Attention variants, state-space and recurrent cells, normalization and token-mixing schemes — each tested against the standard block it replaces.
Replace dense token attention with K asynchronous local nonlinear consensus updates on a sparse token graph. Each selected token is moved to the unique incident p-mean of its neighbors, producing a tunable message-passing operator that is averaging for p=2 and nonlinear for other p.
Replace dense long-range token interactions with sparse message passing on a balanced hierarchical tree. Boundary token features are extended to internal tree nodes by weighted harmonic relaxation, and the resulting weighted normal derivative is used as a learned multiscale interaction or regularization signal.
Replace ordinary feature-only cross-attention between two graph sets with a transport coupling optimized using pairwise structural costs. Use the paper's positive-semidefinite graph-kernel factorization to compute the quadratic GW interaction without materializing an O(N_0^2 N_1^2) tensor, and use the inexact-gradient convergence bound to control truncated or approximate updates. The resulting coupling can be used directly as attention weights or as a soft graph-matching matrix.
Replace dense aggregation over patches, points, or graph neighbors with a state-dependent positive quadrature rule. A latent encoder produces the current manifold coordinate z, a small controller chooses a subset of entities, and a convex redistribution step assigns nonnegative weights so that the subset preserves the full aggregation on a learned feature basis. The method is especially suitable for neural operators and vision attention where many spatial entities are redundant along the data…
Replace a Cartesian grid or singular polar convolution near the center of a disk with a hierarchy of polar-spline coefficient features. Use separate coefficient spaces for scalar fields, vector 1-forms, and 2-forms, together with sparse derivative matrices satisfying the discrete complex identity D1 D0 = 0. Refine only cells with large learned or physics residuals, while retaining the pole constraints that make the representation smooth despite the collapsed angular boundary.
Translate a recursive neural architecture grammar into a combinatorial specification and use its generating function to sample architectures uniformly conditional on size. This removes the severe shape and branching biases of naive random tree generation, producing a better-covered architecture-search population at the same evaluation budget.
Construct a sparse directed attention mask over ordered tokens by treating token pairs as edges labeled by time or position. Apply a recursive three-hop spanner construction so that every dense temporal route has a sparse route using at most three times as many hops. Run attention only on the selected edges and use three sparse message-passing layers, or a small learned hop-composition block, to approximate dense causal attention.
Replace an untied stack of recurrent or polynomial-filter blocks with a delayed recurrence whose coefficients repeat with period r and whose characteristic polynomial has exact (r+1)-fold rotational symmetry. The resulting layer couples the current state to a state r steps back, producing a structured spectrum rather than arbitrary eigenvalues and potentially improving long-horizon propagation with fewer parameters.
Replace an arbitrary graph spectral filter with a constrained transition-band filter that is exactly one on the lowest L graph frequencies and zero above frequency N. Train or initialize the transition coefficients to minimize the induced infinity-norm excess above one, which controls amplification and sign overshoot while retaining low-frequency information exactly. The paper predicts that widening the transition band reduces the best achievable excess quadratically rather than linearly.
Replace repeated high-dimensional quadratic interactions in a neural operator or world model with a reduced latent quadratic map whose basis-pair interactions are precomputed offline. The online computation becomes a small polynomial in latent coefficients, preserving quadratic interaction structure while avoiding repeated full-resolution contractions.
Replace independent per-sensor temporal encoders with a shared two-dimensional transform whose temporal and spatial phase channels remain algebraically distinguishable. Train a neural residual detector and localization head on the transformed coefficients, forcing perturbations with the same temporal signature but different positions to differ primarily in their spatial phase.
Replace independent softmax routing or attention normalization with a differentiable approximate distribution over one-to-one assignments, using the Bethe permanent as the partition-function surrogate. Constrain the allowed token-to-expert or query-to-key support graph to have high girth, which gives an explicit bound on the approximation error and reduces short-cycle-induced correlations.
Build a neural interaction layer for multiple populations whose cross-population kernels are ordered, so the response of population s to population t need not equal the response of t to s. Evaluate the interaction in divergence form and apply an explicit moment-nullspace projection so each layer preserves total mass, momentum, and energy instead of learning these constraints from penalties.
Replace dense attention weights with a compactly supported anisotropic bump derived from the obstacle solution, using one learnable ellipsoid per attention head or feature group. Tokens outside the learned ellipsoid receive exactly zero weight, while tokens inside receive smoothly decaying weights according to a fractional exponent. The learned positive-definite matrix represents orientation, scale, and correlations between feature dimensions.
Use a sparse multivariate Legendre expansion as the geometry-to-network-weights map, rather than an unconstrained MLP that consumes all shape parameters. The hypernetwork predicts only coefficients for a selected set of polynomial multi-indices, allowing high-dimensional or countably parameterized shape uncertainty to be handled with a number of learned terms determined by coefficient decay rather than ambient dimension.
Replace pointwise high-order derivative residuals in an eigenvalue PINN by an assembled dynamic-stiffness residual \(\mathbf W(\omega)q_\theta\), where each element matrix is obtained from homogeneous PDE solutions. The network predicts nodal degrees of freedom or element boundary traces, while the exact frequency-domain operator enforces the physics without differentiating the network multiple times with respect to coordinates.
Add a fixed or lightly parameterized graph-level feature extractor based on short non-backtracking cycle counts, and concatenate its Poisson log-likelihood-ratio features with the output of a graph neural network. The feature scaling uses the paper's explicit means, so the network receives statistics that are approximately independent and correctly normalized rather than raw, highly correlated cycle counts.
Insert a sparse heavy-tailed nonlocal mixing operator into a residual sequence, graph, or spatial network so that information can traverse distant positions without stacking many local layers. Use the critical tail exponent s = 1/2, whose truncated first moment grows logarithmically and predicts an effective propagation distance proportional to depth times log depth rather than merely depth.
Insert a differentiable reaction-diffusion layer that converts dense token or pixel features into sparse, spatially coherent routing masks. Two competing orientations form complexes through conserved monomer reservoirs, so local assignments can cluster while opposite assignments mutually exclude one another instead of independently activating at the same location. The layer can be used as a soft-to-hard MoE router, attention-mask generator, or object-part grouping module.
Replace a sparse graph layer's separate edge transformations with one joint low-rank factorization of all transformations entering each target node. For target node i, concatenate the neighbor matrices horizontally, project all neighbor features into a shared low-dimensional receiving basis, and reconstruct one output; retain the self transformation exactly. This can reduce edge-parameter storage and message-passing FLOPs when the incoming block row has rapidly decaying singular values.
Partition neural modules into two empirically identified reliability or noise classes and restrict their communication graph to a two-block stochastic block model. Allocate a fixed connectivity budget across within-class and cross-class edges using a water-filling update that favors block pairs producing the largest increase in validation utility. The resulting layer is sparse and modular, with a testable prediction that optimal connectivity concentrates on a few block pairs rather than…
Replace random node or token subsampling by a subset selected to preserve the averages of low-frequency graph signals exactly or approximately. Build a graph over nodes or tokens, compute its Laplacian eigenspace, and choose a fixed-cardinality subset whose eigenvector moments match those of the full graph; use the retained subset for graph pooling or downstream attention.
Replace recursive adaptive coordinate traversal in a CPPN-based architecture generator with one static multiresolution candidate tensor and an active boolean mask. Evaluate all candidate coordinates in parallel, compute local CPPN-output variance over each parent cell's children, and retain only children whose parent is active and whose predicted variation exceeds a threshold. This preserves adaptive refinement while enabling population batching, JIT compilation, GPU vectorization, and…
Construct a residual neural network or graph message-passing layer whose skip edges are sampled with probability proportional to their distance as $|i-j|^{-(1+\sigma)}$, while retaining a small local backbone. The paper's mechanism predicts that coarse-grained propagation is governed by the long-range kinetic operator and is therefore asymptotically insensitive to the particular Bernoulli graph realization, yielding controllable superdiffusive information transport without dense all-to-all…