Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Treat local neural-network training as a driven linear system and periodically modulate the learning rate by a small sinusoid. Estimate the transfer function from this modulation to loss or gradient observables, fit its relaxation poles, and set the learning rate below the measured instability boundary.
Approximate the minibatch loss Hessian by a positive-semidefinite bulk curvature plus a small signed transverse correction, and treat only the correction with explicit negative-curvature steps. This imports the paper's observation that all unstable directions can be confined to a low-dimensional subspace, producing a curvature-aware optimizer whose step-size boundary is governed by a small matrix rather than the full Hessian.
Replace a fixed optimizer memory order with a nested family of gradient-integral controllers. Training begins with a first-order update and activates additional accumulated-gradient states only after an exponentially smoothed residual fails to decrease for several decision intervals; newly activated gains are ramped from zero, so the parameter update remains continuous and previously learned states are preserved. The optimizer should use little memory on easy problems and acquire longer memory…
Insert a constraint-reduction layer between a policy network and its executed action. The policy proposes an action, while the layer retains only geometrically extreme collision and obstacle constraints and verifies that every discarded halfspace is implied by the retained ones through nonnegative conic multipliers. The reduced projection or quadratic program is therefore equivalent to the full tightened safety filter whenever certification succeeds, but uses substantially fewer constraints.
Construct a residual network from independently attachable modules, but permit only a number of modules whose aggregate feedback gain lies inside a delay-dependent admissible interval. Estimate deployed end-to-end latency and each module's local Jacobian gain, then reject or bypass additional modules when the predicted delayed-loop stability boundary is crossed. This turns variable-width or depth scaling into a falsifiable control problem rather than an empirical choice.
Replace the dense hidden-state trajectory of a continuous-depth or recurrent neural block by a rank-r factorization F(t) = X(t) S(t) V(t)^T, and evolve the factors with a reversible projector-splitting integrator. During backpropagation, reconstruct earlier hidden states by reversing the factor updates rather than storing all activations.
Model one period of a cyclic optimizer or periodically modulated recurrent network as a discrete-time linear time-periodic system obtained by linearizing the update around its current trajectory. Estimate a periodic Lyapunov matrix sequence and scale the next learning-rate or modulation amplitude so that every phase contracts according to a certified energy decrease. This should prevent delayed divergence caused by resonance with the schedule, even when individual phase Jacobians are…
Turn an iterative optimization or equilibrium computation inside a neural network into a differentiable layer whose backward pass solves the implicit adjoint system with conjugate gradients or GMRES using only automatic-differentiation matrix-vector products. This avoids storing unrolled iterations and avoids explicit Hessian or Jacobian construction, enabling longer solver horizons and lower-memory implicit architectures.
Learn a low-dimensional Koopman operator from successive states of an iterative neural system, such as debate agents, recurrent refinement blocks, or diffusion denoising trajectories. Use the magnitude of the subdominant eigenvalue to predict the remaining number of rounds required for disagreement to fall below a target tolerance, and stop computation when the predicted deadline is reached rather than using a fixed round budget.
Replace an instantaneous diagonal optimizer with a causal convolution of recent gradients, where cross-layer or cross-module gradient correlations define a finite-memory Onsager response matrix. Estimate the response at several parameter-block pairs and lags, integrate it to obtain a finite-time transport matrix, and use its regularized inverse or symmetric part to precondition the update. This targets optimization regimes in which gradients propagate between blocks with measurable delay, such…
Use fast memory as read-only scratch state during the internal pondering iterations of a recurrent block, and apply memory writes only after the latent computation has halted or crossed a write gate. This prevents the transition operator from changing while it is being iterated, reducing self-corruption of the evidence used for subsequent reasoning.
Partition graph nodes into backward-equivalent classes and run message passing on the K-node quotient graph instead of the original N-node graph. If every node in a class receives the same aggregate message from every source class and shares the same local update map, class-constant node representations remain class-constant at every layer, making the quotient computation exactly equivalent to the full GNN on that invariant subspace.
Model a multi-timescale optimizer as a controlled dynamical system and use several Lyapunov-like quantities to regulate loss, momentum energy, and constraint violation simultaneously. The explicit high-order control-Lyapunov feedback becomes a low-cost correction to an SGD-momentum or Adam step. A Hurwitz comparison matrix supplies a measurable stability certificate and predicts the decay rate of the controlled training dynamics.
Replace an EKF or a large particle ensemble inside a neural world model with a fixed-order polynomial chaos representation of the latent state distribution. The transition network is evaluated under quadrature or sampled chaos variables, and Galerkin projection produces the next uncertainty coefficients directly; a coefficient-wise LMMSE update then assimilates observations without backpropagating through resampling.
Replace recurrence or nearest-neighbour analogue lookup with a learned delay-coordinate observer that continuously corrects a latent state using the current observation. Constrain the observer's closed-loop Jacobian or linear state matrix to have spectral radius below one, so prediction error contracts geometrically and required burn-in grows logarithmically with target accuracy.
Replace uniform collocation for a fixed random-feature neural PDE solver with sampling from the leverage-score density of the operator-applied features. Whiten the retained residual feature space before solving for output coefficients, so the sampled least-squares matrix has an identity-like expected Gram rather than inheriting severe anisotropy from the differential operator. The same construction can be used for a linearized neural network by treating Jacobian features as the trial functions.
Replace a single preconditioner with a finite set of stable update operators and switch between them during training to rotate optimization error into directions that later operators remove quickly. The controller should choose a small number of hard switches, including occasional use of a seemingly slower or less aggressive preconditioner, rather than averaging all optimizers at every step.
For a recurrent or graph neural network with known local connectivity, estimate each node's local Jacobian row using only graph neighbors rather than all hidden coordinates. Use the resulting sparse Jacobian both to compute a contraction certificate and to regularize training toward dynamically local interactions, reducing estimator variance and the number of samples required for reliable stability decisions.
Replace simultaneous descent-ascent on a bilinear adversarial subproblem by an implicit midpoint step. The update is a Cayley transform of the skew-symmetric game Jacobian, so it rotates rather than amplifies oscillatory modes and remains bounded for arbitrarily large positive step sizes in the exact bilinear case.
Replace the standard squared pointwise PDE residual in an elliptic PINN by its discrete $H^{-1}$ norm. The residual is passed through an inverse Dirichlet Laplacian, reducing the dominance of rapidly varying residual modes and acting as a mathematically specified preconditioner for the PINN training gradients.
Replace a full neural-network Gauss–Newton solve with a damped solve in an adaptively constructed low-dimensional parameter subspace. The subspace contains the current gradient, recent accepted updates, Krylov curvature directions, and randomized Jacobian-curvature probes, and is enlarged whenever its projected gradient fails to capture enough descent information.
Use the behavior-subspace gap as a trust-region constraint when applying a shared update to multiple recurrent modules or experts. A proposed common update is accepted only when post-update behavior subspaces remain close to their leader and their graph subspaces remain sufficiently transverse, preventing one shared optimizer step from destabilizing dynamically different members.
When a model, dataset, or parameterization has an involutive symmetry, transform gradients and curvature into symmetry sectors before applying adaptive updates. The Hessian and optimizer can then be handled as independent even and odd blocks, preventing curvature from one sector from forcing an unnecessarily conservative learning rate in the other.
Replace a full Hermitian curvature matrix, such as a Hessian or empirical Fisher matrix, by its block-diagonal version only when the paper's perturbation certificate predicts a small eigenvalue change. Use the certificate online to merge poorly separated blocks and retain independent preconditioners for well-separated blocks, yielding a controllable accuracy-memory tradeoff rather than a fixed block-diagonal approximation.