✗ Failed on benchmark
2026
Treat the hidden-state evolution of an RNN or state-space model as a parameterized dynamical system and globally continue its attractors over a grid of inputs, perturbation amplitudes, and training checkpoints. Penalize or stop training when the task-relevant attractor loses basin mass, rather than relying only on local Jacobian eigenvalues at one nominal trajectory.
Useful7/10
Difficulty6/10
Novelty8/10
✗ Failed on benchmark
2026
Replace isotropic parameter penalties and diagonal Fisher estimates with a task-covariance interference budget. The update is damped only in directions where old-task features have large variance, while directions absent from old-task feature support remain available for learning the new task. This may preserve old-task performance with less loss of plasticity than unconditional projection.
Useful7/10
Difficulty5/10
Novelty6/10
△ Mechanism confirmed, baseline not beaten
2026
Equip a neural policy or learned world model with an MPPI-style rollout planner whose perturbation covariance is conditioned on a discretized latent-state cell and updated from observed transition residuals. Apply spatial diffusion to neighboring covariance estimates using a kernel matched to the empirical visitation distribution, so covariance adaptation is smoothing rather than an unstable independent estimate at every state.
Useful7/10
Difficulty6/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Measure how validation forecast error grows with prediction horizon and fit exponential and Mittag-Leffler models. When the Mittag-Leffler fit is decisively better, activate a fractional-memory SSM or long-memory residual branch and use its fitted effective order to set the branch's kernel decay and horizon-loss weights; otherwise retain a conventional recurrent or finite-memory branch.
Useful7/10
Difficulty6/10
Novelty7/10
✗ Failed on benchmark
2026
Treat optimizer configurations as elements of a finite intervention poset and decompose validation loss or training traces into pure causal effects rather than raw ablation differences. The recovered second- and higher-order effects reveal whether, for example, momentum and adaptive preconditioning are complementary, redundant, or destabilizing, and can be used to select a smaller optimizer or construct a better configuration.
Useful7/10
Difficulty4/10
Novelty7/10
✓✓ Beats tuned baseline
2026
Use the differentiable covariance chart to construct a Fisher-information preconditioner for the edge and innovation parameters of a linear-Gaussian neural module. Instead of applying an isotropic Euclidean update, whiten parameter steps according to how strongly they change the predicted Gaussian distribution. This targets ill-conditioning caused by redundant paths, correlated latent nodes, and badly scaled innovation covariances.
Useful7/10
Difficulty6/10
Novelty5/10
✓ Mechanism works
2026
Search for a compact symbolic optimizer instead of selecting among fixed AdamW-like formulas. Encode optimizer programs as token sequences, learn a continuous variational representation of those sequences, and use a Gaussian-process Bayesian optimizer to propose promising update rules based on short neural-network training rollouts.
Useful7/10
Difficulty6/10
Novelty6/10
✗ Mechanism failed
2026
Use an online estimate of the target distribution's remaining coordinate dependence to choose diffusion step sizes instead of using a uniform timestep grid. Take large jumps when the current state is nearly coordinate-factorized and shorten the step when the denoiser detects strong cross-coordinate dependence, targeting the paper's dependence-adaptive discretization scaling.
Useful7/10
Difficulty4/10
Novelty8/10
Audited (legacy)
2026
Replace independent per-action distributional value heads with a critic whose shared latent particle produces a vector of return samples for all actions simultaneously. Train the predicted joint return vector against a Bellman target vector formed from coupled counterfactual reward-transition samples, using a sliced Wasserstein loss. The greedy action is selected by the mean of the corresponding marginal particles, while shared particles retain cross-action dependence for learning and…
Useful7/10
Difficulty6/10
Novelty7/10
Unverified
2026
Treat the correlation decay exponent and Hermite rank as explicit hyperparameters controlling the roughness of the depth-wise residual trajectory. Use smoother long-memory drivers for stable deep propagation and less correlated drivers when optimization needs more layer-wise diversity, while retaining critical scaling so the network does not collapse to the identity.
Useful6/10
Difficulty5/10
Novelty8/10
Unverified
2026
Add a separately identified constant or slowly varying force channel to a neural dynamics model, analogous to the paper's online gravity-trim and mass adaptation. This isolates persistent low-frequency prediction bias from state-dependent dynamics, reducing repeated corrective action by the planner and avoiding expensive network retraining after a systematic shift.
Useful6/10
Difficulty3/10
Novelty5/10
Unverified
2026
Attach a small, online-adapted Taylor-polynomial residual model to a neural dynamics model instead of adapting all network weights. Construct the dictionary from the local Taylor structure, remove monomials forbidden by known joint-odd symmetry, and update its coefficients with recursive least squares. This yields fast adaptation with an interpretable parameter count and a direct test of whether the assumed symmetry is valid.
Useful6/10
Difficulty4/10
Novelty6/10
Unverified
2026
Treat discrete training events such as gradient-norm spikes, curvature changes, rejected steps, or minibatch outliers as jump channels and apply an event-specific parameter update map. The optimizer should be evaluated using both progress and the information cost of selecting the feedback map, because feedback may reduce loss fluctuations or improve adaptation without changing the average update magnitude.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Regularize the state-transition or input-output Jacobian of a recurrent, state-space, or implicit neural network so that its complex eigenvalue cloud belongs to a selected non-Hermitian symmetry class and has the corresponding unfolded pair statistics. Combine this statistical-shape constraint with an explicit spectral-abscissa or spectral-radius margin, preventing the network from obtaining good average singular values while remaining highly non-normal and transiently unstable.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Add a low-dimensional, trusted sentinel state to the optimizer or recurrent inference controller. The sentinel is driven by a secret probe and a protected gain, so unauthorized gain changes produce a predictable shift in its equilibrium even when the main neural dynamics remain numerically stable. Monitor the estimated equilibrium and trigger rollback or quarantine when the measured shift exceeds the expected noise envelope.
Useful6/10
Difficulty4/10
Novelty8/10
Unverified
2026
Replace an explicit recurrent transition with a learned descriptor relation in latent space, allowing some latent coordinates to satisfy algebraic constraints rather than being numerically integrated. Fit the relation using total-least-squares or iterative PCA on the jointly observed trajectory, so noise in every channel is treated symmetrically and the model can discover whether the latent system is index-0 or index-1.
Useful6/10
Difficulty6/10
Novelty7/10
Unverified
2026
Train a neural state-space model whose one-step dynamics are linear in a fixed analytic feature vector, and use the empirical feature Gram matrix to detect whether passive trajectories identify the dynamics. Add data collection or replay only when the Gram matrix is poorly conditioned; the analytic-feature assumption predicts that persistent excitation should emerge without deliberately visiting every operating mode.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Build a recurrent or state-space network with time-dependent transition parameters, but train it to forget perturbations at a common exponential rate across all admissible parameter schedules. The model should retain task-relevant long-term signals while suppressing dependence on arbitrary initial hidden states, reducing instability under changing inputs, curricula, or deployment-time dynamics.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Replace constant friction and optimizer noise with a velocity-dependent friction gamma(u) and noise amplitude tied by a fluctuation-dissipation relation. High-speed momentum states can be damped and randomized differently from low-speed states, creating controlled transient exploration while preserving a known equilibrium momentum distribution.
Useful6/10
Difficulty5/10
Novelty7/10
Unverified
2026
Add a temperature-response constraint to stochastic neural predictors so that changes in inverse temperature cannot produce disproportionately large changes in expected loss or energy. This converts the nonequilibrium fluctuation-response inequality into a measurable robustness monitor and a regularizer for beta-conditioned stochastic representations.
Useful6/10
Difficulty5/10
Novelty6/10
Unverified
2026
Replace a static MoE load-balancing penalty with a two-stage capacity allocator. First compute each expert's technically feasible token capacity from latency, memory, and overflow constraints; then redistribute capacity using cumulative proportional fairness so experts that were repeatedly under-served receive more capacity later. Constrain the redistribution by an explicit efficiency budget, so fairness cannot silently cause an uncontrolled increase in routing loss or expert compute.
Useful6/10
Difficulty5/10
Novelty5/10
Unverified
2026
Use Fisher width as a branch coordinate in addition to training loss. During a short reference run with SGD, fit the expected Fisher-width curve as a function of loss, then add a soft penalty to Adam or another optimizer when its width at the same loss deviates from that reference branch. This directly tests whether optimizer-induced geometric displacement is responsible for differences in training dynamics or generalization.
Useful6/10
Difficulty5/10
Novelty6/10
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
Use SPINE's nested entropy profile on the singular values of each trainable weight matrix to discover spectral bands online, rather than choosing a fixed rank or a fixed number of learning-rate groups. Assign smaller step sizes or stronger decay to dominant singular-value bands and larger step sizes to weak bands, while updating the grouping only when the entropy-boundary signal is persistent.
Useful6/10
Difficulty6/10
Novelty7/10