Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Use the paper's three-periodic linear-gradient construction as a cheap online detector for Adam's failure mode. When recent gradients exhibit the pattern \((c,-1,-1)\), with \(c>2\), and the adaptive update repeatedly moves in a harmful direction, freeze Adam's normalization and use a short SGD or AMSGrad fallback before returning to Adam.
Replace all pairwise consistency comparisons between m augmented views by a single group-valued relative-difference vector with m−1 components. Add a learned contractible-chart penalty so that the relative-difference map remains locally simple rather than merely numerically small. The construction is invariant to simultaneous left multiplication of every view, providing a useful gauge-invariant consistency signal.
Add a fractional oscillation penalty to scalar functions produced by a neural network on an ordered grid. Unlike a derivative penalty, this remains meaningful for nonsmooth or nowhere-differentiable outputs and interpolates between total-variation-like behavior and Sobolev-like smoothness.
Insert a rearrangement operation on scalar feature maps sampled along an ordered coordinate such as time, spatial position, or a neural-field input grid. The operation sorts values into non-increasing order, preserving the empirical histogram exactly while provably not increasing the Riesz fractional variation in the ideal one-dimensional continuous setting.
Insert a piecewise Möbius transformation as a deterministic latent mixing layer, using the paper's exact branch structure rather than a generic unconstrained MLP. The transformation repeatedly moves points between branches while preserving a known reference density, creating a cheap chaotic mixer with analytically computable Jacobian factors. Use a truncated, normalized version in practice so that the sigma-finite invariant measure becomes a valid finite training distribution.
Train a neural function under hard pointwise constraints by projecting its desired output-space update into the Clarke tangent cone of the admissible set at every sampled input. Fit the resulting feasible measurable direction with a parameter update instead of repeatedly allowing the network to violate constraints and repairing it with a penalty.
Split the trainable state into an explicit scalar scale coordinate and a residual perturbation, then update them with separate time scales. Penalize residuals according to their distance from the scale-dependent core, so the optimizer cannot obtain apparent progress by destabilizing the scale mode. The method is a neural optimization analogue of the paper's modulation argument, not a direct consequence of the geometric singularity theorem.
Use the logarithmic exhaustion as a geometry for bounded hidden representations rather than only as a parameter constraint. A representation approaching the boundary receives an increasingly large metric, making ordinary Euclidean motion expensive and discouraging brittle saturation while preserving a bounded intrinsic gradient for the boundary coordinate.
Treat each directed attention matrix as a graph transition matrix and form its Laplacian L = I - A. Compute the principal-cofactor vector to identify tokens with weak global access to the rest of the layer, and regularize the nonzero-eigenvalue product so attention does not become reducible or nearly singular. This targets pathological attention heads that isolate token groups and produce unstable or poorly propagated representations.
Replace ordinary dot-product attention logits with a strictly totally positive kernel evaluated on positive, ordered scalar coordinates attached to queries and keys. Use the modified-Bessel kernel K(x,s)=I_s(x), whose every ordered minor is positive, then row-normalize it as an attention matrix. This creates an attention operator with a mathematically enforced anti-oscillatory structure rather than merely positive entries.
Build an attention or positive-mixture module whose output ratio at two control settings is provably monotone in an ordered index such as token distance, retrieval rank, or discretized uncertainty. Use normalized-positive-series identities to replace an unstable quotient derivative with a difference of expectations, and penalize violations of the resulting stochastic-order condition during training.
Construct a robust central region of each class or domain embedding cloud by intersecting halfspaces whose discarded cap mass is at most a prescribed fraction. Use this floating-body region to define prototypes or consistency targets, suppressing one-sided outliers without assuming Gaussian covariance structure. The centerpoint level 1/(d+1) provides a principled default depth parameter.
Represent the prediction as a sum of a smooth interior branch and a fractional boundary branch: u_theta(x)=u_int_theta(x)+d(x)^a u_bd_theta(x). This mirrors the paper's direct-sum solution structure and allocates separate network capacity to the globally regular component and the boundary layer.
Calibrate the maximum attention logit in each head against the log-correlated extreme-value law instead of applying fixed clipping or a fixed max-norm penalty. Penalize only maxima that exceed the predicted log N minus three-quarter log log N baseline by an unusually large order-one fluctuation, allowing ordinary sharp attention while suppressing rare pathological spikes.
Add a Vandermonde conditioning objective to a mixture-of-experts router so that experts acquire distinct scalar routing signatures instead of collapsing onto the same score region. The regularizer uses powers of one learned scalar score and directly penalizes near-coincident expert scores, providing a finite-mode identifiability signal complementary to load balancing.
Apply the paper's sharp sumset lower bound to the active discrete supports of multiple additive branches in a sparse neural layer. Penalize cases where the support of the combined output is smaller than the mathematically guaranteed minimum implied by the branch supports, discouraging destructive overlap and representational collapse.
Insert a two-mode residual mixer whose mode is selected by a delayed sign variable rather than an instantaneous sign or sigmoid. The delayed mode creates a hysteresis-like effect that prevents high-frequency switching when the latent state is close to the decision surface, while the paper's reduced equations provide a constraint for choosing the delay and mixing strength so the latent energy contracts.
Estimate how often a representation lies on a separating hyperplane for alternative separable dichotomies, and use this quantity as a boundary-concentration penalty. Unlike a single classifier margin, the score measures whether many admissible separators consider the point ambiguous.
Construct a sparse attention or message-passing mask by sampling edges with preferential weights (d_u+alpha)(d_v+alpha), while keeping the edge count below the predicted connectivity threshold. This creates hub-like local communication patterns but prevents one giant component from forcing dense information mixing, reducing attention cost and potentially mitigating oversmoothing.
Use the Bernoulli corank asymptotic to choose sparsity for binary or sparse linear layers and reject initial matrices with excessive numerical rank deficiency. The layer should also explicitly prevent zero columns, because the paper's probability law indicates that zero-column events are a leading mechanism behind large corank in the sparse regime.
For two monotone prediction heads receiving binary features, penalize cases where their covariance is smaller than the sharp degree-weighted collision of their Fourier spectra. This discourages uncontrolled agreement on high-order interaction patterns while preserving low-order shared structure, and can be used either as a constraint or as a diagnostic for monotone multi-task models.
Add a differentiable penalty to a graph generator or graph predictor when its soft higher-order clique density violates the sharp lower bound implied by its lower-order clique density. The regularizer encourages generated graphs to have mathematically consistent motif statistics without hard-discretizing the predicted adjacency matrix.
Use linking number as a diagnostic and optional regularizer for representations of paired closed data manifolds. The probe identifies layers that collapse or separate class geometry through collisions and folds, giving an architecture-selection signal beyond loss and Jacobian singular values.
Prepend an adaptive Savitzky-Golay derivative bank to a temporal neural network. For each input channel and derivative order, select the local window by minimizing Stein's unbiased risk estimate, then concatenate the raw signal with the estimated derivatives. This supplies denoised velocity and acceleration features without requiring clean derivative targets or forcing the backbone to learn unstable finite-difference filters.