Attention variants, state-space and recurrent cells, normalization and token-mixing schemes — each tested against the standard block it replaces.
Preprocess a noisy input graph into a high-degree core and compute a Jordan-center anchor in that core. Feed each node its distance to the anchor, and optionally use the anchor to bias graph-transformer attention; the hypothesis is that this suppresses spurious low-degree noise and gives the network a stable global coordinate system.
Replace dense spatial pooling or integral evaluation over a planar domain by a sparse cubature layer whose nodes are poles of a rational approximation fitted only on the domain boundary. For analytic or nearly analytic neural-field channels, the same learned field can then be integrated using substantially fewer evaluations than a uniform grid, while the boundary approximation residual supplies a cheap reliability signal.
Use the paper's structure-inheriting crossover to construct discrete token-to-expert assignments from two parent routers instead of randomly reinitializing routing assignments. Build a sparse token-similarity graph and optimize an objective combining within-expert similarity, cross-expert separation, and expert-load balance; use the resulting assignment to initialize router logits or to periodically repair overloaded experts. The method is especially suitable for small calibration batches or…
Replace an ordinary graph diffusion or message-passing operator with a positive-semidefinite Laplacian whose kernel contains a prescribed node-wise subspace. The layer smooths only feature components orthogonal to that subspace, preserving global constants, positional modes, or other structural signals even when graph edges are dynamically added or removed.
Build a single-hidden-layer network whose hidden weights and biases are sampled from a non-continuous distribution supported on a dense subset of parameter space, then train only the output coefficients. The result motivates discrete or mixed-precision hidden parameters without requiring a continuous Gaussian initialization; finite-width experiments can test whether this retains accuracy while reducing hidden-layer storage and arithmetic cost.
Add a preprocessing and inference module to a permutation-labeled graph network that computes the states globally compatible with all cycle transports. The module masks node or root-state logits to this fixed-point set, replacing exponential global assignment search with graph traversal plus permutation-table operations. A soft version can use the fixed-point mass as an auxiliary compatibility regularizer during training.
Replace a fixed-norm Fourier feature layer by a Fourier transform followed by spatially varying modular normalization. Use a baseline exponent approaching the endpoint regime at large coordinates and permit only bounded, smooth deviations so the transform remains controlled while the network can emphasize localized details.
Construct positional features from a self-similar digit system whose Fourier characters are orthogonal under a prescribed nonuniform measure, rather than sampling frequencies independently. Use several admissible multiplier values to create frequency bands while preserving the underlying Hadamard structure, giving a deterministic multiscale encoding with a better-conditioned feature Gram matrix on fractal or highly clustered coordinates.
Periodically project a rectangular neural-network weight matrix onto an approximately orthonormal-column matrix using LU-preconditioned CholeskyQR rather than ordinary QR or a polar iteration. Pivoted LU handles badly scaled and nearly dependent columns, while Householder orthogonalization of the LU factor produces a triangular preconditioner that makes the subsequent Cholesky step safer in fp16 or bfloat16.
Use the isolated positive spectral mode created by a finite branch defect on an otherwise long cycle as a graph positional feature. The feature should concentrate around structurally unusual vertices while remaining insensitive to the total cycle length, providing a principled alternative to raw Laplacian eigenvectors for cycle-with-branch graphs.
Represent each sequence chunk by a distribution over local descent-set patterns, then compose two chunks using the shuffle-algebra product rather than concatenating their embeddings alone. The module explicitly models how local order statistics change under interleaving, providing an inductive bias for event streams, code tokens, time-series segments, or mixtures of independently generated sequences.
Replace a locally oriented three-channel feature frame by its positive-definite polar factor, removing arbitrary SO(3) basis rotations before the feature enters an MLP, attention block, or graph message-passing layer. Process the resulting SPD matrix in log coordinates so the downstream network receives a globally unconstrained symmetric representation rather than a gauge-dependent frame.
Learn or select sparse cyclic convolution or relative-attention offsets whose pairwise differences collide less often modulo the sequence length. The paper's Fourier fourth-power identity turns this combinatorial objective into an FFT-computable differentiable loss, enabling fixed-K sparse patterns with lower aliasing and interference than random offsets.
Add a structural positional channel formed from the Krylov sequence generated by the graph adjacency matrix and the all-ones vector. For graphs with k main eigenvalues, this sequence has rank k, so a GNN can retain all information obtainable from global walk counts using only k node features rather than storing many adjacency powers.
Replace an unconstrained linear map on matrix-valued features by an exact operator-norm isometry assembled from parallel copies of X and its transpose. Contractive compression matrices and unitary basis changes allow a wider family than ordinary orthogonal layers, while a contractive remainder can increase output width without increasing the layer's spectral norm.
Build a graph diffusion or neural-operator encoder whose sparse-observation loss is weighted according to graph distance from the observed nodes. For early diffusion times, suppress supervision or cross-attention demands that are geometrically impossible because signals at distance \(d\) are attenuated like \(e^{-d^2/(2t)}\); gradually release those constraints as diffusion time grows.
Replace an unconstrained bilinear matrix fusion or covariance head with \(\Phi(A,B)=\sum_{r=1}^R V_r^*(A\otimes B)V_r\). The output is PSD by construction, and the stronger block-level property makes the layer compatible with minibatches, mixtures, and Gram-matrix inputs rather than merely preserving positivity pointwise.
Construct a shared latent transformation as a neural monad-like operator Γ=Ω∘Σ, and expose its iterates Γ^{q+1}Y as a refinement trajectory rather than stacking unrelated layers. Aggregate the resulting representations with a learned or fixed realization weighting, while training an algebra-action map θ:ΓY→Y to make one-step refinement compatible with the original representation. This creates a shallow-parameter, arbitrarily deep computation path with explicit compositional…
Replace homogeneous feature propagation with a discretized wave equation containing a positive, spatially varying learnable potential. The potential changes Hamiltonian trajectories so that feature energy reaches the layer's readout or sensor region instead of remaining in dynamically hidden modes. Train the potential jointly with the task objective and an empirical observability penalty.
Use the paper's finite-habitat approximation as a warning and design principle: averaging token- or state-dependent routing environments can reduce the persistence of specialized subnetworks. Partition inputs into environments, estimate environment-specific interaction kernels, and retain the heterogeneity that produces positive invasion margins instead of replacing it with one global average.
Replace or augment relative-position attention with a positive fractional-integration mixing kernel whose radial behavior has separate inner and outer power laws. Tokens close to one another interact through the usual fractional singularity, while tokens near different radial scales receive a ground-state correction that can improve multiscale information transport without introducing a dense learned positional table.
Construct the latent transition from a nondegenerate bilinear form phi and a form-compatible operator instead of from an unconstrained dense matrix. The resulting SSM has an exact orthogonal or symplectic algebraic structure, reducing transition parameter redundancy and testing whether preservation of a latent pairing improves extrapolation on reversible, parity-sensitive, or Hamiltonian-like sequence tasks.
Add a permutation-invariant positional channel to a graph neural network by encoding each node through the histogram of shortest-path distances to a selected landmark set. Unlike standard ordered landmark distances, this representation is unchanged when landmarks are permuted and can be optimized to reduce node collisions. Use a small learned projection of the histogram alongside ordinary node features, with an optional collision penalty during training.
Construct a periodically driven hybrid recurrent state-space model whose vector field is piecewise smooth across learned switching surfaces. Engineer a transverse homoclinic intersection around a hyperbolic recurrent state; the resulting shift-like invariant set provides a controllable symbolic reservoir for sequence prediction and long-horizon generation.