✗ Failed on benchmark
2026
Replace repeated full-curvature factorizations when evaluating many Bayesian neural-network hyperparameter settings with one shared low-rank approximation of the data-induced precision update. Reuse the same curvature subspace while changing prior precision, observation-noise precision, or layerwise regularization, evaluating each candidate through a small rank-r matrix and Woodbury solves.
Useful6/10
Difficulty5/10
Novelty5/10
✓ Mechanism works
2026
Replace a flat MoE or graph-pooling assignment with recursive partitions selected by interaction evidence after removing each item’s expected degree effect. Tokens, nodes, or examples that are frequently active for purely popularity-related reasons should not automatically form an expert or cluster. Recursion stops when a candidate split has nonpositive degree-corrected evidence, producing an adaptive hierarchy rather than a fixed number of equally sized groups.
Useful6/10
Difficulty6/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Replace ordinary one-token-to-one-expert or one-token-to-one-attention routing with a local latent subset router: a pooled observation can be explained by a compatible subset of tokens. Pairwise compatibility scores assign probability to subsets, and each token receives the marginal probability that it belongs to the selected subset. This should help when tokens represent overlapping objects, occluded entities, or multiple features that should be processed jointly.
Useful6/10
Difficulty5/10
Novelty6/10
✗ Mechanism failed
2026
Replace a flat stochastic MoE router by a two-level continuous-time routing model: experts within a group communicate rapidly, while transitions between groups occur slowly. Use the effective class-level stationary distribution as a soft load-balancing prior, reducing routing oscillation while preserving expert specialization.
Useful6/10
Difficulty6/10
Novelty6/10
✗ 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
✗ 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
Unverified
2026
Replace a dense weak pairwise interaction over tokens or graph nodes by an analytically resummed Gaussian quadratic term plus explicitly evaluated corrections containing only subsets of size at most R. Choose R from the target approximation error and interaction scale instead of fixing the interaction order heuristically. This is intended for dense global attention, graph interactions, or latent-variable energy models with individually small couplings.
Useful5/10
Difficulty7/10
Novelty8/10
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
Replace a one-step recurrent update with a causal convolution over past affine innovations using an exponential-fractional kernel. Add mean reversion and calibrate the innovation amplitude so that activation mean and variance remain approximately invariant across sequence position while retaining long-range, power-law-like memory.
Useful5/10
Difficulty6/10
Novelty5/10
Unverified
2026
Calibrate random edge dropout in a GNN or sparse-attention layer using the spectral radius of the underlying communication graph. Retain edges with probability p chosen so that p lambda(A) is at least 1 plus a safety margin, preventing the random computation graph from entering a subcritical fragmented regime while retaining high sparsity.
Useful5/10
Difficulty4/10
Novelty7/10
Unverified
2026
Replace dense attention on tree-structured inputs with stochastic attention neighborhoods formed by metric balls of sampled radii. Use the paper's exact trimming rule to ensure that every sampled cover remains valid while its total radius budget is bounded, then average predictions over several independent covers during training. This creates sparse, globally covering attention masks with an explicit locality-versus-coverage control.
Useful5/10
Difficulty6/10
Novelty8/10
Unverified
2026
Use Samuels' exact lower bound as a differentiable certificate for the probability that a random neural-network cost remains below a hard budget, under independent nonnegative component costs and known means. This can regularize stochastic MoE loads, activation memory, dynamic depth, or per-example loss decompositions without assuming variances or bounded support.
Useful5/10
Difficulty4/10
Novelty8/10
Unverified
2026
Build a binary hierarchy over tokens by recursively splitting each active block with a beta-splitting rule, then perform dense attention only inside small leaf blocks and communicate between leaves through learned summaries at internal nodes. The beta parameter controls how balanced the partition is, while the paper's maximum-depth asymptotic supplies a principled depth budget and a way to detect pathological trees.
Useful5/10
Difficulty6/10
Novelty5/10
Unverified
2026
Replace dense token-to-token attention on a 2D token grid with local attention plus sparse horizontal and vertical communication axes. Tokens at intersections of selected axes receive extra cross-axis attention edges, creating a reinforced sparse graph that can transmit information across large blocks while using far fewer edges than dense attention. The mask should use light-tailed, approximately geometric spacing in both directions rather than heavy-tailed spacing in one direction.
Useful5/10
Difficulty5/10
Novelty6/10
Unverified
2026
Replace an unrestricted additive recurrent or fast-weight memory with a sign-selectable update: for each incoming update vector, choose between adding and subtracting it so that a smooth compact potential of the memory state is minimized. This is appropriate when the memory representation has sign symmetry, such as signed random features or a learned linear sketch; it is not a drop-in replacement for ordinary gradient updates where the sign carries semantic information.
Useful5/10
Difficulty5/10
Novelty9/10
Unverified
2026
Regularize a hard MoE router so that assignments remain block-jumbled: every group of token positions sends approximately the expected number of tokens to every group of experts or capacity slots. The condition detects localized routing collapse that ordinary global load balancing can miss, while requiring only a small block-count matrix rather than expensive pairwise or pattern statistics.
Useful5/10
Difficulty4/10
Novelty5/10
Unverified
2026
Replace or augment an exponential state-space memory branch with a causal convolution whose lag-j weight is exp(-lambda j) ell(j)/j. The 1/j boundary provides broad logarithmic memory, while lambda supplies an explicit finite memory scale and prevents uncontrolled accumulation from an untempered long-memory kernel.
Useful5/10
Difficulty5/10
Novelty5/10
Unverified
2026
Replace a uniformly discretized recurrent or continuous-depth model with hybrid hidden-state dynamics: integrate a learned drift between event times, then apply a one-sided reflection update at each irregular observation or constraint event. The reflection prevents the hidden state from violating a lower obstacle, while the explicit jump decomposition avoids smearing abrupt information changes across many small residual steps.
Useful5/10
Difficulty4/10
Novelty5/10
Unverified
2026
Use the model's non-monotonicity result to make graph connectivity a learned resource rather than assuming that every extra edge helps. An edge router assigns transmission scores but also charges a source-side collision cost for exposing an infected node to many susceptible neighbors. The resulting router can prune edges that increase competition and reduce useful reachability.
Useful5/10
Difficulty5/10
Novelty7/10
Unverified
2026
Replace independent softmax expert choices with a collision-free Markov router whose particles occupy expert positions on a one-dimensional or circular index lattice. A particle can move only to an empty neighboring expert, and the move rate contains a product of sine ratios that globally repels nearby assignments; this should reduce expert collapse and produce more evenly spread routing without requiring a separate pairwise diversity loss.
Useful5/10
Difficulty7/10
Novelty8/10
Unverified
2026
Replace raw hyperbolic embedding-radius regularization with a dimension-aware effective-radius target. For embeddings concentrated near hyperbolic radius rho in an n-dimensional hyperbolic space, regulate s times log(sinh(rho) / sqrt(n)) rather than rho itself, and use the same quantity to calibrate distance-logit temperature. This should make hyperbolic metric-learning behavior more invariant when embedding dimension, curvature, or model scale changes.
Useful5/10
Difficulty4/10
Novelty6/10
Unverified
2026
Construct a nested sequence of representations in which each coarser representation is obtained from the previous one by a 1-Lipschitz projection. Train prediction heads at multiple scales so coarse predictions remain stable and approximately recoverable from the finer representation, enabling early exit, token pooling, and controlled multiresolution inference.
Useful5/10
Difficulty6/10
Novelty6/10
Unverified
2026
Use Oja's streaming eigenvector estimate on a parameter block's incoming gradient stream, but activate its rank-one preconditioning correction only after the mathematically predicted d log d sample threshold. Before that point, the estimate is treated as unreliable and the optimizer remains close to AdamW or SGD. This prevents early noisy spectral directions from destabilizing training while retaining an O(d)-memory alternative to storing a full gradient covariance matrix.
Useful5/10
Difficulty5/10
Novelty7/10
Unverified
2026
Represent an input perturbation, feature mask, or expert route as a subset of a ground set of size n. Collect a useful family F of k-subsets, then generate larger l-subsets only by adding l-k elements to members of F; these are the upper-shadow augmentations. The paper's explicit profile predicts a guaranteed fraction of distinct l-masks covered by this procedure, allowing an augmentation system to replace inefficient random mask sampling with targeted combinatorial expansion.
Useful5/10
Difficulty4/10
Novelty9/10