Architecture ideas

Attention variants, state-space and recurrent cells, normalization and token-mixing schemes — each tested against the standard block it replaces.

Unverified 2026

Threshold-Projection Recurrent Memory

Replace or augment a recurrent cell with multiple hysteresis memory branches whose states remain unchanged while the input stays within a branch-specific radius, then move toward the current input only when that radius is exceeded. The resulting cell has explicit persistence and bounded state changes, giving it an inductive bias for temporal hysteresis and reducing the need for the network to learn long-term memory behavior from scratch.

Useful6/10
Difficulty4/10
Novelty7/10
Paper: Accounting for Hysteresis and Eddy Currents in Finite Element Simulations of Ferromagnetic Laminated Cores using a Recurrent Neural Network arXiv:2607.14321
Unverified 2026

Spectral latent geometry for sparse attention

Build a sparse graph by thresholding normalized token or item inner products, then use the leading eigenvectors of its centered adjacency matrix as geometric features or a low-rank attention-logit bias. The graph avoids storing all pairwise similarities, while the paper's spectral bound supplies a concrete signal-to-noise test for deciding whether the resulting embedding is trustworthy.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Spectral Concentration and Recovery in Sparse High-Dimensional Random Geometric Graphs arXiv:2607.14304
Unverified 2026

Capacity-aware compressive-plus-indexed memory

Replace a purely recurrent or state-space history summary with two explicitly separated paths: a fixed-size state channel for compressed sequence mixing and a query-dependent indexed channel for exact or near-exact retrieval. Train a lightweight gate to invoke top-k retrieval only when the recurrent state has insufficient evidence for the current query, preserving near-constant cost on ordinary tokens while preventing catastrophic failures on long-range exact-recall tasks.

Useful6/10
Difficulty5/10
Novelty4/10
Paper: The Capability Convergence Hypothesis: Capability from Access Structure, Not Scale arXiv:2607.14144
Unverified 2026

Gaussian Simplex Classification Head

Replace the unconstrained final classifier with equal-norm regular-simplex class directions and train it under explicit isotropic Gaussian feature noise. At fixed signal energy and equal class priors, the paper's Gaussian-max theorem predicts that this geometry maximizes finite-noise maximum-likelihood decoding probability, making it a concrete candidate for robust classification heads.

Useful6/10
Difficulty4/10
Novelty4/10
Paper: Stochastic Domination of Gaussian Maxima: A Resolution of the Weak Simplex Conjecture arXiv:2607.14087
Unverified 2026

Signature-memory neural CDE

Replace an unconstrained recurrent memory with a truncated path-signature state that is updated continuously from the input control path. Feed this structured state to a learned vector field, allowing the model to represent path-dependent dynamics through iterated integrals of the entire history rather than only the latest hidden state.

Useful6/10
Difficulty5/10
Novelty5/10
Paper: Dynamic Universal Approximation via Signature Controlled Differential Equations arXiv:2607.13886
Unverified 2026

Hermite-Schatten spectral layer

Replace a dense learned linear operator on continuous or image features by a truncated Hermite projection expansion whose coefficients are directly regularized in a Schatten-p norm. The layer becomes a structured low-rank operator, while the radial Hermite-Laguerre correspondence provides an analytically tractable parameterization and an exact spectral penalty.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Quantitative Fourier Restriction Estimates for Weyl Operators: Fourier-Support Dependence and Lower Bounds arXiv:2607.13697
Unverified 2026

Deadline-Aware Fair-to-Greedy Router

Use deadline objectives to train or control a router that explicitly trades off completion probability against completed work by a fixed horizon. Begin with fair allocation for robust exploration, then anneal toward a feedback-greedy rule once per-item difficulty estimates have sufficient evidence.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Meeting Uncertain Threats with Feedback arXiv:2607.13648
Unverified 2026

Finite-Orbit Circulant State Core

Replace the linear state transition in a small recurrent or state-space module by a circulant matrix acting on a vector over a finite field. The hidden state then has only finitely many possible values and follows an exactly periodic orbit after at most \(q^n\) states, eliminating numerical drift on modular-counting and symbolic-memory tasks. A learned real-valued encoder and decoder can surround the discrete core, while the transition itself is fixed, searched, or trained with a…

Useful6/10
Difficulty6/10
Novelty8/10
Paper: Periodicities in the Riordan arrays of polynomials over finite fields arXiv:2607.13442
Unverified 2026

Active-Set Constraint Projection Layer

Replace generic projected-gradient iterations for equality-plus-bilateral constraints with an active-set semismooth Newton layer. Each iteration fixes currently active lower and upper bounds and solves one structured saddle-point system, potentially converging in a few iterations when only a small subset of constraints is active.

Useful6/10
Difficulty6/10
Novelty6/10
Paper: Duality Framework for Flux Constrained Flow in Porous Media: Analysis and Numerics arXiv:2607.13256
Unverified 2026

SBP Energy-Stable Sequence Mixer

Replace a dense token-mixing matrix in a sequence model with a fixed or learnable SBP derivative operator D=P^{-1}Q. The discrete integration-by-parts identity makes the interior mixing energy-neutral or boundary-dissipative, reducing exploding activations in deep residual stacks while preserving directional information along the sequence.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Gaussian FSBP operators: Comparison and application to numerical methods for hyperbolic conservation laws arXiv:2607.13224
Unverified 2026

Lie-Exchange Configuration Mixer

Add a profile-preserving exchange layer that mixes representations through pairwise state swaps, with weights shared over all positions. Unlike ordinary token-wise MLP mixing, the layer explicitly models interactions between two categorical states while respecting the symmetry of position permutations.

Useful6/10
Difficulty6/10
Novelty6/10
Paper: The Action of the Lie Algebra $\mathfrak{sl}_n$ on Colored Graphs and Multicolored Johnson Graphs arXiv:2607.13208
Unverified 2026

Saturating Trail Memory for Asynchronous Multi-Agent Networks

Equip multiple recurrent agents with a shared spatial or token-level trail field whose influence is a bounded function of accumulated visitation, rather than an unbounded additive memory. Use the paper's simultaneous/sequential invariance as a falsifiable design target: parallel and randomly ordered asynchronous agent updates should produce nearly identical predictions when trail occupancy is saturated, while deliberately nonsaturating controls should show order dependence. This can enable…

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Exact collective first-passage statistics of N trail-interacting walkers arXiv:2607.13213
Unverified 2026

Arcosh-convex PINN output

Replace the usual unconstrained PINN output u_theta(x) with a latent field w_theta(x), and reconstruct the physical solution as u_theta(x) = -2 log cosh(w_theta(x)). Train w_theta to be convex while enforcing the Liouville or real two-Hessian PDE residual and the boundary condition u = 0. The transformation automatically gives u less than or equal to zero, matching the target solutions, while convexity supplies a strong global shape prior.

Useful6/10
Difficulty5/10
Novelty8/10
Paper: Strict Convexity for Solution of Liouville-Type Dirichlet Problems arXiv:2607.12849
Unverified 2026

Endpoint-transformed Hermite feature layer

Replace ordinary Fourier, polynomial, or raw-coordinate features for a bounded scalar coordinate with Hermite functions evaluated after a monotone endpoint transform. The transform sends endpoint singularities to localized tails on the real line, while a learnable scale controls how many Hermite modes are needed. This is suited to coordinate MLPs, neural operators, and implicit fields whose targets have square-root, logarithmic, boundary-layer, or derivative singularities.

Useful6/10
Difficulty4/10
Novelty7/10
Paper: Hermite spectral approximation for functions with endpoint singularities using exponential transforms arXiv:2607.12648
Unverified 2026

Hessian-guided orthogonal sparsification

Train a square orthogonal neural mixer while maximizing its entrywise fourth-power concentration. When optimization reaches a non-permutation stationary configuration, explicitly test rank-two row or column rotations and take a rotation with positive exact second variation, using the paper's constructive saddle-escape mechanism.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Local Maxima of the Entrywise $\ell_4$ Norm on the Orthogonal Group arXiv:2607.12431
Unverified 2026

Analytic KL spatial adapter

Replace a dense spatial parameter field in a neural field or convolutional adapter by a truncated squared-exponential KL expansion with analytic Gaussian-Hermite modes. The amplitude and correlation length remain trainable, but changing them only rescales coefficients and basis parameters instead of triggering a numerical eigensolve.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Hierarchical Bayesian inversion using the Karhunen-Loève expansion with analytical eigenpairs of the squared exponential kernel arXiv:2607.12387
Unverified 2026

Horizontal Symmetry Quotient Layer

Replace explicit quotient construction by a differentiable projection that removes learned group-orbit directions from both source and target features. The paper's reduction argument shows that a closed level constraint makes the restricted form horizontal, so the network can operate on invariant coordinates while retaining a measurable residual for symmetry leakage.

Useful6/10
Difficulty6/10
Novelty5/10
Paper: Reduction of relative multisymplectic manifolds arXiv:2607.12350
Unverified 2026

Contour Levy-Area Feature Branch

Add a geometric branch that converts ordered image contours into truncated signatures and feeds first-order displacement and second-order antisymmetric area features into the detector backbone. The area channel captures orientation and enclosed-region structure that ordinary edge magnitude or convolutional filters may miss, making the module suitable for thin cracks, scratches, bent boundaries, and small industrial defects.

Useful6/10
Difficulty4/10
Novelty6/10
Paper: Rough Path Signature-Guided Geometry Augmentation for Few-Shot Industrial Surface Defect Detection arXiv:2607.12245
Unverified 2026

Crystal Parenthesis Expert Balancer

Use the paper's crystal parenthesis matching as a deterministic local search over a sequence of MoE expert assignments. For adjacent experts i and i+1, pair assignments according to the crystal rule and move one unpaired i to i+1 only when this reduces expert-load imbalance. This provides a structured alternative to relying exclusively on an auxiliary load-balancing loss.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Contractions and applications of crystal skeletons: Young quasisymmetric and Stanley symmetric functions arXiv:2607.12232
Unverified 2026

Reflected-random-walk expert ecology

Turn a sparse expert layer into a stochastic birth-death population. Each expert receives a bounded fitness score from recent routed-token performance; at each update, a candidate expert is activated with probability p, while one expert is removed with probability q = 1 - p, preferentially removing the lowest-fitness expert. The paper's critical threshold f_c = q/p predicts which fitness levels can maintain a growing surviving population, providing a principled control knob for expert turnover.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Models for species evolution with random deaths arXiv:2607.12061
Unverified 2026

Decoration-Iteration Graph Coarsening

Construct a graph-neural layer that analytically eliminates fast auxiliary nodes inside repeated decorated motifs and replaces each motif by an effective edge or hyperedge. The effective interaction is computed from the log-partition function of the eliminated variables, while a residual neural correction can model violations of the assumed local motif structure.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Thermal phase transitions in a mixed-spin Ising model on the Lieb lattice: Exact results beyond zero magnetic field arXiv:2607.11661
Unverified 2026

Algebraic-Invariant Residual Layer

Represent a rational-like feature transformation with an auxiliary state y constrained by polynomial equations G(x,y)=0, and update x and y jointly along the tangent space of that constraint manifold. This creates residual blocks in which nonlinear feature identities remain consistent over many layers or time steps, reducing auxiliary-variable drift and potentially stabilizing rational activations and implicit recurrent dynamics.

Useful6/10
Difficulty5/10
Novelty8/10
Paper: Algebraic Invariant Quadratization Schemes for Cahn--Hilliard Equations arXiv:2607.11569
Unverified 2026

Log-Fractional Scale-Derivative Layer

Add a feature transformation that approximates the derivative of a fractional diffusion operator with respect to its order. Instead of only smoothing features with one fractional order, the layer exposes whether a feature changes rapidly across spatial scales, which can help with textures, edges, and multiscale patterns.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Regularity for the fractional logarithmic $p$-Laplacian arXiv:2607.11462
Unverified 2026

Boundary-Only Cell-Complex Network

Parameterize a cell-complex neural network by features on p-cells and derive lower-dimensional boundary features using the cellular boundary map over F2. For a 2D square complex, neighboring plaquette bits determine each link feature through XOR, reproducing the paper's exact gauge-law reconstruction and preventing the network from representing inconsistent open boundary configurations.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: $p$-Form Gauge Dynamics and Digital Quantum Simulation -- Flux and Cosmological Constant Neutralization arXiv:2607.10950