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 independent architecture generation with a diffusion mutation kernel that starts from a known valid neural architecture, re-noises it for only a fraction of the diffusion horizon, and denoises it conditionally toward a new architecture. The resulting candidates should remain closer to the parent and retain validity at low mutation strength, while larger re-noising fractions should produce greater novelty and access to distinct architectural basins.
Replace the usual sum of pairwise modality similarities with a higher-order score based on the coordinatewise Hadamard product of all normalized modality embeddings. For modalities indexed by i=1,...,m, score a tuple using s(x_1,...,x_m)=\omega^\top(\bar g_1(x_1)\odot\cdots\odot\bar g_m(x_m)), where \omega is learned and \odot is coordinatewise multiplication. This adds explicit m-way interactions without concatenating raw features or introducing a joint encoder.
Construct a graph and its spectral positional features using affinities between inputs after optimally aligning one input over the known symmetry group. Feed these quotient-space eigenvectors to a transformer or graph neural network, so symmetry-equivalent examples receive the same structural coordinates without storing augmented copies.
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.
Constrain a student policy to transform its action in the same way that the input state is transformed, while constraining its value estimate to remain unchanged. During distillation, augment every teacher-student pair with several symmetry-transformed copies and penalize disagreement after transforming the student action back to the original frame.
Construct an efficient recurrent or state-space layer whose impulse response follows Mittag-Leffler relaxation instead of a single exponential. A bank of stable diagonal state channels approximates the long power-law tail, allowing the layer to retain information over widely separated timescales with only \(K\) states per feature.
Attach a deterministic supervisory automaton to a neural policy or sequence model and mask every event disabled by the current supervisor state. Use a short receding-horizon planner over admissible events to resolve conflicts between neural preferences and shared-resource constraints. The network scores useful actions, while the automaton supplies an exact safety layer.
Freeze a randomly initialized single-layer transformer and use a constructed soft prompt to make its attention weights equal Gaussian-kernel weights over support examples. The resulting model performs Nadaraya-Watson regression in one forward pass, so task adaptation stores prompt tokens rather than modifying network weights. Prompt length becomes the number of kernel centers, while hidden dimension and prompt norm determine whether the required logits can be represented accurately.
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.
Replace random or greedy one-expert-at-a-time activation with a deterministic binary van der Corput sequence. At each training or inference step, the schedule chooses an expert whose cumulative usage remains close to its proportional target, while recursively balancing nested expert groups rather than only balancing individual experts.
Replace a portion of quadratic key-value attention or an external episodic table with a per-sample matrix fast memory updated by rank-one delta corrections. The memory directly learns a linear key-to-value map and can be carried across sequence segments, providing cheap online adaptation with constant state size per head.
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.
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.
Add a deterministic complex-valued state-space bank whose mode detunings become progressively smaller with mode index, Delta_n=c n^{-p}, while input couplings decay as B_n=b n^{-kappa}. For slowly varying or constant forcing, the summed state follows the paper's subresonant response and grows like t^{1-alpha}, providing controllable power-law memory with only O(N) recurrent state updates. This should improve long-context retention compared with a same-size unconstrained RNN or uniformly spaced…
Replace uniform embedding dimensions with a globally budgeted allocation based on the estimated spectral complexity of each categorical feature. Tables whose category representations have large leading singular-value energy receive more dimensions, while high-cardinality tables are penalized because each extra dimension consumes more parameters.
Use cumulative control measures as the input to a neural surrogate instead of the full sequence of control values. For a quasi-static free-boundary system satisfying the paper's average-pressure path-independence assumption, two nonnegative control histories with identical integrals up to time t should produce the same state at t, allowing a smaller training input and fewer distinct control trajectories in the dataset.
Replace an unconstrained Fourier-domain linear mixer with a bank of positive spectral kernels and a max-times erosion aggregator. For a nonnegative Fourier magnitude f, each kernel produces a quotient response f/psi_k and the layer takes the pointwise supremum over kernels, giving exact positive homogeneity and monotonicity. This is most suitable as a drop-in spectral mixing block in a CNN, vision transformer, or state-space model.
Replace ordinary absolute positional embeddings with coordinates on a learned flat torus and use dual-lattice Fourier characters as positional features. Control the covariance of the coordinate fundamental domain so that the paper's inequality guarantees a lower bound on the smallest nonzero positional frequency, preventing the learned periodic coordinate system from developing arbitrarily weak or nearly constant modes.
Replace a fixed confidence-threshold early-exit rule with a finite-horizon optimal-stopping policy over the model's evolving posterior confidence. The controller stops when the calibrated expected terminal error is no greater than the cost plus expected value of executing another neural block, permitting time-dependent and nonmonotone stopping regions.
Use a fixed spherical t-design as the direction codebook for a directional attention or feature-aggregation module instead of independently sampled random directions. Equal weights provide exact zero mean and isotropic second moments, while exactness for spherical polynomials up to degree t reduces directional aliasing and seed-dependent anisotropy.
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.
Augment standard Transformer positional embeddings with coordinates generated by the paper's signed-base digit expansion. Previous binary digits determine the sign and scale of later contributions, while a two-state Markov chain controls correlations between digits. This supplies multiscale positional structure using a small number of transition and base parameters.
Build a low-rank trilinear attention module in which query, key, and value factors are constrained to the unit sphere and refined through a few proximal alternating sweeps. The proximal terms suppress factor oscillation and make each sweep improve a well-defined tensor interaction objective, offering a stable alternative to unconstrained tensor-power iterations.