Unverified
2026
Represent each token value in an attention head as a bounded three-dimensional Lie-algebra vector and impose a weighted polygon-closure condition on the aggregate value vectors. The module is invariant to a common \(SU(2)\simeq SO(3)\) rotation of all token vectors, preventing the head from spending capacity on an arbitrary global orientation. A soft closure penalty gives a drop-in experiment, while projection onto the zero-sum manifold provides a harder constrained variant.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
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.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Train an energy-based or probabilistic classifier with inverse temperature \(\beta\) matched to the precision \(\Delta\) of injected observation or label noise, following the exact higher Nishimori condition \(\beta=\Delta\). Use two independently sampled network replicas to measure an Edwards-Anderson-style parameter and detect whether training is entering a paramagnetic, ordered, or replica-disagreement regime rather than tuning regularization only by validation loss.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Augment a neural field or neural operator with a bank of localized, divergence-free moving packets whose radius and amplitude follow the Hill scaling rather than ordinary Gaussian scaling. The packet coefficients can represent unresolved flow corrections while keeping their L^2 contribution approximately invariant under refinement, preventing fine-scale features from becoming numerically negligible or explosively large.
Useful6/10
Difficulty5/10
Novelty8/10
Unverified
2026
Replace an MoE router's single softmax distribution with a normalized coordinate-wise product of several simplex-valued routing factors. The product preserves positivity and normalization but, as depth grows, concentrates mass on a small subset of experts, creating a mathematically controlled heavy-tailed routing prior rather than relying only on an auxiliary load-balancing loss.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Add a stable linear latent state-space block whose controllability Gramian is trained toward a chosen positive-definite target using squared Bures–Wasserstein distance. Direction-specific semidefinite constraints can suppress disturbance amplification in nuisance coordinates while preserving controllability in coordinates needed for prediction.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Track the variance of information content in a neural representation or routing distribution and use its interior maximum as a data-driven transition signal. The monitor distinguishes collapse, where nearly all probability occupies one state, from unstructured noise, where all states are equiprobable; both have low complexity, while structured intermediate distributions have high complexity.
Useful6/10
Difficulty4/10
Novelty7/10
Unverified
2026
Add a Renyi divergence penalty between the current network output distribution and a frozen reference distribution representing the pretrained model, a teacher, or a retained-data equilibrium. The Renyi order k becomes a control parameter: k greater than 1 strongly penalizes examples on which the new model assigns disproportionately more probability than the reference, while orders below 1 emphasize support mismatch and low-probability regions. Sweep or anneal k and detect a transition between…
Useful6/10
Difficulty4/10
Novelty6/10
Unverified
2026
Replace ordinary stride-2 pooling by a stochastic block-to-center map that is equivariant under global sign reversal and monotone in every input spin. For a binary feature channel, the layer computes the probability of a positive coarse feature from the number of positive fine features, samples or relaxes the resulting Bernoulli variable, and learns only a constrained scalar rather than an unconstrained pooling kernel. The same construction can be applied independently to channels or to graph…
Useful6/10
Difficulty4/10
Novelty6/10
Unverified
2026
Add a Jacobian cone-field regularizer to recurrent dynamics so that tangent directions expand and remain aligned with an unstable cone outside a designated critical neighborhood. The network is not forced to be uniformly expanding: the regularizer is disabled near the critical set, allowing controlled bifurcation-like behavior while exposing where long-horizon sensitivity changes.
Useful6/10
Difficulty7/10
Novelty8/10
Unverified
2026
Train an MoE router against demand shifts in which only a limited number of token groups can simultaneously move to their maximum deviation from nominal frequencies. For each adversarial demand, compute a balanced routing equilibrium or approximate system-optimal dispatch, and penalize the resulting worst expert congestion rather than only the nominal batch load. This should make routing robust to correlated bursts while avoiding the excessive conservatism of perturbing every token group…
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
For a neural block with matrix-valued activations and transformation Y = A X B, regularize the exact coupled spectrum of the two-sided map instead of penalizing A and B independently. A large singular direction in A is penalized more strongly when the corresponding singular direction in B is also large, directly controlling joint feature amplification.
Useful6/10
Difficulty5/10
Novelty5/10
Unverified
2026
Treat the router state as a symbolic base process and expert transformations as nonstationary expanding fiber maps. Add a relative entropy/free-energy constraint so that the router's conditional entropy is calibrated against the empirically measured growth rate of distinguishable expert trajectories, preventing premature expert collapse while retaining useful specialization.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Add a stochastic pairwise regularizer that penalizes only coordinate pairs whose normalized neural-field difference exceeds a threshold. Unlike a conventional fractional Sobolev penalty, the weak-type functional uses an indicator and a distance weight, and its Gamma-limit guarantees convergence toward a local gradient energy as the threshold grows.
Useful6/10
Difficulty4/10
Novelty7/10
Unverified
2026
Build each nonlinear correction in an inverse neural operator from explicit bilinear products of learned operator features, following the inverse Born expansion instead of using an unconstrained pointwise MLP. Use a square activation to implement multiplication exactly, and truncate the interaction order so the model has a controllable polynomial structure.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Add a hybrid geometric penalty between an attention distribution at one layer or training step and a reference distribution, such as detached attention from the preceding layer or optimization step. The penalty allows attention mass to move between nearby token positions at a transport cost while separately charging for local compositional changes, producing a structured alternative to KL or entropy regularization.
Useful6/10
Difficulty7/10
Novelty7/10
Unverified
2026
Insert a projection step after a graph neural network proposes edge weights, replacing the proposed Laplacian by the closest valid Laplacian with a prescribed block-component structure. The projection removes cross-block interactions while minimally changing within-block weights, and a block spectral-gap constraint guarantees that each block is connected rather than accidentally splitting into smaller components.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Replace the first several convolutional blocks of a small image model with a finite-depth convolution-modulus scattering stem built from a Parseval filter bank. Enforce exact energy accounting and use the paper's polynomial residual law to choose the smallest depth that captures the desired fraction of input energy, avoiding unstable or redundant deep scattering paths.
Useful6/10
Difficulty5/10
Novelty5/10
Unverified
2026
Use the localization theorem to turn a detected pointwise simulator error into a guaranteed region that must contain similarly large error, then place verification samples inside that region instead of sampling uniformly. The same bound can guide a training regularizer: errors with large amplitude and large local Lipschitz constants are penalized because they create planner-exploitable disagreement regions.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Use a window of observed neural-network update trajectories to identify the set of local quadratic objectives and preconditioners that are consistent with the observed optimizer behavior. Rather than selecting one arbitrary curvature model, retain the nullspace of compatible parameters and use its dimension or smallest singular value as an identifiability and stability diagnostic.
Useful6/10
Difficulty6/10
Novelty8/10
Unverified
2026
Apply a set-oriented graph analysis to the latent state dynamics of an RNN, SSM, or world model. Partition latent trajectories into compact cells, estimate the multivalued transition graph and its Markov matrix, then regularize the model so that recurrent latent modes form coherent strongly connected components with controlled transition entropy rather than spurious unstable wandering. This preserves meaningful metastable modes while preventing long-horizon rollout statistics from drifting away…
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Attach a finite mixture of zonotopes to each uncertain neural input or hidden state, and propagate every mixture component through affine layers and conservative nonlinear relaxations. When the number of components grows, merge components only with an enclosing zonotope and sum their probability masses, preserving a formal lower bound on the probability that the true activation lies in the represented set.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Represent hierarchical or tree-structured hidden states on words over d symbols and replace a dense mixing layer by a noncommutative Toeplitz operator composed of shared word shifts. Coefficients are reused at every tree location, so the parameter count depends on maximum interaction depth rather than the number of nodes; an optional spectral penalty controls the amplification profile of finite-depth truncations.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Add a differentiable fourth-order-mass penalty to coefficient vectors used by randomized signed projections, sign-noise layers, or stochastic quantizers. The penalty controls the effective number of active coefficients and therefore the distribution of the injected random fluctuation: diffuse coefficients generate nearly Gaussian perturbations, whereas concentrated coefficients generate larger non-Gaussian deviations.
Useful6/10
Difficulty3/10
Novelty7/10