Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Replace count-only MoE load balancing with greedy balancing of aggregate token-feature vectors. A token is assigned to the expert for which adding its feature vector produces the smallest increase in that expert's squared aggregate norm, encouraging experts to receive complementary semantic mixtures rather than identical token counts.
Replace the ordinary update in a differentiable minimax game with a Halpern-anchored second-order operator step. The current game iterate is first corrected using the local Jacobian of the game gradient, and the corrected point is then contracted toward a fixed anchor with a decreasing Halpern weight. This is intended to reduce cycling in adversarial training while preserving the faster asymptotic behavior associated with second-order monotone-operator methods.
Grow a mixture-of-experts layer by splitting one expert into two children while conserving its routing mass, and choose the split ratio to minimize the worst imbalance over all intermediate expert counts. Use the paper's sharp threshold as a hard design target: with n experts, some intermediate stage must have capacity ratio at least D_n = 2^{1-1/\lceil n/2\rceil}, so schedules substantially below this are impossible rather than merely difficult to discover. Initialize child router logits with…
Replace a continuous allocation or routing decision with a lattice-valued decision whose unit size is explicitly normalized by total capacity. Round allocations downward rather than to the nearest lattice point, preserving per-example capacity feasibility, and train or evaluate against the resulting granularity ratio rather than treating discretization as an implementation detail.
Replace a Gaussian or point-estimate regression head with a heteroscedastic Student-t head whose scale and degrees of freedom depend on the learned state. This gives the model a principled way to absorb abrupt, nonmonotone events and operating-condition shifts without forcing the central degradation trend toward rare extreme residuals.
Replace the scalar learning rate of SGD or Adam's outer update by a blockwise Barzilai--Borwein estimate computed from consecutive parameters and gradients. Use gradient smoothing, denominator checks, and clipping so that the curvature estimate remains usable with stochastic neural-network gradients.
Replace backpropagated gradients in a Langevin sampler with a simultaneous-perturbation estimate obtained from two noisy evaluations of a neural energy or simulator-defined negative log-density. This enables posterior or latent-space sampling when the energy contains nondifferentiable code, stochastic simulation, discrete operations, or an inaccessible neural-network gradient.
Train a neural coefficient-recovery model with an additional loss that rewards observation sensitivity in every learnable coefficient direction. Instead of only minimizing the reconstruction error of the observed trajectory, explicitly discourage a nearly singular parameter-to-observation Jacobian, which should reduce ambiguous reconstructions and improve robustness to noise.
Add a small constrained equilibrium layer whose response depends on the current neural state and recent exogenous history, then cache responses keyed by a learned history embedding. For a new history, reuse a cached response only when an empirical Wasserstein distance to the cached history is below a threshold; otherwise run a few inner optimization iterations. The paper's local Holder and trajectory-stability results motivate graceful degradation rather than catastrophic errors for nearby…
Replace a fixed maximum-wait timeout in dynamic neural inference batching with a service-aware timer. The timer estimates the marginal reduction in computation or communication cost from adding another request and delays service only while that saving justifies the additional oldest-request delay.
Use the paper's stationarity-floor scale to set the SAM radius from a desired gradient tolerance, and reduce the radius when training approaches that tolerance. This turns an otherwise opaque SAM hyperparameter into a curvature- and accuracy-aware schedule.
Replace fixed-period federated averaging or distributed all-reduce with a Bernoulli communication decision whose probability is selected from estimated network connectivity and optimization conditioning. Local workers continue making corrected updates between communication events, while the contraction theorem exposes when communication is worth its cost.
Train a neural average-reward actor-critic that turns robustification on only when the estimated uncertainty scale σH₀ is comparable to or larger than the desired critic accuracy ε. In the high-tolerance regime use an ordinary nominal Bellman target; in the low-tolerance regime add a total-variation pessimism penalty proportional to the learned bias span. This avoids injecting a large robustness penalty when it is statistically unnecessary while retaining protection against transition…
Represent stochastic layer execution, branching, retries, and early exit as a finite continuous-time Markov chain, with the completed-prediction state absorbing. Learn transition rates jointly with neural-network weights, but use MFPT sensitivities to allocate rate changes according to their available control budget rather than allowing one routing edge to dominate halting-time control.
Add a supervised anchor that forces a conditional generative predictor to output the expected target when its noise input is set to the mean of the noise distribution. The model remains stochastic for nonzero noise, but its zero-noise trajectory becomes a stable estimate of the conditional mean, which should reduce rollout drift and make the learned transition easier to optimize.
Use the Bregman objective's exact residual-dependent curvature to build a positive-semidefinite Gauss-Newton preconditioner for a neural network's scalar regression head. Negative curvature weights are clipped or damped before solving the update, preserving the original gradient while preventing residual patterns from producing unstable parameter steps.
Replace globally backpropagated hidden-layer losses with independent recurrent layers that receive bottom-up, top-down, and lateral inputs. Train each layer to assign low activity to correctly paired input-label examples and high activity to mismatched examples, then classify by selecting the label with the lowest accumulated surprise.
Replace a fixed-batch SGD or proximal-gradient update by a stochastic proximal-subgradient step whose step size is backtracked against an empirical sufficient-decrease condition. If the condition is too noisy or repeatedly fails, enlarge the batch and retry; otherwise retain the current batch, allowing sample size to grow only when needed.
Replace uniform or purely loss-driven update allocation with a scheduler that targets both the mean update rate and the temporal variance of updates for each parameter group, task, or expert. At every training step, assign the available minibatch slots or accelerator workers to groups with the largest weighted deficits, preventing starvation while avoiding highly bursty update streams that can produce optimizer oscillations.
Insert a differentiable equilibrium layer between a neural payoff/state encoder and the final action recommendations. The layer parameterizes a joint recommendation object and enforces all unilateral-deviation inequalities as positive-semidefinite constraints, preventing the network from producing recommendations that agents have a strict incentive to disobey. A quantum-inspired density-matrix parameterization can model correlated recommendations using PSD matrices rather than factorized action…
Replace a fixed global learning-rate estimate in an accelerated optimizer with a curvature envelope that depends on the current estimated optimality gap. Use phase restarts and a descent backtracking test so that the method remains safe when the gap or \(H_1\) estimate is inaccurate. The expected benefit is faster progress on objectives whose curvature is large early in training but decreases substantially near a good solution.
Add a periodic coarse optimization phase to SGD or Adam that operates on a compressed parameterization and returns a prolongated correction to the full network. Retain nonsmooth constraints or regularizers explicitly through a primal-dual update instead of relying on penalty smoothing. Accept the correction only when it improves a cheap fine-batch merit test, making the method useful even when the coarse objective is only approximately coherent with the fine objective.
Use enumerated weighing matrices as sparse orthogonal channel-mixing operators inside MLPs or residual blocks. Their ternary entries reduce multiplication to signed additions, while exact orthogonality prevents amplification or attenuation of feature norms; a trainable fixed-support version can recover expressivity without giving up computational sparsity.
Train the critic on the action that the environment actually received after safety filtering, not only on the actor's nominal action. Prioritize transitions whose estimation residual, barrier proximity, or filter intervention is large, so replay concentrates on the distribution shift introduced by the safety controller instead of repeatedly sampling benign nominal behavior.