Attention variants, state-space and recurrent cells, normalization and token-mixing schemes — each tested against the standard block it replaces.
Replace real diagonal state-space channels with complex damped oscillators whose hidden states encode both amplitude and phase. Train with parallel causal convolution and deploy with the equivalent one-step recurrence, allowing the same layer to support efficient batched training and low-memory streaming inference.
Replace dense token-to-token attention in selected layers with communication through a small number of multi-token hyperedges. Each hyperedge aggregates its incident token states and broadcasts the resulting message back to those tokens, allowing higher-order interactions while reducing the number of pairwise links. Reconstruct hyperedges periodically from cumulative token displacement so stable tokens retain useful groups while rapidly changing tokens are regrouped.
Add a small constrained equilibrium layer whose response depends on the current neural state and recent exogenous history, then cache responses keyed by a learned history embedding. For a new history, reuse a cached response only when an empirical Wasserstein distance to the cached history is below a threshold; otherwise run a few inner optimization iterations. The paper's local Holder and trajectory-stability results motivate graceful degradation rather than catastrophic errors for nearby…
Replace independent Gaussian attention noise or unconstrained token routing with a directed-polymer path distribution over positions and layers. The router aggregates exponentially many monotone paths through temporally correlated random edge scores, producing heavy-tailed but spatially coherent routing and preventing attention from collapsing onto a single token. The paper's t^{2/3} wandering and t^{1/3} free-energy fluctuations become measurable diagnostics and tunable targets rather than…
Represent periodic input-output behavior using a compact real vector of Fourier coefficients and learn an invertible neural map from input coefficients to output coefficients. Inference then obtains the input representation for a desired periodic output by a single inverse pass instead of iterative optimization through a nonlinear forward model, while the Fourier representation reduces sequence dimensionality when high-rate signals are spectrally sparse.
Add a causal memory branch whose weights are generated by the paper's power-type Volterra kernel rather than learned independently at every lag. Learn or softly constrain the exponents so the model can select rough short-memory behavior or smoother long-memory behavior while using only a few parameters. The branch can be implemented as a truncated causal convolution, a multiresolution approximation, or a recurrent state-space realization.
Use Kemeny’s constant as a diffusion-quality gate when adding shortcut edges or cliques to a graph used by a GNN. Candidate augmentations are accepted only when they reduce estimated average hitting time, preventing rewiring operations that superficially shorten paths but make the random walk mix more slowly.
Replace repeated full-dimensional matrix-exponential or ODE solves in a conditioned continuous-time state-space layer with contour quadrature evaluated in a projection basis. The same reduced basis and contour nodes can serve many conditioning vectors, while shifted reduced resolvents provide a stable and differentiable approximation over a prescribed time window.
Use attention-graph hitting times to identify tokens whose information has not mixed through the network, then route only those tokens through additional Transformer blocks. Tokens with fast reachability exit early, while slow or isolated tokens receive more computation.
Construct a recurrent module with a phase variable and a transverse memory coordinate modeled on a perturbed twist map. Train the transverse state to lie on an invariant graph over the phase, while the phase follows an approximately irrational rigid rotation. A KAM-inspired graph correction and residual penalty should reduce long-horizon drift in recurrent prediction.
Augment a neural policy with deterministic DFA states for the task objective and safety constraint, then select among objective-specific policy heads using those states. Before either target is reached, execute a mixed policy; after one target is reached, switch permanently to the policy specialized for the remaining target.
Replace repeated fine-scale message passing on a recursively constructed graph by a finite-state compositional module that summarizes each cell through boundary connectivity states. Child summaries are combined only through admissible configurations, so the network can represent globally consistent paths and forests without enumerating full walks. The same update map is shared across all hierarchy levels, giving logarithmic-depth computation in the graph diameter and a strong inductive bias for…
Construct a 2D recurrent or residual neural lattice with slowly varying local couplings, while parameterizing those couplings so that an anisotropy invariant remains constant across all spatial and depth locations. The network obtains controlled local propagation velocities rather than arbitrary inhomogeneous amplification, enabling depth-dependent receptive fields while preserving near-critical signal propagation.
Construct a neural layer with two independently ordered transformations and train its operators to satisfy the paper's diamond equations, so that applying direction 1 then direction 2 gives the same result as direction 2 then direction 1. Unlike ordinary weight sharing, the mixed identity permits noncommuting operators whose interaction defects cancel exactly.
Build a recurrent ReLU module that evaluates a binary refinement cascade using a fixed-dimensional state and shared cell weights. Replace hard binary digit selection with two overlapping circle coordinates; switch between their affine state updates only at points where the two candidate updates agree, so the switch is an exact continuous piecewise-linear ReLU operation rather than a multiplicative gate.
Project each FFN residual update onto the tangent space of the current token residual direction before adding it to the stream. This preserves the component that changes representation direction while suppressing norm-only motion, which may reduce residual-norm drift and aggregation-induced representation collapse.
Equip a recurrent or state-space layer with multiple noncommuting transition operators and regularize the span of finite operator words applied to the input injection matrix. This discourages hidden directions that cannot be reached from the input and may improve long-range input influence, gradient propagation, and robustness under operator switching.
Construct an unrolled phase-retrieval network that begins with an isotropic Gaussian estimate rather than a spectral initializer. Retain the AMP residual correction and Onsager subtraction, but learn the scalar measurement denoisers and step sizes; use several random starts and select the iterate with the lowest measurement residual.
Replace a learned sequence-mixing matrix with a structured lower-triangular Sprugnoli operator whose square is exactly the identity. Applying the same operator in reverse reconstructs activations exactly, so it can be used as a reversible Transformer mixer or reversible channel permutation while retaining nontrivial long-range mixing.
Build a neural feature-mixing block from a sparse exchange matrix and a fixed sequence of cluster-style mutations. Transpose-related channels share parameters through orbit tying, while positive multiplicative-additive updates are evaluated with log-sum-exp. The block is a concrete alternative to a dense MLP projection and can be tested at equal parameter count and compute.
Replace a learned nonnegative banded transition matrix by an ordered product of learnable stochastic bidiagonal factors. Each factor performs one local left or right transport step, so a product of p lower and q upper factors creates an effective bandwidth of p+q while retaining nonnegative entries, row sums equal to one, and a highly structured propagation kernel.
Store a finite library of successful robot configurations or action-conditioned waypoints and construct a smooth soft minimum of their distances. Use the negative distance gradient as a structured action prior, add a learned residual policy, and pass the combined action through a quadratic-program safety layer. This gives a neural controller an explicit attraction basin toward demonstrated solutions while preventing violations of known state constraints.
Construct a routed neural state update from a collection of branch maps whose inverse-style refinement operators are uniformly contractive. The discrete routing sequence acts as an expandable symbolic code, while contraction makes the continuous state associated with a long routing history insensitive to initialization and earlier perturbations. Use a finite active top-\(K\) set during training, but retain an expandable branch table so the model can represent increasingly complex or rare modes.
Replace unconstrained low-rank compression of a neural state with an augmented basis that always contains vectors representing known conserved quantities or diagnostically important linear statistics. After each learned transition, project the state back onto the affine constraint set with an exact minimum-norm correction, preventing rank truncation and model error from accumulating in those statistics.