Machine-learning ideas tagged Initialization in the ML taxonomy of the Math2NN corpus.
Replace Cox-de Boor evaluation of each cubic B-spline edge activation with its fixed truncated-power expansion. Normalize each scalar edge input to a bounded knot coordinate, evaluate the five shifted cubic positive-part terms in parallel, and contract them with the learned spline coefficients inside one fused kernel.
Replace a single neural representation of a rapidly oscillatory PDE solution by a macroscopic network plus an explicitly oscillatory corrector network. Feed the network both the slow coordinate $x$ and fast coordinate $y=x/\varepsilon$, and train the resulting composite field in a variational energy objective. This targets the paper's scale-robust approximation bound rather than forcing the optimizer and finite sample set to discover oscillations of wavelength $\varepsilon$.
Replace uniform collocation for a fixed random-feature neural PDE solver with sampling from the leverage-score density of the operator-applied features. Whiten the retained residual feature space before solving for output coefficients, so the sampled least-squares matrix has an identity-like expected Gram rather than inheriting severe anisotropy from the differential operator. The same construction can be used for a linearized neural network by treating Jacobian features as the trial functions.
Use a spectral eigenvalue-counting function to bracket each target mode before neural optimization. The network then solves only within an interval containing exactly one eigenfrequency, preventing optimization from repeatedly collapsing to the lowest mode or jumping between modes.
Train a residual network on a coarse depth mesh, estimate a dual-weighted residual for every layer interval, and insert new layers at intervals with the largest estimated contribution to objective error. This replaces uniform depth expansion or expensive neural architecture search with targeted refinement driven by both forward-dynamics error and downstream loss sensitivity.
Replace the unconstrained transition of a recurrent or state-space neural network with a DMDc-initialized linear latent transition plus a learned nonlinear residual. Estimate the transition from a short warm-up dataset using Hankel delay coordinates, retain eigenmodes with decay rates near the unit circle for long-term memory, and let the neural residual model dynamics not explained by the linear backbone. This should make long-horizon prediction and slowly varying signals easier to learn while…
Use the paper's parameterized invariant-torus residual and pseudo-arclength Newton correction to train a neural ODE across a continuous family of latent dynamical regimes. The continuation constraint allows the solver to pass through saddle-node folds, where stepping a physical control parameter alone would fail or jump to a different branch.
Replace ordinary momentum with a semi-implicit velocity update containing viscous damping and a proximal dry-friction step, while evaluating the gradient at a look-ahead parameter point. The dry-friction proximal operator exactly zeros sufficiently small velocities, which may suppress late-training oscillations and create finite-time stationarity instead of merely asymptotic velocity decay.
Parameterize an orthogonal or semi-orthogonal neural weight matrix directly on the Stiefel manifold and update it with a Cayley retraction instead of unconstrained SGD plus a penalty or QR projection. The update preserves orthogonality exactly, is second-order accurate for the appropriate metric, and avoids the cubic QR factorization at every optimizer step.
Replace the recurrent transition or state-space mixer with a reversible transport followed by complementary relaxation of symmetric and antisymmetric feature components. The construction preserves a weighted energy and damps both parity sectors, giving bounded long-horizon powers without requiring the learned transition matrix itself to be symmetric. A numerical-range ellipse can be used as a cheap training-time certificate against transient growth.
Replace a trainable shallow hidden layer by a deterministic feature dictionary generated from Chebyshev-spaced scalar parameters and quasi-uniform ridge directions. Train only the output linear map, or use the frozen layer as the first stage of a larger network, thereby eliminating hidden-layer backpropagation while retaining a constructive smooth-function approximation guarantee.
Replace an unconstrained deep residual recurrence by a discretized diffusion system over feature or token positions, with trainable source terms and analytically constrained boundary feedback. The state remains nonnegative under nonnegative inputs, while negative boundary gains enforce exponential decay of perturbations and prevent exploding activations in very deep stacks.
Initialize an invertible neural linear layer from a bounded discrete random matrix only after checking that every leading principal submatrix is nonsingular and that its LU growth factor is below a prescribed threshold. This replaces blind random initialization with a cheap resampling rule designed to prevent zero pivots and excessive finite-precision amplification in reversible or flow-based networks.
Replace raw polynomial or Fourier-like features in a small MLP with basis functions orthonormal under a Sobolev inner product that jointly measures feature magnitude and input derivative magnitude. This explicitly controls feature smoothness while preserving decorrelation, potentially improving conditioning and reducing the need for large derivative-regularization coefficients.
Use the network-space inverse inequality to choose derivative order, collocation resolution, and feature separation jointly instead of enforcing arbitrarily high-order residuals on an under-resolved network. This creates an anti-aliasing rule: a network whose parameters are separated by \(\underline h\) cannot represent high Sobolev frequencies without a factor \(\underline h^{-(r-s)}\), so derivative penalties above the resolvable order should be disabled or accompanied by refinement.
Replace eigenvalue-only stability checks for a continuous-time recurrent or state-space layer with an explicit finite-horizon transient-growth test. Penalize state matrices that have small spectral decay but large induced norms of exp(tA), exp(tA^{-1}), or their discretized transition operators. This targets the paper's phenomenon in which a system is exponentially stable in continuous time yet numerically and inversely unstable because its eigenbasis is highly conditional.
Add a periodic coarse optimization phase to SGD or Adam that operates on a compressed parameterization and returns a prolongated correction to the full network. Retain nonsmooth constraints or regularizers explicitly through a primal-dual update instead of relying on penalty smoothing. Accept the correction only when it improves a cheap fine-batch merit test, making the method useful even when the coarse objective is only approximately coherent with the fine objective.
Construct one reference Lanczos basis for a symmetric propagation operator H, then derive several seed-specific spectral responses for Q_a(H)x_0 through finite-band polynomial connectors. With degree-r seeds, each transformed basis vector uses at most 2r+1 neighboring reference basis vectors, avoiding a separate Lanczos factorization for every seed.
Replace a dense token-mixing matrix in a sequence model with a fixed or learnable SBP derivative operator D=P^{-1}Q. The discrete integration-by-parts identity makes the interior mixing energy-neutral or boundary-dissipative, reducing exploding activations in deep residual stacks while preserving directional information along the sequence.
Replace ordinary Fourier, polynomial, or raw-coordinate features for a bounded scalar coordinate with Hermite functions evaluated after a monotone endpoint transform. The transform sends endpoint singularities to localized tails on the real line, while a learnable scale controls how many Hermite modes are needed. This is suited to coordinate MLPs, neural operators, and implicit fields whose targets have square-root, logarithmic, boundary-layer, or derivative singularities.
Construct a reusable ReLU trunk that emits approximate univariate powers or Legendre-polynomial features for each input coordinate, then combine them with a linear or low-rank polynomial head. This gives a compact explicit basis for smooth functions and can replace a large generic MLP in low-dimensional scientific regression or serve as a frozen or partially trainable front-end.
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 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.
Use diagonal similarity scaling to reduce the induced infinity norm of a sparse recurrent transition or companion operator before or during training. Similarity preserves eigenvalues while changing coordinate-wise amplification, so the method can suppress non-normal transient growth without changing the represented linear dynamics.