# Curvature-Defect Step Control

- ID: 2131
- Canonical URL: https://synthcore.org/idea/2131/curvature-defect-step-control
- API JSON: https://synthcore.org/api/idea/2131.json
- API Markdown: https://synthcore.org/api/idea/2131.md
- Verification status: unverified
- Source: [arXiv:2608.11486](https://arxiv.org/abs/2608.11486)
- Category: optimization
- Solves: stability, speedup
- ML areas: optimizer, training-dynamics
- Math tags: optimization, convex-analysis, dynamical-systems
- Ratings: usefulness 6/10; difficulty 4/10; novelty 6/10

## Idea description

Add a cheap directional curvature-defect estimator to an SGD or AdamW optimizer and shrink the step size only when the local gradient field loses the nominal contraction margin. Unlike a Hessian-norm trust-region rule, this directly measures the quantity that appears as additive instability in the Euler coupling estimate.

## Mathematical statement

For a differentiable potential V and nominal curvature parameter alpha > 0, the paper defines the curvature defect between x and y as d_alpha(x,y) = [alpha ||x-y||^2 - <grad V(x)-grad V(y), x-y>]_+, where [z]_+ = max(z,0). If grad V is L-Lipschitz and x^+ = x - h grad V(x), y^+ = y - h grad V(y), expanding the squared distance and using <grad V(x)-grad V(y),x-y> >= alpha||x-y||^2 - d_alpha(x,y) gives the one-step bound ||x^+-y^+||^2 <= (1 - 2h alpha + h^2 L^2)||x-y||^2 + 2h d_alpha(x,y). Here h is the learning rate, L is a local or global gradient-Lipschitz estimate, and d_alpha is the additive loss of contraction. The proposed optimizer uses the normalized defect r = d_alpha/(alpha||x-y||^2 + epsilon) as a directly measurable instability score and reduces h when r is large.

## Key formulas

- $$\mathfrak{d}_{\alpha}(x,y)=\bigl[\alpha\|x-y\|^{2}-\langle\nabla V(x)-\nabla V(y),x-y\rangle\bigr]_{+}.$$
- $$\|x^{+}-y^{+}\|^{2}\leq\left(1-2h\alpha+h^{2}L^{2}\right)\|x-y\|^{2}+2h\,\mathfrak{d}_{\alpha}(x,y),\qquad x^{+}=x-h\nabla V(x),\ y^{+}=y-h\nabla V(y).$$
- $$r_t=\frac{\mathfrak d_{\alpha}(\theta_t,\tilde\theta_t)}{\alpha\|\theta_t-\tilde\theta_t\|^2+\varepsilon},\qquad h_{t+1}=\operatorname{clip}\!\left(\frac{h_t}{1+\gamma r_t},h_{\min},h_{\max}\right).$$

## Implementation notes

Integrate this into the learning-rate line of SGD or AdamW, after computing the ordinary minibatch update but before applying it. Maintain a cheap probe parameter vector tilde_theta_t that shares the current weights but receives a small normalized perturbation: tilde_theta_t = theta_t + rho u_t, where u_t is a fresh Rademacher vector normalized to unit norm. Compute two gradients on the same minibatch when possible, g_t = grad_theta L(theta_t) and tilde_g_t = grad_theta L(tilde_theta_t). Estimate the directional defect with d_t = max(0, alpha rho^2 - dot(tilde_g_t-g_t, tilde_theta_t-theta_t)). Set r_t = d_t/(alpha rho^2+epsilon), smooth r_t with an exponential moving average, and update the base learning rate by h <- clip(h/(1+gamma r_t), h_min, h_max). For AdamW, apply this multiplier to the bias-corrected adaptive step before decoupled weight decay; do not alter the Adam moments. Estimate alpha from a desired nominal contraction margin, for example alpha = 0.01 times a running median of positive directional curvature dot(tilde_g-g,tilde_theta-theta)/rho^2, clipped to a positive range. Estimate L from the same secant as L_t = ||tilde_g-g||/rho and optionally enforce h <= 0.9*2*alpha/(L_t^2+2*alpha) when alpha is positive. The defect formula and additive-error interpretation come from the paper; secant estimation, gamma, rho, and clipping are empirical choices. First test on CIFAR-10 with a six-layer CNN and a small ViT, comparing SGD and AdamW against cosine decay at equal gradient evaluations. Log r_t, learning rate, gradient norm, loss spikes, and time per step. Success means fewer divergence or loss-spike runs and faster loss decrease at equal final learning rate. A negative result is increased gradient-computation cost without improved stability or time-to-accuracy.

## Disclaimer

AI-generated research hypothesis, automatically tested. Not peer-reviewed.