Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Build a sequence or graph module whose state is a canonical set of affine tropical pieces rather than an opaque hidden vector. Compose consecutive segments by Minkowski addition of their lifted supports, merge alternatives by union followed by lower-hull reduction, and evaluate the resulting piecewise-linear function with a minimum.
Replace uniform embedding dimensions with a globally budgeted allocation based on the estimated spectral complexity of each categorical feature. Tables whose category representations have large leading singular-value energy receive more dimensions, while high-cardinality tables are penalized because each extra dimension consumes more parameters.
Use Adam normally, but periodically estimate the spectrum of the Adam-preconditioned Hessian and add a damped low-rank Newton correction when the preconditioned curvature is strongly ill-conditioned or the gradient is concentrated in flat directions. The correction is computed only in a small Lanczos subspace, so the method targets cross-coupled ill-conditioning without materializing or inverting the full Hessian.
Replace an opaque MLP vector field with a stack of trainable symbolic primitives that can express linear terms, monomials, products, and related analytic operations. Apply an L1 penalty and prune small primitive coefficients after rollout training, yielding a compact dynamics module that is cheaper to evaluate and easier to inspect.
Propagate uncertain inputs or parameter perturbations through a neural network with zonotopes, and choose the reduction rule after each layer using short lookahead that predicts downstream certified-bound error. Candidate reducers can include largest-generator retention, box conversion, and norm-based merging; unlike a fixed policy, the chosen rule depends on the current generator geometry and remaining network depth. The goal is tighter robustness bounds at the same generator budget, or the…
Replace a Gaussian regression likelihood with a chart-pulled exponential-family head. The network predicts a natural parameter eta, while each target y is mapped to a bounded probability coordinate u=G(y). Because the sufficient statistic T(u) is bounded, extreme targets cannot create arbitrarily large parameter gradients. The chart determines the tail class, while eta controls the distribution in probability space.
For bandlimited or progressively higher-frequency targets, construct a nested random Fourier layer whose frequency window grows according to the paper's bandwidth laws instead of sampling all frequencies from one fixed range. Train with a low-bandwidth subset first, then activate additional frequency blocks. This creates a mathematically motivated spectral curriculum and may reduce early optimization difficulty while preserving high-frequency accuracy.
Replace a finite-dimensional latent prior for fields or trajectories with a continuous flow acting on function coefficients in a separable Hilbert space. A Gaussian reference with an explicit covariance spectrum is transformed by a learned vector field, so the model can represent smooth, non-isotropic function distributions while sharing parameters across discretization resolutions.
Replace a dense Gaussian random-feature layer by a sparse Poisson-chaos layer with the same learned kernel, and penalize the layer's normalized fourth add-one energy. The penalty suppresses rare single-atom jumps, so the sparse layer approaches the Gaussian same-kernel representation in Wasserstein distance while preserving computational advantages from sparse Poisson sampling.
Parameterize a scalar or vector implicit neural field on each spatial cell with Bernstein polynomials and constrain its coefficients instead of sampling many points to enforce output bounds. The Bernstein convex-hull property gives a deterministic pointwise bound everywhere in the cell, making the method useful for neural fields representing densities, concentrations, masks, or material parameters.
Insert a fixed expander channel before a covariance-dependent feature transformation. The channel repeatedly conjugates the feature covariance by a constant number of sparse Pauli/CNOT unitaries, preserving total feature energy while contracting anisotropic covariance components. Use the mixed covariance for whitening or as a regularized normalization statistic, and test whether it gives more stable training than dense whitening or an explicit isotropy penalty.
Add a multiscale Besov penalty to the output of a shallow ReLU^k network, targeting the smoothness threshold that the paper proves is sufficient for finite ridge-variation representation. This suppresses pathological high-frequency output while preserving low-frequency approximation, providing a principled alternative to ordinary parameter weight decay.
Insert a projection-space mixer that combines several fixed or learned directions using reciprocal correlations with the current feature, then normalize the result. The exact construction has a universal beta law for its squared input-output cosine, so it can create controlled angular diversity while remaining deterministic and independent of the chosen direction dictionary.
Use a sparse multivariate Legendre expansion as the geometry-to-network-weights map, rather than an unconstrained MLP that consumes all shape parameters. The hypernetwork predicts only coefficients for a selected set of polynomial multi-indices, allowing high-dimensional or countably parameterized shape uncertainty to be handled with a number of learned terms determined by coefficient decay rather than ambient dimension.
Replace an unconstrained hidden-to-hidden interaction in an MLP or transformer feed-forward block by two gauge-related branches. Split channels with an orthogonal involution Θ, constrain the learned interaction K to anticommute with Θ, and use opposite signs of K in paired branches. This creates a testable inductive bias in which the learned interaction only transfers information between the two channel subspaces.
Partition a neural network into heterogeneous parameter blocks or maintain several worker replicas, and model each block's optimizer state as a constrained linearized dynamical agent. At every synchronization interval, jointly optimize a finite sequence of parameter updates and a feasible common terminal parameter target, while enforcing consensus through distributed primal-dual iterations. Unlike ordinary gradient descent toward a fixed or implicit target, the target is selected together with…
Parameterize an orthogonal or semi-orthogonal neural weight matrix directly on the Stiefel manifold and update it with a Cayley retraction instead of unconstrained SGD plus a penalty or QR projection. The update preserves orthogonality exactly, is second-order accurate for the appropriate metric, and avoids the cubic QR factorization at every optimizer step.
Replace a neural controller's pointwise action outputs over a finite horizon with Bernstein control points whose convex hull satisfies actuator and trajectory constraints. The network predicts the control points, while a robust margin accounts for bounded tracking or model-prediction error, making continuous-time actuator feasibility checkable from finitely many inequalities.
Construct a generative or recurrent neural architecture with several contractive or mildly expanding branches, and explicitly control the geometric complexity of its invariant set using the sub-additive singular-value pressure of branch-Jacobian products. Instead of regularizing only the operator norm, the model can preserve anisotropic directions while targeting a desired attractor dimension, potentially improving coverage of structured data without uncontrolled folding or collapse.
For a neural network predicting A coupled concentration or density fields, decompose Fourier-space fluctuations into a charge direction and its charge-neutral composition subspace. Hard-project the predicted fields to eliminate the global charge mode, and regularize their low-wavenumber covariance so that neutral modes retain finite susceptibility while the charge structure factor follows the Coulombic suppression S_ZZ(k) proportional to k squared. This should improve long-range physical…
Represent binary nuisance variables or augmentation bits as coordinates of a Hamming cube and penalize the model response that remains predictable from a random k-coordinate subset. Use the theorem's derivative-plus-global-norm certificate as the regularizer, retaining its p/log p dependence instead of using an arbitrary masking penalty. Random coordinate permutations provide a cheap stochastic approximation to the subset average.
Build actuator or parameter constraints directly into the neural controller using a differentiable hard-saturation map rather than penalizing violations after the fact. This makes the Lyapunov certificate apply to the actual bounded controller and prevents training from exploiting unrealistically large actions.
Parameterize a complex linear layer as a product of sparse triangular network factors whose positive modulus version is totally nonnegative. The layer can use phase cancellation for expressive transformations, while selected minors remain bounded by explicitly computable positive minors, giving a structured alternative to unconstrained dense complex weights.
Replace the single backpropagated subgradient of a piecewise-smooth network loss by a minimum-norm convex combination of gradients evaluated at nearby parameter perturbations. Shrink the perturbation radius geometrically and restart the schedule when the sampled Goldstein direction becomes small, following the paper's INGD motivation.