Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Restrict a fine-tuning adapter or output head to the subspace invariant under a prescribed monodromy, analogous to the paper's unbroken flavor lattice. This removes update directions intentionally changed by the domain-loop transformation, producing a parameter-efficient adapter with an explicit algebraic constraint.
Insert a fixed reversible lattice shear into a residual network so successive blocks follow a structured monodromy orbit rather than using unrelated learned transformations. Apply the transformation to a small learned subspace of hidden channels while leaving the remaining channels unchanged. This creates deterministic phase-dependent feature mixing with no additional trainable parameters.
Apply a weighted Caffarelli–Kohn–Nirenberg deficit to selected intermediate feature channels. The penalty discourages features that obtain large weighted responses only by becoming sharply localized or highly sensitive to small input perturbations. It can be evaluated with input-Jacobian estimates and added to the ordinary task loss.
Replace ordinary isotropic residual noise in a normalized continuous-depth block with projected Brownian forcing on the unit sphere. Apply a shared random symmetric quadratic drift to all tokens, plus a small token-specific tangent perturbation; the shared term preserves structured antipodal dynamics while the independent term removes persistent symmetry and cluster degeneracy. This is intended as a controlled stochastic regularizer, not merely additive Gaussian noise.
Group neural features into small Hermitian matrix elements and scale each group with the paper's tracial spectral Lp norm rather than independently normalizing scalar channels. This introduces a coupled spectral geometry while remaining implementable with ordinary eigendecompositions in the associative Hermitian-matrix special case.
Use the flat-torus covariance bound as a representation regularizer that controls the largest covariance eigenvalue while maintaining a prescribed total variance. This creates a directional anti-collapse constraint rather than only a scalar variance penalty, and can be applied to encoder outputs, VAE latents, or Transformer sequence representations.
Encode each scalar or discrete code t by the strictly convex lift γ(t)=(t,t²), optionally followed by a learned linear projection and normalization. Because three-code sums on this curve have only near-minimal additive energy, the representation should produce fewer collisions when a model composes three tokens, codes, or retrieved items by addition.
Regularize a spatiotemporal neural model with spectral penalties corresponding to several temporal-spatial scaling laws rather than using a single isotropic smoothness penalty. The model can remain spatially detailed while suppressing temporal oscillations, or learn the opposite preference when the data demand it.
Replace an unconstrained geometric multiscale codebook by features generated from a finite digit set and a Pisot scale factor. The contracting algebraic-conjugate directions should suppress near-collisions between representations at different scales, producing a discretely separated hierarchy that can be used for embeddings, recurrent memory, or quantized transformer states.
Whiten intermediate feature vectors and constrain several gauge moments to remain in the dimension-dependent interval predicted by the paper's Gaussian/log-concave comparison. Apply the penalty only to moderate orders, where the paper gives a uniform bound independent of the particular log-concave distribution; this should suppress heavy activation tails without forcing all features to be exactly Gaussian.
Add a learnable orthogonal rotation to a hidden representation and train it to make every channel projection have a small ψ2/L2 ratio. Unlike variance normalization, this explicitly suppresses directions with unusually heavy empirical tails while preserving the total quadratic energy of the representation.
Build a differentiable assignment layer whose rows represent tokens and whose columns represent experts, memory slots, or attention slots. Each row has unit probability mass, but no column receives positive mass from two rows; maintaining at least one vacant column makes assignments continuously deformable through elementary vacancy moves instead of abrupt softmax switches.
Add a measurement-conditioned attention layer with two explicitly separated fields: a geometry-only inverse-temperature profile that controls interaction strength and an outcome-dependent chemical-potential bias. For a region bounded by coordinates a and b, force the interaction gate to vanish as the square root of the distance from either boundary, while allowing a separate potential channel to encode measured values.
Replace a dense channel-mixing matrix in a sequence layer with alternating diagonal propagation and sparse unipotent Stokes jumps. The diagonal part carries independently controlled exponential phases, while the unipotent factors implement cheap residual-like mode conversion without changing determinant or requiring a dense matrix multiply. Constrain the phase magnitudes and jump coefficients during training to obtain a reversible, norm-monitorable mixer.
Replace a softmax transition or mixture-of-experts router by probabilities obtained from squared amplitudes of an isometric latent transition. Each input state is mapped to an orthogonal latent subspace, and summing probability over the latent index produces the desired expert or next-state distribution. The latent amplitudes can retain information that would be destroyed by directly averaging expert outputs, while normalization is guaranteed by construction.
Build a parameter-free spectral channel mixer whose channels are arranged as components of an l-form and whose multiplier is the trace-free Beurling--Ahlfors transform. At every nonzero spatial frequency it mixes the exact and coexact channel subspaces with opposite signs, preventing a uniform channel-direction bias and preserving a structured cancellation property. Insert it as a residual branch before a convolution, MLP, or attention block, with one learned scalar gate controlling its…
Replace a large flat positional-embedding table with a recursively decoded nine-way address whose child transformations contract coordinates by exactly 1/3. Encode an input position using features attached to the address prefix at several depths, guaranteeing that increasing depth produces a geometrically localized representation and that an infinite valid address cannot ambiguously represent two distinct points. This is especially suitable for 2D vision tokens, maps, point clouds, or…
Add a low-dimensional spectral regularizer to an encoder or transformer representation by estimating the first N nonconstant modes of its Gaussian-weighted diffusion operator. Penalize excessive reciprocal spectral mass and unequal low-frequency eigenvalues, using a Gaussian-ball reference calibrated to the representation's effective mass; this discourages latent directions from becoming weak, collapsed, or strongly anisotropic.
Replace raw hyperbolic embedding-radius regularization with a dimension-aware effective-radius target. For embeddings concentrated near hyperbolic radius rho in an n-dimensional hyperbolic space, regulate s times log(sinh(rho) / sqrt(n)) rather than rho itself, and use the same quantity to calibrate distance-logit temperature. This should make hyperbolic metric-learning behavior more invariant when embedding dimension, curvature, or model scale changes.
Regularize the hidden-state trajectory of a sequence model so that the distance between states at positions i and j follows a controlled power-law profile in |i-j|. This explicitly prevents representation collapse over long contexts while avoiding the requirement that all distant states be maximally separated. Use alpha as a tunable geometry parameter and compare alpha against the effective hidden dimension using the paper's Euclidean realizability threshold.
Replace raw polynomial interactions between neighboring feature vectors with central polynomial interactions computed after subtracting the local feature mean. Keep separate second-, third-, and fourth-order channels and apply independent residual gates to them, so a uniform shift of every feature in a neighborhood cannot create artificial cross-order responses. This is a drop-in higher-order mixer for a small transformer or graph neural network.
Represent a sequence of hidden states as points on a Riemannian sphere and penalize discrete geodesic curvature rather than merely penalizing adjacent-state differences. The regularizer discourages sharp bends in representation trajectories while remaining comparatively insensitive to uniform traversal speed, making it suitable for transformer depth trajectories or diffusion denoising paths.
Represent sequence positions as vertices on a circle and use a maximal family of mutually non-crossing d-arcs to define the allowed attention interactions. Rotate the family by d positions, or use several phase-shifted families across successive layers, so each layer has only O(N) edges but repeated layers propagate information over long distances. The geometric compatibility rule replaces arbitrary local-window or hand-designed sparse masks with a structured family whose maximality gives broad…
Add a hypergraph p-Laplacian penalty to hidden representations of samples or tokens grouped by a known relation, such as augmentations of one image, mentions of one entity, or tokens in one retrieved semantic cluster. Unlike mean pairwise smoothing, the penalty targets the maximum weighted discrepancy within each hyperedge, preventing a single representation from becoming an outlier while allowing moderate variation among the remaining members.