Replace ordinary dot-product attention logits with a strictly totally positive kernel evaluated on positive, ordered scalar coordinates attached to queries and keys. Use the modified-Bessel kernel K(x,s)=I_s(x), whose every ordered minor is positive, then row-normalize it as an attention matrix. This creates an attention operator with a mathematically enforced anti-oscillatory structure rather than merely positive entries.
Build a sparse neural mixing layer from colored directed strands rather than a dense all-to-all matrix. Feature channels are assigned ordered colors, local trivalent junctions conserve every color, and an edge width is the weighted sum of the colors carried by that edge; a differentiable penalty favors monotone, crossing-free routings that define a canonical leading term. This creates a structured routing prior that can be compared directly against dense attention and unconstrained sparse…
Represent a sparse higher-order attention head by a 3-uniform hypergraph whose hyperedge $(v,x,y)$ allows anchor token $v$ to aggregate a pairwise interaction between tokens $x$ and $y$. During mask construction, greedily reject edges that would create a $4$-cycle in the link graph $L(v)$, so the same pair of source tokens cannot reach an anchor through multiple redundant pairings. This produces a diversity-constrained sparse attention pattern with an explicit, measurable collision bound.
Add a norm-controlled feature mixer that applies a polynomial spectral filter to the channel covariance of a transformer or MLP block. A quadratic filter centered at \(\rho\) suppresses covariance eigenmodes far from the target and preserves modes near it, providing a tunable alternative to purely variance-maximizing mixing or standard normalization.
Construct metric-graph Laplacian positional encodings only at frequencies whose empirical eigenvalues are statistically stable under the paper’s $(n v_\mu(h))^{-1/2}$ law. Use local ball-mass estimates and empirical eigengaps to gate or downweight unreliable eigenvectors, preventing small-sample spectral noise from entering a GNN or graph transformer.
Replace an unconstrained linear residual adapter by an operator \(T\) satisfying a polynomial relation in the commutator operator \(\Delta_A(X)=AX-XA\). Choose the polynomial roots in a stable half-plane so that repeated commutators become nilpotent, making repeated adapter application terminate algebraically and permitting a finite-polynomial inverse of \(I+T\).
Insert a fixed or lightly gated lower-triangular binomial-transform layer into a sequence model to create global interactions across positions without forming attention logits. For a sequence of length N, mix each output position with all earlier positions using coefficients determined by an integer m; initialize the layer fixed and optionally learn a diagonal channel gate or a small mixture over m values.
Paper: $m$-Bell and $m$-Stirling numbers: Iterated binomial transforms, hyper-Bessel functions, and moments of the Conway--Maxwell--Poisson distributionarXiv:2608.12011
Replace explicitly stored independent random positional features with deterministic multiplicative phase features generated from a small table of random phases indexed by primes. Restrict positions to integers whose prime factors exceed a slowly growing threshold, because the paper's central-limit result says periodic weighted sums over these rough integers recover Gaussian behavior despite strong multiplicative dependence.