Architecture ideas

Research ideas extracted from mathematics papers, categorized as Architecture.

Unverified 2026

Markov-Increment Window Encoder

Replace a collection of overlapping sliding-window features with approximately orthogonal incremental features: the length-m feature contains information not predictable from shorter consecutive windows. Use the paper's transition-operator Toeplitz precision matrix to decorrelate the resulting sequence of window features before attention, suppressing duplicated local evidence and improving conditioning.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Block-hierarchical covariance decompositions for finite-block additive functionals arXiv:2607.25949
Unverified 2026

Gain-aligned branch shunting

Replace an additive nonnegative feature readout by several local divisive branches, where each branch divides a signal pathway by a positive pool chosen to estimate shared multiplicative gain. Initialize or constrain each pool toward the dominant nuisance covariance direction while retaining an additive bypass so the model can reject harmful normalization. This should improve robustness when nuisance gain is shared across features, but not when the pool support is shuffled or its measurements…

Useful6/10
Difficulty5/10
Novelty5/10
Paper: When Branch-Local Shunting Helps: A Gain-Load-Alignment Principle for Dendritic E/I Networks arXiv:2607.24990
Unverified 2026

Period-Resolvent Recurrent Layer

Construct a recurrent or graph-neural layer on a finite state space with a known bijection T, such as a modular cat map, and use the diagonal resolvent gain (1 − α^kx)^−1 as a state-dependent self-return or memory coefficient. States on short periodic orbits receive larger amplification, while long-period states receive weaker amplification, producing deterministic localization without learned disorder.

Useful6/10
Difficulty5/10
Novelty8/10
Paper: Arithmetic Landscape Functions of a Discrete Cat Map arXiv:2607.24857
Unverified 2026

Double-Bracket Projector Refinement

Represent an attention or routing state as a symmetric projector or fixed-spectrum positive semidefinite matrix and refine it using the paper's double-bracket flow instead of unconstrained gradient steps. The update rotates the state toward a task-derived Hermitian cost matrix while preserving its eigenvalues, so rank, trace, and spectral diversity remain fixed by construction.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: A Path Integral Model of Cognition arXiv:2607.24807
Unverified 2026

Canonical Zero-Defect Density Layer

Convert a density-dependent operator acting only on active coordinates into an ambient neural layer by embedding with sqrt(rho) and setting its action to zero on the zero-density defect. This produces the minimum-norm extension and prevents arbitrary or unstable outputs on coordinates that the current density declares unobservable.

Useful6/10
Difficulty4/10
Novelty7/10
Paper: Density-Dependent Operators on Density-Projection Condensation Spaces: Ambient Extensions, Zero-Density Defects, and Stability arXiv:2607.24540
Unverified 2026

Proximal Tensor Attention Refinement

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.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: A Homogeneous Tensor Framework for High-Order Trust-Region and Spherical Polynomial Optimization arXiv:2607.24046
Unverified 2026

Matching-safe sparse layers

Construct the connectivity mask of a sparse MLP or MoE layer so that every active feature group is covered by a matching to an independent input or sample group. If the mask contains unmatched vertices, repair it with the fewest additional edges or low-rank skip connections before training. The goal is to avoid width- or sparsity-induced singular regimes that can produce sharp interpolation-like loss and generalization spikes.

Useful6/10
Difficulty5/10
Novelty8/10
Paper: The Zero Pattern of a Design Matrix Drives Multiple Descent in Over-parameterized Regression arXiv:2607.24041
Unverified 2026

Regular Hypergraph Masks for Balanced Sparse Attention

Construct sparse attention masks as regular or almost-regular hypergraphs: every query selects exactly w keys, while every key receives nearly equal incidence within each score or locality class. This preserves a fixed sparsity budget while removing systematic positional and head-level load imbalance that can cause hot keys, uneven memory traffic, and unstable gradients.

Useful6/10
Difficulty6/10
Novelty6/10
Paper: Optimal Dynamic Cooling of Multiple Qubits arXiv:2607.23641
Unverified 2026

Connectivity-aware ATC gradient tracking

Use the paper's mode decomposition to choose between ordinary DIGing and augmented ATC-DIGing/AugDGM according to the communication spectrum and curvature. The augmented scheme changes the disagreement feedback gain mode by mode, which can reduce the dominant pole radius on well-connected graphs without changing the neural-network architecture.

Useful6/10
Difficulty5/10
Novelty5/10
Paper: Exact Worst-case Convergence Rates of Distributed Gradient Tracking Methods arXiv:2607.23601
Unverified 2026

Berrut-Coded Neural Evaluation

Encode K input vectors at Chebyshev nodes with a Berrut rational interpolant, send different evaluations of that interpolant to N workers, and let every worker run the same neural function on its encoded input. The master reconstructs the desired outputs at the original data nodes from worker responses, allowing stragglers to be omitted and enabling robust rejection of inconsistent responses.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Robust Berrut-Approximated Coded Computing via Discrete Cosine Transforms arXiv:2607.22427
Unverified 2026

Imaginary-Axis Gramian Compression for Neural SSMs

Replace a large stable linear state-space or recurrent layer by a lower-order balanced realization computed from frequency-targeted controllability and observability Gramians. Use generalized low-rank ADI with imaginary-axis shifts concentrated at frequencies that dominate the training data, then retain states associated with the largest approximate Hankel singular values. This should reduce recurrent inference cost while preserving the layer's input-output response in the selected frequency…

Useful6/10
Difficulty6/10
Novelty7/10
Paper: A New Low-Rank Cholesky-Factor ADI Algorithm Allowing Shifts Anywhere in the Complex Plane with Applications to Data-Driven Model Reduction arXiv:2607.21969
Unverified 2026

Noncommutative adjacency-degree moments

Augment a graph neural network with features generated by noncommutative words in the adjacency matrix and diagonal degree matrix. Ordered patterns such as AD^2A and DADA distinguish where degree information occurs along a walk; the paper proves that the full scalar moment family determines every tree.

Useful6/10
Difficulty4/10
Novelty6/10
Paper: Adjacency-degree algebras and spectral determination of graphs arXiv:2607.21494
Unverified 2026

Linear-solve ensemble controller

Add a shallow neural interpolation controller to a neural ODE or state-space model so one shared vector field matches prescribed derivatives at several anchor trajectories. At every control time, compute controller weights from a small linear system instead of learning all task-specific parameters by backpropagation.

Useful6/10
Difficulty5/10
Novelty5/10
Paper: Exact ensemble controllability for neural differential equations via neural interpolation arXiv:2607.21112
Unverified 2026

L2-Certified DAG Attention Ordering

Add a learned scalar ordering to a directed graph attention layer and retain only forward edges, producing a DAG attention mask without requiring a supplied topological order. Train the ordering with a differentiable surrogate for weighted surplus, and regularize it toward the paper's explicit half-weight-minus-l2 certificate. This supplies a principled alternative to random masking or unconstrained bidirectional graph attention when causal or hierarchical information flow is desirable.

Useful6/10
Difficulty5/10
Novelty8/10
Paper: The optimal constant for minimum weight feedback arc sets in oriented graphs arXiv:2607.20996
Unverified 2026

Degree-Capped Simplicial Residual Step

Set the residual propagation coefficient of a simplicial neural layer from a cheap upper bound on the operator spectrum instead of tuning it blindly. The degree-majorization theorem supplies a bound on the largest eigenvalue, while the Brouwer-type inequality supplies a topology-count-based bound on sums of the top eigenvalues.

Useful6/10
Difficulty3/10
Novelty6/10
Paper: Degree Majorization and Laplacian Eigenvalue Sums for Simplicial Complexes arXiv:2607.20910
Unverified 2026

Separable Tensor-Product Spline Trial Layer

Use the paper's correspondence between KAN splines and finite-element or isogeometric shape functions to build coordinate-separable tensor-product trial layers. Replace additive coordinate aggregation with a multiplicative contraction of one-dimensional spline expansions, yielding an explicit tensor-product basis without storing a dense multidimensional grid.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: PG-KINN: A Physics-Informed Petrov-Galerkin Kolmogorov-Arnold Network for Solving Forward and Inverse PDEs arXiv:2607.20378
Unverified 2026

Constraint-Free Skew Coupling

Compose independently parameterized neural dynamical modules through power-preserving skew coupling instead of equality penalties or projected constraints. This creates a modular graph or world model in which information exchanged between modules is antisymmetric, so internal coupling cannot create or destroy total latent energy.

Useful6/10
Difficulty6/10
Novelty6/10
Paper: Mixed finite element discretization of intrinsic geometrically exact beams for explicit multibody dynamics arXiv:2607.20245
Unverified 2026

Hemifield sum-difference orientation channels

Compute separate doubled-angle orientation order parameters for left and right image regions, then expose their sum and difference as symmetric and antisymmetric global features. This gives a network a low-dimensional inductive bias for global vertical structure versus left-right imbalance, while retaining magnitude channels that indicate when either readout is undefined because orientations cancel.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Perceived vertical and eye level as one orientation order parameter: a closed-form account of the Li-Matin rules for egocentric space arXiv:2607.19681
Unverified 2026

Hard-Rod Symmetry Invariant Module

Construct a scalar feature or critic for oscillator-based neural dynamics that is invariant under the transformations imposed by free harmonic motion and elastic collisions. For finite-size rods, the module should represent only quantities compatible with common oscillator-phase rotations and momentum permutations, preventing a learned world model from inventing coordinate-dependent pseudo-conserved quantities that disappear after collisions.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Absence of hidden analytic conserved quantities in harmonically confined rods arXiv:2607.18872
Unverified 2026

Volume-Weighted Hodge Convolution

Replace the ordinary combinatorial Hodge propagation in a simplicial neural network with a geometry-induced weighted Hodge Laplacian built from Euclidean simplex volumes. The operator preserves the harmonic/topological subspace while changing the positive spectrum according to the shape and scale of the simplices, allowing message passing to distinguish geometrically meaningful cells that have identical incidence patterns.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Geometry-Induced Hodge Stars on Rips and Dowker--Rips Complexes arXiv:2607.18692
Unverified 2026

Ramanujan Signed Ring Mixer

Replace an unsigned two-hop cyclic mixer by the paper's alternating signed circulant. The sign pattern preserves one-step and two-step interactions while reducing the exact spectral radius from 4 to 2√2, allowing a larger raw mixing coefficient under the same operator-norm stability constraint.

Useful6/10
Difficulty3/10
Novelty6/10
Paper: Signed circulants at the Ramanujan bound arXiv:2607.18334
Unverified 2026

Regular Linear Hypergraph Attention

Construct attention groups as hyperedges of a linear r-uniform hypergraph: every pair of tokens is allowed to share at most one group, while each token participates in approximately the same number of groups. Apply local attention inside each group and aggregate the outputs across groups. The construction inherits the paper's sharp capacity bound and prevents both redundant pair interactions and high-degree token hubs.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Linear Turán Numbers of Uniform Hypertrees arXiv:2607.16854
Unverified 2026

Cycle-certified nonnegative rank-one attention

Represent a nonnegative attention or routing score matrix by two nonnegative vectors, X = uv^T, and learn only entries on a sparse bipartite graph of important query-key or token-expert interactions. Complete the remaining entries multiplicatively and monitor cycle residuals as a certificate of whether the sparse representation is compatible with rank one. Use local ratio violations to trigger additional edges or relax the rank-one approximation only where needed.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Tight Conic Relaxations for Rank-one Doubly Nonnegative Matrix Completion arXiv:2607.16796
Unverified 2026

Dyson Diagonal Scaling for Directed Message Passing

Replace ordinary row-degree or symmetric normalization in a directed graph neural network with a nonlinear Dyson scaling. For a nonnegative directed adjacency matrix A, solve a positive vector equation and propagate with B = D A D, where D is the diagonal matrix of the solution. The resulting operator has row sums strictly below one, giving an explicit bound against exploding directed message propagation while retaining asymmetric edge information.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Non-symmetric vector dyson equations arXiv:2607.16333