Linear Stochastic Systems with i.i.d. uncertainties: Exact Covariance Characterization, Stability Analysis and State-feedback Design
arXiv:2608.07028
2026
Dynamics
2 ideas extracted · analyzed Aug 31, 2026
What the math gives to ML
The paper provides an exact deterministic recursion for second moments of linear systems with i.i.d. multiplicative uncertainty, and proves that mean-square stability is equivalent to Schur stability of a Kronecker-lifted matrix. This mechanism transfers directly to recurrent, state-space, and residual neural networks whose weights, activations, or deployed parameters experience multiplicative noise. The most useful implementation is a covariance-propagation monitor and regularizer that estimates the lifted second-moment operator, penalizes or constrains its spectral radius, and predicts a sharp boundary between bounded and exploding activation variance. This is more informative than ordinary spectral normalization because it captures stochastic products and cross-covariances across perturbation channels.
Ideas from this paper
Unverified
2026
Treat multiplicative weight noise, quantization error, or structured parameter uncertainty in a recurrent or state-space layer as an i.i.d. random linear operator and explicitly control its second-moment growth. Add a differentiable penalty or projection based on the spectral radius of the Kronecker-lifted operator, so the network can tolerate stochastic perturbations without exploding hidden-state variance or collapsing useful memory.
Useful8/10
Difficulty6/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Use the lifted second-moment operator to adapt the residual step size of a deep residual network or neural ODE under multiplicative layer noise. Instead of choosing a fixed residual coefficient, shrink or enlarge it online to keep the predicted covariance-growth factor below a target margin, producing a stochastic stability controller for depth and inference time.
Useful7/10
Difficulty5/10
Novelty8/10