Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Use sign choices over redundant gradient or adapter proposals to keep the accumulated residual update small in the coordinatewise maximum norm. Constrain the sign controller to preserve a positive projection onto the desired descent direction, so it suppresses coordinate spikes without completely canceling optimization progress.
Represent a hidden state as complex-valued points on a two-dimensional lattice and replace unconstrained local updates by the exact harmonic-quadrilateral completion rule from discrete conformal geometry. Given three corners of a plaquette, compute the fourth corner by a Mobius-rational formula enforcing cross-ratio minus one, then use a learned readout or forcing term for task-specific predictions. The layer supplies a hard geometric inductive bias and a directly measurable local constraint…
Use the paper's scattering energy balance as a measurable regularizer for an existing recurrent or state-space model instead of replacing its architecture. Penalize positive violations of the per-step energy inequality and, for paired examples, penalize violations of incremental passivity so that the model learns not to amplify perturbations over long sequences.
Replace an optimizer's endpoint-only step acceptance rule with a robust envelope rule that requires all monitored neural-network constraints to remain feasible for every interpolation point between the old and proposed parameters. This targets transient instability during a large update, such as exploding activations, loss spikes, negative curvature, or violation of a spectral-norm budget, even when the final endpoint appears acceptable.
Build a recurrent or continuous-depth block from a dissipative vector field and project every state derivative onto the tangent cone of a closed convex hidden-state set. Unlike ordinary clipping, tangent-cone projection removes only the outward component at the boundary and preserves admissible motion. Under the paper's maximal-dissipativity result, the continuous flow is nonexpansive in its initial state.
Replace a standard proximal-gradient or Adam-style update for a composite neural-network objective with a golden-ratio primal-dual iteration. The optimizer separates a nonsmooth regularizer from a locally smooth loss, estimates local curvature from successive gradients, and uses dual variables for explicit constraints instead of forcing all structure into penalty coefficients. The experiment is falsifiable: at equal gradient evaluations, the method should tolerate larger initial steps and show…
Build the diffusion drift as a contractive linear term plus a spectrally controlled residual, so trajectories are pulled toward a state-dependent center while retaining nonlinear expressivity. This directly targets the paper's sharper one-sided dissipative regime rather than hoping that ordinary weight decay produces dissipativity.
Replace a fixed optimizer learning-rate field by a positive state-dependent scaling rho(theta) and penalize expansion of weighted parameter-space volume. The optimizer is encouraged to contract regions of parameter initializations that have high weighted divergence, potentially reducing sensitivity to initialization and stabilizing training near sharp or anisotropic loss landscapes.
Use the paper's mean and variance dynamics to control exploration in a population of neural-network adapters. Estimate local reward curvature from the current candidates, then choose mutation strength so selection contracts diversity only when the reward landscape is locally reliable. Increase diffusion when reward noise or selection causes population collapse.
Use the paper's negative-semidefinite interaction curvature to detect and compensate for destructive coupling among layerwise learning-rate, momentum, or preconditioner mechanisms. Instead of independently tuning mechanism amplitudes, estimate their reduced curvature after hidden optimizer states relax, then apply a low-rank trust-region step or freeze mechanisms whose interaction curvature is too negative.
Replace an unconstrained latent transition by a layer with a distinguished scalar coordinate \(t\) and a symplectic leaf state \(x=(q,p)\). The layer advances \(t\) through a Reeb drift while updating \(x\) with a symplectic Hamiltonian step, preventing arbitrary mixing between progression and content coordinates and potentially improving long-horizon stability.
Model the scalar feedback route in a recurrent layer as a rank-one perturbation of its open-loop transition. Regularize the frequency response of that route so that no mode reaches unit loop gain, directly targeting oscillatory and slowly decaying instabilities rather than relying only on gradient clipping.
Replace standard heavy-ball momentum with an update derived from a discrete kinetic-minus-loss action and a discrete viscous force. The force discretization produces a rational damping factor that remains controlled over a specified range of step sizes, potentially reducing oscillations and instability without Adam-style second-moment state.
Track the implicit l2 regularization induced by adversarial SGD and explicitly correct it when the optimizer drifts toward an undesirable ridge strength. Apply the correction first to the final linear head or a low-dimensional adapter, where feature covariance and ridge estimates are tractable.
Add a two-sided cone-restricted spectral penalty to a recurrent or state-space model. Instead of estimating growth using a symmetric singular-value surrogate, jointly optimize a positive right vector and positive left vector in the extended quotient from the paper, targeting a real generalized eigenvalue of the learned non-selfadjoint transition operator.
Replace pointwise spectral normalization of an RNN transition with a stability constraint on the entire family of input-conditioned matrices. Use a learned positive-definite metric P so every transition contracts in the same state geometry, approximating the paper's uniform exponential stability and input-forgetting guarantee.
Replace spectral-radius-only stabilization of a recurrent or state-space transition matrix with a numerical-range constraint. Penalize directions in which the Hermitian part of a rotated transition matrix has a large maximal eigenvalue, controlling nonnormal transient amplification and polynomial state propagation.
Replace the usual softmax router or soft one-hot penalty with a vector-valued phase-field regularizer whose low-energy states are exactly the expert one-hot vectors. Component-wise barriers create stable categorical phases, while a weaker coupling term suppresses invalid states such as the all-zero vector or multi-expert activation; annealing \(\varepsilon\) produces increasingly discrete routing.
Insert a weighted negative-semidefinite fourth-order mixing operator into a residual or state-space layer. Instead of learning an unconstrained token-mixing matrix, parameterize its dissipative component as Q = -a W^{-1} B^T W B, ensuring that this component cannot increase the chosen weighted feature energy. Use a boundary-aware finite-difference matrix B along the sequence axis, optionally with learnable banded coefficients while preserving the factorization.
Add a diagnostic and optional regularizer that measures whether a neural block's multi-step directed interactions differ strongly when traversed forward versus backward. This catches transient directional amplification in deep acyclic or nearly nilpotent networks, which eigenvalue or spectral-radius penalties can miss because all eigenvalues may be zero even though short directed walks are large.
Replace the random or gradient-aligned perturbation in sharpness-aware minimization with a unit perturbation direction selected by a polynomial of the local Hessian. With \(\mathscr{P}(s)=(s-\rho)^2\), the direction converges toward Hessian eigenspaces whose eigenvalues are closest to the target curvature \(\rho\), allowing regularization of a chosen curvature band instead of indiscriminately penalizing only the sharpest direction.
Replace a contractive recurrent transition by its explicit Schäffer isometric lift, optionally augmenting it with a second operator satisfying the nonlinear covariance relation $V_1V_2=V_2f(V_1)$. The lifted state preserves or nearly preserves hidden-state energy, while the covariance penalty or parameterization imposes an algebraic structure on multiple recurrent channels.
Replace one deterministic residual update with a short cyclic composition of learned vector fields evaluated for randomized, short run times. Because finite compositions of noncommuting flows generate directional-derivative and Lie-bracket terms, changing the cycle order gives the network an explicit, low-cost way to learn drift directions that are unavailable from the individual vector fields alone.
Add a loss term requiring a neural optimizer or recurrent module to decrease a nonnegative Lyapunov-like energy over M update steps, rather than forcing monotonic one-step decrease. The term includes an empirically estimated mismatch allowance, so stochastic or delayed updates are tolerated while persistent instability remains penalized.