✗ Mechanism failed
2026
Represent candidate two-dimensional attention windows as dyadic rectangles and penalize local regions where many deeply embedded windows overlap. Use complementary horizontal and vertical depth exponents rather than independently penalizing one coordinate. The resulting router should reduce pathological concentration of sparse attention computation while preserving access to multiscale context.
Useful6/10
Difficulty6/10
Novelty7/10
✓✓ Beats tuned baseline
2026
Replace one deterministic residual update with a short cyclic composition of learned vector fields evaluated for randomized, short run times. Because finite compositions of noncommuting flows generate directional-derivative and Lie-bracket terms, changing the cycle order gives the network an explicit, low-cost way to learn drift directions that are unavailable from the individual vector fields alone.
Useful6/10
Difficulty5/10
Novelty7/10
✗ Failed on benchmark
2026
Represent an intermediate feature as a low-rank PSD matrix and compress it using nonnegative measurements \(\langle A_i,X\rangle\), while penalizing the empirical ratio between maximum and minimum measurement distortion over low-rank feature pairs. This directly discourages collapsed directions and excessively amplified directions in a covariance or Gram-feature bottleneck.
Useful6/10
Difficulty5/10
Novelty6/10
✗ Failed on benchmark
2026
Add a loss term requiring a neural optimizer or recurrent module to decrease a nonnegative Lyapunov-like energy over M update steps, rather than forcing monotonic one-step decrease. The term includes an empirically estimated mismatch allowance, so stochastic or delayed updates are tolerated while persistent instability remains penalized.
Useful6/10
Difficulty4/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Construct a sparse attention support by solving multiple small perturbed assignment problems between query and key embeddings and taking the union of the selected optimal matchings. Use the resulting spanning tree as the only set of cross-token edges, with edge biases determined by empirical assignment frequency.
Useful6/10
Difficulty7/10
Novelty7/10
✗ Mechanism failed
2026
Treat a recurrent or state-space layer as a finite-state Markov cocycle and constrain optimizer steps using the paper's inverse-logarithmic sensitivity of Lyapunov exponents near a zero exponent gap. Instead of enforcing a crude spectral-norm bound, allow updates that are harmless for long-run growth while shrinking steps that could substantially change the recurrent stability profile.
Useful6/10
Difficulty5/10
Novelty7/10
✗ Failed on benchmark
2026
Replace the usual hand-designed expert-load penalty with a heterogeneous survival penalty derived from a susceptibility distribution. Each expert receives an availability factor q_e=G(A_e), where A_e is its cumulative recent routing pressure and G_e is a learned or fixed mixture of exponentials; highly used experts are suppressed smoothly, while heterogeneous experts can have different resistance to pressure. The mixture produces adaptive curvature and long-tailed penalties that may reduce…
Useful6/10
Difficulty4/10
Novelty6/10
✗ Failed on benchmark
2026
Regularize a circular recurrent kernel by directly controlling the growth rate and phase velocity of its Fourier modes. This converts replay-speed selection into a low-dimensional spectral control problem and can suppress unstable or excessively slow modes without adding recurrent parameters.
Useful6/10
Difficulty6/10
Novelty8/10
✗ Mechanism failed
2026
Choose the consensus gain and gradient-tracking gain in decentralized training from the communication Laplacian spectrum rather than tuning them independently. The gains minimize the worst asymptotic pole radius for the paper's exact quadratic model, providing a principled initialization and a conservative stability safeguard for neural-network optimization.
Useful6/10
Difficulty6/10
Novelty6/10
△ Mechanism confirmed, baseline not beaten
2025
Regularize a neural dynamical map so that its log-volume expansion is cohomologous to a constant rather than forcing the Jacobian determinant to be constant at every state. Learn a scalar potential that explains transient expansion and penalize only the non-telescoping component, which should reduce long-horizon gradient explosion or collapse while retaining useful average expansion.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Apply a regularizer that penalizes feature disagreement under a finite set of known transformations. The paper's spectral-gap inequality gives a quantitative reason that this local consistency penalty controls distance from the subspace invariant under the transformation group, while the task loss prevents undesirable collapse.
Useful5/10
Difficulty5/10
Novelty5/10
Unverified
2026
Represent an original neural-network block and a proposed rewritten block as constrained optimization formulations over inputs and trainable parameters, then certify that the rewrite preserves feasibility and the ordering of losses over a bounded domain. This gives a compiler or pruning pipeline a formal reject/accept gate instead of relying only on numerical regression tests.
Useful5/10
Difficulty7/10
Novelty8/10
Unverified
2026
Regularize a scalar network output so that its superlevel sets are approximately quasiconcave in input or latent space. Instead of penalizing the full Hessian, penalize positive curvature only in directions orthogonal to the output gradient, matching the paper's projected-Hessian and weighted 1-Laplacian structure.
Useful5/10
Difficulty6/10
Novelty6/10
Unverified
2026
Train an implicit neural field with a regularizer that evaluates its level-set minimality operator after several nonuniform diagonal coordinate dilations. Instead of penalizing only the aggregate operator at the original coordinates, invert the resulting Vandermonde system and penalize every coordinate-pair coefficient separately. This suppresses hidden curvature cancellations and should produce level sets that remain geometrically simple under anisotropic rescaling.
Useful5/10
Difficulty6/10
Novelty8/10
Unverified
2026
Use the reciprocal arrangement as a probe of whether a learned representation has the intended angular response, and penalize deviations from the paper's universal beta distribution. This converts the theorem into a distribution-level regularizer rather than assuming that the reciprocal layer itself improves task loss.
Useful5/10
Difficulty4/10
Novelty7/10
Unverified
2026
Use the graph-coloring stability concept to route graph nodes to experts. Nodes rank experts by router logits, adjacent nodes are constrained to use different experts, and a blocking cycle is a directed cycle in which every node prefers the expert currently assigned to the next node. Eliminate profitable feasible cycles or penalize their existence so routing reaches a locally stable assignment instead of oscillating between equally plausible expert allocations.
Useful5/10
Difficulty6/10
Novelty9/10
Unverified
2026
Represent each example or minibatch by two positive semidefinite feature maps, such as teacher and student covariance operators, and penalize their noncommutative operator-valued f-divergence rather than only a scalar KL or Frobenius distance. The matrix-valued penalty preserves directional disagreement in feature space and is compatible with positive postprocessing, making it a candidate replacement for covariance matching in distillation and representation regularization.
Useful5/10
Difficulty5/10
Novelty6/10
Unverified
2026
Replace a static top-k MoE capacity rule with a router whose expert allocation evolves through a finite-domain coverage process. Experts with larger current occupancy can either receive more future capacity, intentionally amplifying specialization, or receive less capacity by reversing the size dependence, allowing a controlled test of the paper's asymmetry-amplification mechanism.
Useful5/10
Difficulty6/10
Novelty6/10
Unverified
2026
Build a constrained autoregressive model whose initial logits are generated from a translation-invariant MPS associated with a local zero-mode construction. The MPS supplies a structured valid distribution before a Transformer residual is added, so the model starts on the constraint manifold instead of learning validity through a penalty.
Useful5/10
Difficulty6/10
Novelty6/10
Unverified
2026
Add a Michael-Simon-inspired penalty to a neural implicit surface, neural renderer, or differentiable mesh generator. The penalty suppresses large-area sheets whose anisotropic first variation is small, which should reduce spurious folds, floating components, and geometrically unstable solutions while preserving surfaces required by the task loss.
Useful5/10
Difficulty6/10
Novelty8/10
Unverified
2026
Represent a neural model's particle ensemble, latent samples, or routing prototypes as an empirical probability measure and penalize its Wasserstein total variation across training or inference steps. Discrete resampling and particle replacement remain allowed, but their mass-distance cost is made explicit so the model cannot obtain a cheap distributional change through untracked teleportation. A weak continuity-equation residual can be added as an auxiliary loss or used as a diagnostic.
Useful5/10
Difficulty6/10
Novelty6/10
Unverified
2026
Train two parameter replicas with common low-rank stochastic forcing and an adaptive finite-dimensional Cameron–Martin correction that contracts their discrepancy in a weak parameter metric. Transporting the forcing directions through the loss Hessian is intended to make a rank-k perturbation influence more than k raw parameter directions, while damped momentum suppresses high-energy divergence.
Useful5/10
Difficulty7/10
Novelty7/10
Unverified
2026
Represent a time-dependent Hamiltonian system on the reduced state $(q,t,p_q)$ rather than on the redundant extended state $(q,t,p_q,p_t)$. A neural Hamiltonian section predicts one canonical representative of each affine cotangent fiber, while an optional symmetry loss enforces consistency under transformations that translate time.
Useful5/10
Difficulty5/10
Novelty8/10
Unverified
2026
Constrain a channel-mixing layer to be a product of nonnegative bidiagonal matrices, rather than an unconstrained dense matrix. The resulting totally nonnegative operator is predicted not to increase sign oscillations in ordered channel features, potentially reducing high-frequency feature noise and making deep stacks more stable.
Useful5/10
Difficulty4/10
Novelty8/10