Every idea extracted from recent arXiv mathematics papers — verified and unverified. Click an idea to open its full card; badges show the empirical verdict.
Cluster recurrent modules or MoE experts by the geometry of their observed finite-horizon input-output behaviors rather than by parameter distance. Train one shared optimizer/controller or low-rank adapter per cluster while retaining module-specific parameters and routing. This should reduce control and optimizer overhead without merging modules whose temporal responses are dynamically incompatible.
Train a recurrent or neural state-space model on fixed-initial-state subsequences, but select the training horizon and burn-in from an empirically estimated turnpike bound instead of choosing them arbitrarily. If the cumulative discrepancy between fixed-initial-state and free-initial-state optima is bounded, the average discrepancy decreases as 1/N, allowing shorter windows while preserving the long-horizon optimum.
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.
Add an uncertainty-aware observation scheduler to a neural state-space model or recurrent world model. Between expensive observation-encoder updates, propagate the latent state using the learned dynamics; periodically compute a decimated Riccati prediction and choose the largest skip length whose predicted covariance remains below a task-specific bound. This replaces a fixed observation stride with a principled, state-dynamics-dependent schedule.
Build a recurrent or state-space model with a base state carrying task-relevant dynamics and an explicitly contracting auxiliary state. If the training loss or energy depends on the auxiliary state, replace it by a quotient loss plus an analytically known telescoping correction; long-run optimization and invariant averages are then unchanged, while transient fiber effects decay geometrically.
Choose the neural operator's input-history length from the measured correlation time of the unresolved closure signal produced by coarse-graining. This avoids under-memory, which causes systematic closure error, and over-memory, which increases attention cost and can destabilize training. The same diagnostic can drive adaptive memory truncation across physical regimes.
Use the distance-matrix filtration of a sequence embedding as a cheap proxy for state-space persistent homology, and map its persistent recurrence cycles into explicit latent-space loops. Train a recurrent, state-space, or Transformer encoder so that important recurrence cycles have geometrically coherent trajectory paths rather than being artifacts of isolated pairwise returns. This avoids building a Vietoris-Rips complex over every latent window while retaining a mathematically controlled…
Train a small encoder and latent Koopman predictor to forecast whether a neural sequence model will enter a high-error or high-instability region, then execute an expensive refinement block only when the forecasted risk exceeds a threshold. The base model remains active at every step, so the learned preview model controls computation rather than directly replacing the main predictor. Add a bounded-rate interpolation when the gate switches off, preventing abrupt changes in recurrent state or…
Use the critical-droplet mechanism to control noise injection and perturbation-based switching in bistable recurrent networks or diffusion samplers. Instead of applying uniform noise, estimate front speed and interface cost, then create the smallest spatially localized perturbation expected to exceed the critical droplet size and trigger deterministic growth toward the target attractor.
Replace a large diagonalizable recurrent or state-space transition operator by a sparse set of retained oscillatory modes selected according to their contribution to the output autocorrelation. Unlike magnitude-based pruning, the objective is to preserve the power-law return signal generated by pairwise spectral differences, enabling long memory with far fewer modes.
Choose the retained context length by balancing the statistical complexity of adding lag j against the squared prediction bias from discarding it. Unlike a fixed context window, the rule uses both the remaining-horizon spectrum and the estimated tail energy, allowing a model to expand or shrink its memory online.