Architecture ideas

Research ideas extracted from mathematics papers, categorized as Architecture.

Unverified 2026

Spectrally screened polynomial pseudoinverse layer

Replace an SVD-based pseudoinverse of a learned rectangular matrix with a low-degree polynomial initialization followed by a few Newton–Schulz iterations. The polynomial approximates the inverse Gram operator, while a cheap residual test accepts it only when the iteration is contractive and otherwise selects a conservative transpose-scaled initialization.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Iterative Methods for Computing the Moore--Penrose Inverse of Split-Quaternion Matrices with Applications arXiv:2607.29270
Unverified 2026

Pullback-Commuting 3-Axis Network

Use three learned state-transition operators corresponding to three data axes, and train them to satisfy the paper's pullback-style interchange rule. For every local pair of axes, two successive updates should reach the same square state; for triples of axes, all six update orders should agree. This reduces sensitivity to scan direction and limits long-horizon drift caused by inconsistent local transitions.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Higher-Dimensional Symbolic Dynamics: A Textile Framework For 3-graphs arXiv:2607.29233
Unverified 2026

Landmark Distance-Profile Adapter

Add a metric-aware front end that represents an arbitrary object x by its distances to a fixed set of reference objects rather than forcing x into a Euclidean or Hilbert embedding. Feed the resulting profile through a learned projection and concatenate it with the ordinary neural representation. This should be useful for graphs, trees, distributions, and sets where generic vectorization loses geometry or requires an expensive object-specific encoder.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Distance Profile Embedding for Independence and Conditional Independence Testing of Random Objects arXiv:2607.28981
Unverified 2026

Null-form quadratic wave layer

Replace an unconstrained quadratic interaction between channel derivatives with a learnable combination of Lorentzian and antisymmetric null forms. For wave-equation surrogates, this enforces exact cancellation when two interacting features have parallel null directions, suppressing resonant derivative products that otherwise cause unstable long-horizon rollouts.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Recovery of a Null Form in the Wave Equation from Scattering Data arXiv:2607.28917
Unverified 2026

Demographic Synchronizing Expert Layer

Replace static mixture-of-experts routing weights with positive expert abundances that undergo phase-dependent birth, death, and crowding. Each expert has an internal phase and natural frequency; experts aligned with the population order parameter receive larger effective abundance, while a logarithmic penalty prevents runaway replication. The mechanism creates a measurable synchronization transition and can serve as a differentiable alternative to hard top-k routing.

Useful6/10
Difficulty6/10
Novelty8/10
Paper: Synchrony by Birth and Death arXiv:2607.28867
Unverified 2026

Random Tree Feature Layer

Generate many random symmetric decision trees and encode each input by the one-hot indicator of its reached leaf. Use the resulting fixed random feature vector as an additional input to an MLP, or train only a ridge/linear prediction head on it. The tree ensemble's Gaussian-process-limit interpretation predicts that increasing the number of independent trees should approximate a stable kernel while avoiding MCMC and difficult optimization over discrete split structures.

Useful6/10
Difficulty4/10
Novelty7/10
Paper: Seeing the Forest for the Trees: The Gaussian Process Limit of BART arXiv:2607.28844
Unverified 2026

Group-Lifted Message Passing

Turn a small base message-passing graph into a family of larger graphs by replicating every base node across a finite fiber and wiring replicas with permutations derived from a group extension. Use one shared local neural update on every lifted copy, so parameter count stays that of the base graph while the lifted graph supplies additional global paths and larger effective receptive fields.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Lifting Lifted Product Codes arXiv:2607.28621
Unverified 2026

Geometrically Random Transport Network

Construct a deep sequence model as a layered channel network with fixed random K-regular connections between neighboring depth layers, instead of dense or independently random weight matrices. Use norm-preserving edge normalization and a reversible residual update so that geometric randomness controls information transport while trainable nonlinear readouts provide task-specific computation. The architecture exposes a tunable crossover between quasi-one-dimensional ballistic or localized…

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Quantum Chaos and Diffusive Transport from Geometric Randomness arXiv:2607.28579
Unverified 2026

Decision-Diagram Convex Hull for Neural Gates

Replace independent linearizations of binary neural-network gates and their higher-order interactions with a compact decision-diagram flow formulation. This preserves the exact convex hull of feasible gate configurations whenever the DD is exact, making MILP-based pruning, quantization, or architecture search substantially less vulnerable to fractional gate solutions.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Beyond Hand-Derived Inequalities: Decision Diagrams for Cut Generation in Binary Polynomial Optimization arXiv:2607.28511
Unverified 2026

Normalized Scheduling-Degree Truncation

Use normalized scheduling variables and explicitly cap the degree of their products in a neural LPV or mixture-of-dynamics model. Instead of allowing every multiplicative interaction between scheduling coordinates and past or future features, retain only monomials below a chosen degree threshold. This produces a controllable approximation knob between a purely linear model and a full lifted predictor, while avoiding unstable extrapolation caused by poorly scaled high-degree features.

Useful6/10
Difficulty5/10
Novelty5/10
Paper: A subspace approach to data-driven predictive control for linear parameter-varying systems arXiv:2607.28490
Unverified 2026

Distance-Powered Boundary Ansatz for Fractional PINNs

Represent the predicted solution as $u_{\theta}(x)={\rm d}_{\Omega}(x)^s v_{\theta}(x)$, where $v_{\theta}$ is an unconstrained neural network and ${\rm d}_{\Omega}$ is the distance to the boundary. This builds the fractional Dirichlet boundary layer into the architecture and leaves the network to learn the smoother quotient $u/{\rm d}_{\Omega}^{s}$, which the paper proves extends Hölder-continuously to the boundary when the reaction has sufficient integrability.

Useful6/10
Difficulty4/10
Novelty6/10
Paper: On boundary regularity for the fractional p-Laplacian with unbounded reactions arXiv:2607.28436
Unverified 2026

Caret-Invariant Tree-Pair Mixer

Represent a token sequence or image region partition by a pair of planar rooted binary trees and a permutation of their leaves, following the tree-pair model of Thompson-type groups. Train the mixer to produce the same output whether a leaf is left intact or refined into two child leaves that are immediately merged by a matching inverse refinement. This creates a variable-resolution hierarchical module with an explicit consistency constraint instead of relying on an arbitrary fixed tree depth.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Thompson's Group $V$ and Virtual Link Theory arXiv:2607.28406
Unverified 2026

Coherence-to-Diffusion Graph Layer

Replace a single graph or token-mixing operator with two coupled channels: an antisymmetric, coherence-preserving transport channel and a state-dependent dissipative diffusion channel. The local feature state controls the dissipative edge rates, so strongly occupied or conflicting regions are smoothed while weakly interacting regions retain rapid coherent propagation.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Interacting Quantum Symmetric Exclusion Process arXiv:2607.28255
Unverified 2026

Entropy-Maximized WL Anchor Selection

Replace random or degree-based graph positional anchors with a greedy set chosen to maximize the entropy of the graph's Weisfeiler-Lehman equivalence partition. Give each selected anchor a unique trainable embedding, then run several WL-inspired refinement rounds to produce positional cell features. The resulting positional signal should distinguish nodes that ordinary message passing treats identically while using only a small anchor budget.

Useful6/10
Difficulty4/10
Novelty5/10
Paper: On Leader Selection for Strong Structural Controllability in Matrix-Weighted Networks arXiv:2607.28168
Unverified 2026

Interval-Lifetime Feature Modules

Represent a multiparameter feature stream as a direct sum of interval modules over the nonempty-subset lattice of several corruption, scale, or context variables. Each learned channel is active only on a connected order-convex region of the lattice, replacing a dense bank of unrelated per-condition adapters with reusable feature channels having explicit parameter-space lifetimes.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Interval Decompositions for Multipersistence Modules over Finite Posets and Robustness of Sheaf Data on Simplicial Complexes arXiv:2607.28134
Unverified 2026

Layered Structural Reachability for Neural States

Treat the hidden-state Jacobian of an RNN, SSM, or graph neural network as a directed matrix-weighted network and decompose repeated block couplings into scalar interaction layers. Use layer-specific structural controllability to select input, skip, reset, or readout channels that can reach all hidden dimensions, and reject architectures with structurally unreachable states before training.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: On the Strong Structural Controllability of Matrix-Weighted Networks arXiv:2607.27852
Unverified 2026

Screened Disordered Mixing Layer

Replace a dense token or state-mixing matrix with an inverse-capacitance operator whose couplings decay with graph distance, while introducing trainable heterogeneous diagonal capacitances to break spatial symmetries. The layer is cheap because the capacitance matrix is sparse and banded, but its inverse produces global responses with controllable locality.

Useful6/10
Difficulty6/10
Novelty6/10
Paper: Nanoparticle Networks for Neuromorphic Computing arXiv:2607.27844
Unverified 2026

RIP-Circulant Sparse Projection Layer

Replace a dense Gaussian or learned projection from dimension N to m with a normalized partial circulant projection generated by a single Gaussian vector. For K-sparse hidden states, the restricted-isometry guarantee predicts approximate norm preservation while reducing stored projection parameters from O(mN) to O(N). The projection can be evaluated with an FFT and should be combined with explicit top-k gating so that the sparse-input assumption is enforced.

Useful6/10
Difficulty4/10
Novelty5/10
Paper: Improved RIP Bounds for Gaussian Partial Circulant Matrices arXiv:2607.27676
Unverified 2026

Risk-cost efficient frontier routing

Build a sparse expert cascade whose router uses empirical progress, uncertainty, and compute cost to construct an efficient frontier of experts. Instead of always choosing the expert with the largest immediate gain per FLOP, route different confidence states through an ordered sequence of frontier experts, allowing cheap high-variance experts early and safer or more cost-effective experts near the final decision.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: Strategies for Milestone-driven Start-ups in Multi-activity Settings arXiv:2607.27563
Unverified 2026

Unique-witness two-hop attention

Replace direct source-to-target attention connectivity by two sparse incidence relations through a set of latent witness tokens. A source token attends only to a small set of witnesses, and each witness attends only to a small set of target tokens; the composed relation is trained to contain exactly one witness for desired pairs and no witnesses for undesired pairs. This produces a controllable sparse attention pattern whose errors can be measured entrywise against a dense teacher or known mask.

Useful6/10
Difficulty5/10
Novelty6/10
Paper: On $\varepsilon$-Matrix Product Factorization of graphs arXiv:2607.27407
Unverified 2026

Decoder branch witness regularizer

Apply the paper's mechanism-contrast idea to ReLU decoders by requiring each piecewise-affine branch to produce a detectable and distinctive change across at least one activation boundary. Penalize branches with vanishing Jacobian jumps or nearly identical boundary signatures, discouraging observationally interchangeable decoder mechanisms.

Useful6/10
Difficulty6/10
Novelty8/10
Paper: Beyond ICA: Identifiability by Symmetry Breaking arXiv:2607.23182
Unverified 2026

Structure-Preserving Profile Layer

Replace unconstrained output coordinates with a neural parameterization whose outputs are valid monotone profiles by construction, analogous to representing a Young diagram through nonnegative ordered row increments. Train the network against an explicit energy or negative log-probability while preserving the feasible geometry, rather than relying on penalties that permit invalid intermediate states.

Useful6/10
Difficulty4/10
Novelty6/10
Paper: Neural variational framework for random Young-diagram limit shapes arXiv:2607.27061
Unverified 2026

Essential-Edge Harmonic Attention

Compute one-dimensional persistent homology and minimum-norm harmonic representatives, then use their absolute edge coefficients as topology-aware saliency in a graph transformer. Add the saliency to attention logits or use it as a soft regularizer so the model preferentially propagates information along edges that are essential to persistent cycles.

Useful6/10
Difficulty6/10
Novelty7/10
Paper: Essential Simplices Dominate in Harmonic Representatives of One-Dimensional Persistent Classes arXiv:2607.26378
Unverified 2026

Fourier Turing Recurrent Layer

Replace one spatial convolution block by a recurrent Fourier-domain layer that couples every mode k to its opposite mode -k and gives the strongest amplification to a nonzero selected wave number k*. The layer crosses a controlled Turing-like instability at k* and uses cubic saturation to produce bounded structured features instead of unbounded activation growth.

Useful6/10
Difficulty5/10
Novelty7/10
Paper: Quantum Turing Patterns arXiv:2607.26331