Unverified
2026
Replace an unconstrained attention or MoE routing matrix with a row-stochastic matrix that preserves a learned Gibbs-like distribution over groups. Anneal its temperature so that the router converges to a block-upper-triangular operator: mixing remains flexible within equal-energy groups, while cross-group traffic becomes directional rather than oscillating or collapsing. This creates a hierarchy-aware inductive bias and can reduce the number of active expert or attention connections.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Add a consistency loss for a neural model processing irregularly sampled functions or sequences, comparing its latent covariance representation before and after a time-coordinate warp. The target discrepancy should be near zero for affine transformations but should grow with the normalized curvature, bandwidth, and local sparsity of a nonlinear warp.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Use a small set of known or trusted pairwise costs to estimate the effective entropic temperature of a Sinkhorn attention or mixture-of-experts routing layer directly from its observed transport plan. This provides a calibration controller that can detect over-concentrated routing and adjust epsilon without backpropagating through a costly temperature search.
Useful6/10
Difficulty4/10
Novelty8/10
Unverified
2026
Insert a projection-space mixer that combines several fixed or learned directions using reciprocal correlations with the current feature, then normalize the result. The exact construction has a universal beta law for its squared input-output cosine, so it can create controlled angular diversity while remaining deterministic and independent of the chosen direction dictionary.
Useful6/10
Difficulty5/10
Novelty8/10
Unverified
2026
Replace an unconstrained residual block by a first-order gradient-flow correction whose energy contains first-, second-, and third-difference penalties, mirroring the paper's higher-gradient gravitational energy. The correction suppresses high-frequency modes while retaining a trainable nonlinear residual branch, and its step size can be chosen from an explicit spectral stability bound.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Replace an unconstrained hidden-to-hidden interaction in an MLP or transformer feed-forward block by two gauge-related branches. Split channels with an orthogonal involution Θ, constrain the learned interaction K to anticommute with Θ, and use opposite signs of K in paired branches. This creates a testable inductive bias in which the learned interaction only transfers information between the two channel subspaces.
Useful6/10
Difficulty5/10
Novelty8/10
Unverified
2026
Replace an ordinary local convolution or token-mixing block by two nonnegative feature populations A and B that diffuse and drift along the spatial or token axis, with transport rates increasing quadratically with local population and with an optional directional bias. Add a local A plus B to empty reaction so mutually conflicting feature mass is removed rather than merely averaged. The module should produce adaptive competition, and its isolated relaxation should exhibit a measurable t raised…
Useful6/10
Difficulty5/10
Novelty8/10
Unverified
2026
Impose a screened pair-correlation dynamics on stochastic neural replicas so that correlation fluctuations relax locally instead of propagating across the entire representation. The key control knob is a learned or scheduled relaxation rate \(\mu_{FB}\), which predicts a measurable correlation length \(\xi_{FB}=\sqrt{D_{eff}/\mu_{FB}}\). This can be used as a locality regularizer for token representations, diffusion trajectories, or recurrent hidden states.
Useful6/10
Difficulty5/10
Novelty8/10
Unverified
2026
Replace the raw position/channel basis of a one-dimensional sequence module by eigenvectors of the projected cubic radial kernel matrix. Penalize or truncate coefficients in eigenmodes with many sign changes, giving a mathematically ordered smooth-to-oscillatory inductive bias while preserving the two-dimensional nullspace corresponding to affine trends.
Useful6/10
Difficulty4/10
Novelty6/10
Unverified
2026
Replace the linear phase progression in a positional encoding or rotary attention mechanism with a deterministic quadratic phase. The resulting position signal is generated by an irrational rotation with linearly changing increments, and the paper proves that its infinite diffraction measure is purely absolutely continuous, suggesting disorder-like spectral coverage without random sampling.
Useful6/10
Difficulty4/10
Novelty6/10
Unverified
2026
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.
Useful6/10
Difficulty4/10
Novelty6/10
Unverified
2026
Add a local Jacobian spectral regularizer and an initialization sweep to steer a looped transformer away from uncontrolled near-unit dynamics. The goal is to prevent examples from entering a fold-critical regime with very long relaxation times, or alternatively to deliberately target a controlled critical regime when adaptive test-time compute is useful.
Useful6/10
Difficulty6/10
Novelty5/10
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
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
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
Attach each token or graph node a learned scalar charge q_i and add a fractional stable kernel K_ij = exp(-tau D |q_i-q_j|^alpha) to the interaction mechanism. Constrain 0 < alpha <= 2, the exact range in which the kernel is positive semidefinite for arbitrary finite real charge sets, and optionally make tau layer-dependent to obtain multiscale interactions. This provides a principled alternative to unconstrained learned distance biases and can be used either as an attention-logit bias or as a…
Useful6/10
Difficulty4/10
Novelty5/10
Unverified
2026
Construct training and evaluation examples whose every (d-1)-variable marginal is exactly independent, but whose full d-variable distribution contains a parity interaction. This isolates genuine high-order reasoning from shortcuts based on pairwise or lower-order statistics and can expose whether attention or MLP architectures learn the intended interaction.
Useful6/10
Difficulty3/10
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
Parameterize a sequence attention graph by a finite graph motif on positions [r], then lift that motif to arbitrary sequence lengths using strictly increasing embeddings. Instead of learning an independent positional mask for every context length, the model learns a small collection of order-preserving edge types whose induced graph is valid at every length and can be extended without retraining.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Replace unconstrained or entropy-regularized MoE routing with a minimally disruptive update that preserves a lower bound on the log-determinant of the experts' weighted output span. The router still tracks the desired mixture, but a projection prevents the active experts from becoming linearly redundant or collapsing onto a low-rank subset.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Treat a selected neural submodule as an open dynamical system embedded in the rest of the network. Regularize it to contain internal modes that are simultaneously reachable from many external features and observable through many external outputs, rather than behaving as a one-sided receiver, broadcaster, or disconnected read/write split.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Replace a neural network head that predicts a covariance or other SPD matrix entrywise with regression in the matrix-log domain. The network predicts a symmetric matrix in unconstrained Euclidean coordinates, the matrix exponential guarantees an SPD output, and training can use intrinsic log-Euclidean or affine-invariant errors rather than Frobenius error on raw entries.
Useful6/10
Difficulty4/10
Novelty5/10
Unverified
2026
Add a multiscale penalty to transformer token-mixing activations when they are simultaneously concentrated on a spatial or token subset and on a separated, irregular frequency subset. The penalty uses the fractal uncertainty scaling law to discourage hidden states from collapsing onto narrow token patterns and narrow spectral bands, potentially improving robustness to token masking and frequency-corrupted inputs.
Useful6/10
Difficulty5/10
Novelty7/10