Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Augment neural-network parameters with momentum variables and update the pair using a symplectic map generated by a Hamiltonian. The optimizer approximately preserves a modified Hamiltonian, reducing systematic energy drift and potentially making long unrolled optimization more stable.
Transform local neural residuals into the Ripa model's characteristic coordinates before spatial aggregation, apply a mode-dependent gate based on neighboring characteristic jumps, and transform back. This lets the model damp oscillatory acoustic or equilibrium-mode corrections near discontinuities without globally smoothing every feature.
Parameterize a learned feature-space operator as accretive but not necessarily symmetric, then apply its fractional power through a finite positive mixture of shifted resolvents. This provides a matrix-function layer that can represent directional and rotational interactions while avoiding unstable eigendecomposition of nonnormal matrices.
Replace the usual explicit residual update with a nonstandard general-linear block containing several internal feature stages. The effective step is a positive denominator function rather than the raw depth step, allowing the block to take large nominal steps while damping the update and preserving bounded activations. This is most promising for deep residual MLPs, neural ODE discretizations, and state-space sequence models where exploding hidden states limit usable depth.
Separate a neural network into nonlinear hidden parameters and a linear output layer. Solve the output layer exactly by least squares, then update hidden parameters with a truncated-pseudoinverse Gauss-Newton step that discards numerically singular directions.
Replace explicit RK integration in a stiff neural ODE or continuous-depth residual network with the paper's stiffly accurate SDIRK4 discretization. Instead of performing a dense Newton solve for each implicit stage, solve the diagonal stage equation using a Chebyshev-accelerated residual iteration whose polynomial damps the negative, high-magnitude Jacobian modes responsible for stiffness.
Replace an unpreconditioned conjugate-gradient solve for a damped Gauss–Newton step with a two-level algebraic preconditioner derived from local Jacobian-row supports. Use overlapping local parameter blocks as Schwarz subdomains and a coarse basis containing low-energy local modes, so the optimizer can correct both localized and globally coupled parameter errors.
Replace a dense recurrent transition matrix with a periodic CMV-style product of alternating local 2x2 unitary cores. The transition is exactly norm-preserving, has O(n) trainable parameters under periodic tying, and can be applied through local factor operations rather than stored as an n-by-n matrix. Use turnover refactorization when changing the ordering or boundary connection of cores, enabling a compact cyclic unitary state-space layer.
Replace the usual linear predictor in continuation of an implicit neural state with a fractional-power predictor fitted from recent states, then correct the prediction using a pseudo-arclength constraint. This is designed for equilibrium layers, implicit sequence models, or homotopy training schedules where the state Jacobian becomes nearly singular and ordinary Newton correction or fixed-point iteration becomes unstable.
Replace a fixed or hand-tuned learning-rate schedule with a slowly exponentially increasing schedule, and restart the schedule whenever the update norm grows at least as fast as the schedule itself. The restart preserves the current parameters but resets the learning-rate multiplier, allowing the optimizer to repeatedly approach the largest locally stable step size without requiring a Hessian spectrum or a reliable initial learning-rate guess.
Add a learned Riesz-transform branch that extracts normalized spatial gradients after diffusion by a positive parabolic operator. The diffusion branch carries smooth semantic content, while the Riesz branch represents boundaries, motion changes, and graph discontinuities. Resolvent smoothing makes the derivative branch less sensitive to feature noise than directly applying a finite difference.
Add a spatially weighted TV penalty to a neural inverse solver, where a pixel receives a large penalty when perturbations there are strongly visible to the forward operator and a small penalty when the operator is insensitive. This prevents ordinary TV from suppressing or displacing structures differently across the field of view. The weight can be recomputed per acquisition geometry or cached for a fixed forward operator.
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.
Represent a solution on an unfitted domain with local neural subnetworks and train them using one augmented energy containing the bulk physical energy, symmetric Nitsche boundary or interface terms, and a derivative-jump ghost penalty. Automatic differentiation of this scalar objective supplies all gradients and avoids independently tuning inconsistent PDE residual, flux, and boundary losses. The method is especially suited to moving geometries, cut-cell domains, and domain-decomposed neural…
Augment a CNN with a nonlocal feature-gradient branch that compares each feature vector with a kernel-weighted neighborhood rather than using only pointwise or local convolutional interactions. Regularize this branch using the paper's Fourier multiplier energy, which penalizes feature oscillations according to the kernel spectrum and approaches an ordinary local-gradient operator as the interaction radius tends to zero.
Add a conservative correction after low-rank tensor compression so selected linear moments of an activation or learned state are exactly preserved. This can reduce tensor rank and memory without allowing compression error to accumulate in physically meaningful global quantities.
Treat a coupled neural training loop as a delayed feedback system with two hard delays and two first-order implementation filters. Estimate the dominant coupled Jacobian mode and use the characteristic equation to distinguish a recoverable delay-induced oscillation from a filter-induced instability; then reduce stale-gradient delay only in the former case, and slow or retune EMA or relaxation filters in the latter.
Insert a scalar flux-correction-style limiter after a neural operator predicts a conservative state or residual. Interpolate between a known-admissible baseline state and the learned high-order candidate, choosing the largest coefficient that satisfies a geometric family of linear inequalities encoding positive density, positive pressure, and subluminal velocity. This retains as much of the neural prediction as possible instead of independently clipping physical variables.
Bootstrap the optimizer curvature scale from a deliberately nondegenerate pair of gradient queries, then perform steepest descent in lp geometry with a local secant backtracking rule. The method does not require a supplied learning rate, smoothness constant L, initial distance R, or optimum value f*, and it automatically uses the dual norm associated with p.
Replace scalar neural activations by pairs of nonnegative channels whose ratio represents the signed or unsigned activation. Implement multiplication and addition through pair algebra, and renormalize each pair because the representation is invariant under multiplying both rails by the same positive scalar. This creates an explicitly bounded, cancellation-aware arithmetic layer for deep multiplicative MLPs, rational networks, and neural fields.
Replace a standard nonlinear recurrent transition with a truncated Carleman lift containing levels $z_j\approx u^{\otimes j}$, coupled by linear maps that represent quadratic, linear, and forcing terms. The resulting transition is linear in the lifted state but still expresses nonlinear dynamics in the original state, while the highest-order omitted interaction supplies an explicit truncation-defect signal that can be used for adaptive order selection or regularization.
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.
Replace AdamW or SGD updates on simplex-valued routing probabilities with a logarithmic-barrier mirror step. The update remains strictly positive, avoids projection-induced zero coordinates, and can approach a boundary solution asymptotically while retaining the paper's theoretically motivated O(log k/k) convex convergence behavior.
Replace a dense weak pairwise interaction over tokens or graph nodes by an analytically resummed Gaussian quadratic term plus explicitly evaluated corrections containing only subsets of size at most R. Choose R from the target approximation error and interaction scale instead of fixing the interaction order heuristically. This is intended for dense global attention, graph interactions, or latent-variable energy models with individually small couplings.