Attention variants, state-space and recurrent cells, normalization and token-mixing schemes — each tested against the standard block it replaces.
Replace dense attention over structured object tokens with attention over role-filler tensor-product representations. A learned query specifies both a role and a filler, retrieves objects matching that binding, extracts a target role, and rebinds the extracted filler into an output object.
Replace the usual random or elementwise-positive linear-attention feature map with a rank-one positive-semidefinite feature map derived from query and key vectors. For normalized inputs, the resulting kernel is the squared inner product, which is nonnegative and gives a geometrically structured interference pattern that is better suited to associative recall than an arbitrary low-rank feature map.
Compress each hidden layer by retaining directions that are simultaneously reachable from the observed input distribution and observable at the network output. Unlike PCA or SVD, the retained subspace is weighted by downstream task sensitivity, so high-variance but output-irrelevant directions can be removed while low-variance predictive directions are preserved.
Replace an unconstrained token mixer or feed-forward residual map with a feature-conditioned rational operator whose transfer matrix is positive real on the positive orthant. Apply it through a resolvent, rather than an additive residual, so that the accretivity certificate yields a nonexpansive implicit update and suppresses activation explosions.
Replace ordinary row-softmax attention with a doubly stochastic Sinkhorn attention plan W, and periodically recover a gauge-fixed pairwise cost from W using the exact double-centering identity. Use this recovered cost to initialize or regularize a structured attention score, making the attention geometry identifiable despite arbitrary query and key row and column offsets.
Augment a 1D neural operator or transformer with explicit tokens for detected discontinuities. Advance each front analytically using the local Rankine–Hugoniot speed and train the network only to reconstruct smooth regions and the residual caused by source terms and grid resolution.
Initialize each row of a neural weight matrix as a stationary correlated Gaussian process instead of using independent entries, but constrain its correlation tail to remain on the finite-fourth-moment side of the transition. This creates controllable structured spectra while avoiding the heavy-edge regime predicted for correlations slower than \(t^{-1/2}\).
Replace a conventional whitening transform with a constrained whitening layer that minimizes cross-channel covariance while requiring every output channel to remain correlated with its designated input channel by at least a threshold \(\rho_{\min}\). The layer exploits the orthogonal freedom in whitening to find a rotation that preserves channel identity instead of arbitrarily mixing features. It can be inserted before an MLP, convolution, or attention projection and compared directly against…
Insert a linear Johnson–Lindenstrauss bottleneck around a set of jointly processed representations, choosing its width from the sharp finite-set dimension bound rather than from the model's nominal hidden size. The projection should preserve pairwise distances between tokens, patches, or retrieved items, allowing a downstream attention or MLP block to operate at lower width while retaining the geometry relevant to similarity computations.
Use predicted covariance reduction as a differentiable gate for selecting tokens, views, sensors, or retrieved demonstrations. The gate favors inputs with high expected information gain while accounting for acquisition cost, turning attention and data collection into active observability optimization.
Train a cheap shared multi-task probe briefly, extract one semantic embedding per task, and use density-based clustering to determine which tasks should share a neural trunk. After clustering, replace the globally shared trunk by one trunk per discovered cluster, with task heads remaining separate; this preserves cooperation among related tasks while isolating destructive task interactions.
Replace deterministic LoRA importance scores with one-sided tests of whether each rank-one update has population contribution at least a user-selected threshold. Maintain empirical contribution samples during fine-tuning, estimate their uncertainty, and prune the components with the weakest statistical evidence while respecting the target rank budget. The method should avoid deleting components merely because their latest minibatch gradient was small.
Add a differentiable orientation-pooling layer after steerable filters or an orientation-bin expansion. It aggregates unoriented line evidence using doubled-angle vectors, so a feature at angle θ is identical to one at θ+π, while symmetric orientations cancel naturally instead of producing an arbitrary mean angle. Feed the network both the Cartesian order parameter and its magnitude-based confidence.
Construct overlapping attention windows from a block design instead of using one dense sequence-by-sequence attention matrix. Every token appears in exactly $r$ windows and every token pair co-occurs in exactly $\lambda$ windows, giving uniform coverage and avoiding the uneven connectivity of arbitrary sparse masks.
Add a causal memory branch whose lag-response function is represented by a Bernstein polynomial with coefficients constrained to produce a nonnegative, decreasing, convex kernel. The branch aggregates past hidden states using this kernel, giving the model a learnable long-memory profile while preventing oscillatory, negative, or increasing historical influence.
Partition tokens into a rectangular grid and use Latin-square labels to define several sparse attention heads. Each head attends only to tokens sharing a row, column, or Latin label, while orthogonality guarantees that every pair of labels occurs at most once, reducing systematic blind spots and repeated collisions. The resulting masks are deterministic, reusable across examples, and can be generated without learned routing scores.
Construct a weight-tied transformer loop in which the recurrent state receives a bounded diagonal carry plus a learned block increment, rather than applying a residual identity inside the learned increment. Parameterize the carry so every channel is strictly below one, allowing many recurrent iterations without the state explosion observed with an unconstrained carry.
Replace ordinary pairwise attention similarity by an affinity averaged over transformed keys or values. The resulting attention is invariant to the group action on either input and avoids requiring the network to learn identical attention patterns for every rotated or transformed copy.
Replace a weight-tied residual or neural-ODE stepper with an explicit Runge–Kutta method satisfying the reused-last-stage conditions. The final derivative is evaluated at the exact endpoint and becomes the first derivative of the next step, saving one expensive neural-vector-field call per step while preserving the designed integration order.
Replace dense coarse-to-fine cross-attention at multiresolution interfaces with a sparse, nonnegative overlap operator whose weighted feature average is exactly conserved between the two resolutions. Use this operator as a low-order path and blend it with an unrestricted neural cross-attention path through a convex limiter that keeps features inside a box or simplex domain. The construction is especially suitable for adaptive token grids, hierarchical graph neural networks, neural operators…
Replace or augment geometric attention on spatial or point-cloud tokens with a positive fractional kernel containing the paper's inverse-square origin factor. This gives tokens near a designated singular center a controlled increase in receptive-field influence while preserving a scale-invariant distance decay, which may help models represent cusp-like fields, radial singularities, and multiscale spatial interactions.
Construct a sparse token-interaction graph from attention affinities and recursively split it by removing the closed neighborhoods of a small set of separator tokens. Separator tokens retain global communication, while the resulting anti-adjacent components perform local attention independently, reducing quadratic attention and KV-cache costs. The induced Erdos-Pósa theorem supplies a structural diagnostic: graphs with few anti-adjacent long-cycle or theta packings should admit small…
Replace the usual hand-designed expert-load penalty with a heterogeneous survival penalty derived from a susceptibility distribution. Each expert receives an availability factor q_e=G(A_e), where A_e is its cumulative recent routing pressure and G_e is a learned or fixed mixture of exponentials; highly used experts are suppressed smoothly, while heterogeneous experts can have different resistance to pressure. The mixture produces adaptive curvature and long-tailed penalties that may reduce…
Replace a conventional scalar activation by a geometrically indexed family of affine pieces whose slope changes with the logarithmic magnitude of the input. The same two endpoint parameters are reused across all scales, giving a compact, explicitly scale-aware activation that can represent different responses for exponentially separated activation magnitudes.