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 dense token attention with K asynchronous local nonlinear consensus updates on a sparse token graph. Each selected token is moved to the unique incident p-mean of its neighbors, producing a tunable message-passing operator that is averaging for p=2 and nonlinear for other p.
Replace dense aggregation over patches, points, or graph neighbors with a state-dependent positive quadrature rule. A latent encoder produces the current manifold coordinate z, a small controller chooses a subset of entities, and a convex redistribution step assigns nonnegative weights so that the subset preserves the full aggregation on a learned feature basis. The method is especially suitable for neural operators and vision attention where many spatial entities are redundant along the data…
Replace unconstrained momentum updates for selected neural-network parameter blocks by damped second-order motion on a unit sphere. The update learns directions while preserving the block norm exactly, and includes the centripetal term required for a valid acceleration constrained to the sphere rather than merely projecting gradients after an unconstrained step.
Replace independent softmax routing or attention normalization with a differentiable approximate distribution over one-to-one assignments, using the Bethe permanent as the partition-function surrogate. Constrain the allowed token-to-expert or query-to-key support graph to have high girth, which gives an explicit bound on the approximation error and reduces short-cycle-induced correlations.
Add a recurrent associative matrix to each selected transformer layer so recent key-value relationships can be retrieved without retaining every past token or performing gradient updates. The matrix uses input-dependent retention and write gates, but retrieval is always performed from the pre-write state, preventing the current target from leaking into its own prediction. Frobenius-norm clipping makes the recurrent memory bounded and provides a direct stability control.
Insert a differentiable reaction-diffusion layer that converts dense token or pixel features into sparse, spatially coherent routing masks. Two competing orientations form complexes through conserved monomer reservoirs, so local assignments can cluster while opposite assignments mutually exclude one another instead of independently activating at the same location. The layer can be used as a soft-to-hard MoE router, attention-mask generator, or object-part grouping module.
Replace an ordinary group-L1 penalty on structured neural components with a powered ratio-of-norms penalty applied to their nonnegative importance magnitudes. The ratio encourages importance to concentrate on a small number of heads, channels, or experts while being less sensitive to arbitrary rescaling of the underlying weights. After training, components with small importance can be physically removed and the model can be fine-tuned.
Replace fixed-budget token or patch pruning with greedy selection that combines a teacher-derived relevance score and Gaussian-process mutual information. Select an item when it is both relevant and non-redundant, and stop when the largest remaining information gain falls below a calibrated threshold instead of retaining a fixed number of items.
Train a graph neural network with a learnable fractional graph mask constrained toward the tree-double-matching polytope. This gives the model a global connectivity constraint while penalizing attention patterns that concentrate too many incident edges on individual vertices.
Approximate a dense symmetric interaction matrix in a neural layer by \(\widehat A=C\widehat M C^{\top}\), but compute the small core \(\widehat M\) from a two-sided sketched least-squares fit rather than from the landmark principal submatrix. This preserves signed or indefinite directions and avoids exploding outputs caused by an almost-singular \(A(I,I)\).
Replace repeated substrings in long sequences with nonterminal symbols from an acyclic straight-line grammar, then run the transformer on the compressed sequence. Unlike ordinary fixed tokenization, the compression objective explicitly minimizes the number of reusable binary productions, allowing repeated document-specific or corpus-level motifs to become single units. An expansion map lets the model recover token-level outputs for selected positions.
Represent hierarchical routing decisions by one weighted bipartite graph between parent experts or regions and fine cells or token groups. Select a regularized edge set once, then derive both coarse parent activation and fine-grained routing from it, preventing later refinement from invalidating earlier load balancing.
Replace an unconstrained multiplicative interaction between two nonnegative neural features by a lifted gate whose first and second moments satisfy the paper's semidefinite relaxation for the set F = {(x1,x2): x1,x2 >= 0, x1 x2 <= 1}. Insert the gate into an MLP, attention score, or MoE router to prevent explosive feature products while retaining a tractable convex feasible set.
Replace an unconstrained Wasserstein representation-matching loss with a graph-causal transport loss whose coupling at node k is conditioned only on the representations of its parents. This forces domain alignment, distillation, or augmentation consistency to respect the information flow of the model's DAG, reducing spurious matches that exploit descendants or globally visible features.
Use the OT spectral bound as a conditioning signal for optimizing parameters of a neural cost or inverse-OT objective. Adapt the parameter step size and add a covariance floor whenever the estimated Jacobian lower bound collapses, preventing optimization from entering regions where Sinkhorn outputs become insensitive to the learned cost.
Represent a collection of neural directions as generators of a zonotope and reward the volume spanned by their subsets. The objective favors complementary, non-collapsed vectors rather than merely pairwise-separated vectors, making it suitable for attention heads, MoE expert signatures, or embedding prototypes. Use normalized generators and positive gates so the regularizer cannot be increased trivially by scaling.
Replace the fixed number of Sinkhorn iterations used to normalize an attention kernel with a per-example stopping rule based on the local spectral contraction factor. Estimate the remaining marginal error geometrically and stop early on easy examples while retaining extra iterations on difficult or nearly disconnected examples.
Give a token, patch, retrieval-item, or expert-selection module a learned utility f_theta(S) over subsets S, and penalize violations of the paper's submodularity and strong-submodularity inequalities. The resulting selector should prefer complementary elements: the marginal value of adding an item decreases when the current selected set is already rich in similar information. At inference, use greedy marginal-gain selection rather than independently thresholding token scores.
Add KL Tikhonov regularization to simplex-valued attention or routing parameters so that the optimizer selects a stable solution close to a chosen reference distribution instead of collapsing onto a few entries. Anneal the regularization strength to obtain exploration early and specialization later.
Replace independent top-k routing by a tree-structured hypergraph assignment layer. Each candidate route is a singleton or pair of resources, and the router selects exactly q_e routes for every tree edge e while ensuring that no resource is consumed twice. This removes capacity collisions before expert computation instead of repairing them with token dropping or load-balancing penalties.
Replace top-k token pruning by greedy maximization of a diversity-aware monotone submodular utility under a spacing or coverage constraint. The selector repeatedly chooses the feasible token with the largest marginal utility, avoiding the redundant-token failure mode of independent score ranking while inheriting a constant-factor approximation guarantee under the stated 2-system abstraction.
Represent prototypes or attention keys by points p_i on the unit sphere and regularize their configuration with a Green-potential log-partition objective inspired by the TPMS branch-point formulation. The objective penalizes configurations whose positive and negative Gibbs-weighted potentials are concentrated in different regions, providing a smoother alternative to pairwise repulsion or uniformity losses.
Replace raw pairwise attention or graph-edge scores by exact U-centered residuals, removing additive effects attributable to either endpoint. The resulting scores represent interaction beyond independent source and destination biases and satisfy zero row sums, preventing a few high-degree or high-activation tokens from dominating relational aggregation.
Replace Euclidean momentum updates on simplex-valued neural variables with entropy mirror descent when the variable dimension is large and the loss is naturally smooth relative to negative entropy. The lower bound predicts that Nesterov-style acceleration cannot guarantee an asymptotic improvement in this geometry, while the geometry-matched update preserves positivity and can reduce boundary instability.