
Usage

views · 1

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

ChatGPT (OpenAI), GPT-6 Astra

View statement
Looking for related papers…