△ Mechanism confirmed, baseline not beaten
2026
For each frozen weight tensor, append a second tensor of identically shaped zero weights and assign trainable scores to both the real and dummy edges. Select a fixed number of candidates by top-k score in the doubled space; real edges selected by the competition remain active, while selected dummy edges consume the quota without changing the network. The resulting number of active original edges is learned rather than imposed by a separate layerwise sparsity search.
Useful7/10
Difficulty4/10
Novelty7/10
✓✓ Beats tuned baseline
2026
Replace or augment a recurrent or state-space block with a small ensemble of stable linear memory lifts. Each lift stores a low-dimensional state whose repeated matrix powers generate a structured long-range convolution, and a learned gate mixes the experts using both their current predictions and their slowest block-operator mode. The module can represent multiple memory timescales without explicitly storing a long token history.
Useful7/10
Difficulty6/10
Novelty5/10
✗ Failed on benchmark
2026
When a neural state-space model has latent directions that are invisible under normal inputs, add a small structured carrier to the input or hidden-state update during selected training windows. The carrier changes local measurement and transition projections, analogous to the paper's carrier-dependent measurement and force projections, and can reveal modes that passive training leaves unconstrained.
Useful7/10
Difficulty5/10
Novelty7/10
✗ Failed on benchmark
2026
Add finite-horizon observability and reachability objectives to a recurrent or state-space neural model so that its latent modes are both inferable from outputs and influenceable by available inputs. This directly penalizes the failure mode identified in the paper: a large latent perturbation with nearly zero first-order output projection.
Useful7/10
Difficulty6/10
Novelty6/10
✗ Mechanism failed
2026
Replace or augment a recurrent hidden coordinate with a nonnegative bistable autocatalytic state driven by an external control signal. The cell retains information through metastable low and high states, while a periodic or slowly varying control produces a controlled phase lag and hysteresis useful for temporal regime detection. Explicit noise can be injected to test whether it enhances switching near the predicted intermediate-frequency regime.
Useful7/10
Difficulty6/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Replace the inner step of a neural optimizer with a safeguarded cubic local-model solve. Represent the cubic Taylor model as a homogeneous tensor in an augmented coordinate, solve proximal unit-sphere subproblems by alternating tensor contractions, decode a candidate step, and accept it only when the actual neural loss confirms the predicted decrease.
Useful7/10
Difficulty7/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Let a neural policy emit an unconstrained abstract action z, then solve a state-dependent feasibility problem that maps z to an admissible optimal-control parameter p before execution. Unlike coordinate-wise clipping, the mapping accounts for predicted dynamics, coupled state and input constraints, and recursive feasibility, allowing the policy to retain a simple unconstrained output space while the controller enforces plant constraints.
Useful7/10
Difficulty6/10
Novelty5/10
✓✓ Beats tuned baseline
2026
Replace constant decoupled weight decay with a coefficient proportional to the current learning rate divided by the peak learning rate. The optimizer applies ordinary decay at the learning-rate peak but weakens decay during cooldown and late training, preventing unnecessary steady-state parameter-norm shrinkage while retaining early-training stabilization.
Useful7/10
Difficulty2/10
Novelty6/10
△ Mechanism confirmed, baseline not beaten
2026
Augment SGD or momentum with a state observer that estimates the slowly varying component of minibatch-gradient disturbance from one-step parameter-transition residuals. Cancel the estimated disturbance with feedforward correction, then apply a curvature-dependent robust feedback gain whose closed-loop dynamics satisfy a discrete stability or bounded-gain condition.
Useful7/10
Difficulty6/10
Novelty7/10
✗ Failed on benchmark
2026
Implement a momentum optimizer as a contact Hamiltonian splitting rather than as a direct Euler discretization. Introduce an auxiliary scalar contact state and compose exact kinetic, potential, and damping subflows; this produces a second-order conformal integrator whose modified contact energy should decay more reliably at moderately large learning rates.
Useful7/10
Difficulty5/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Replace a soft auxiliary load-balancing loss with a deterministic router correction that balances expert assignments separately within confidence or score shells. The correction preserves the number of tokens assigned to each shell while making every expert receive equal or nearly equal mass inside each shell, preventing high-confidence tokens from monopolizing a subset of experts.
Useful7/10
Difficulty5/10
Novelty5/10
✗ Failed on benchmark
2026
Replace generic learning-rate selection in decentralized or federated gradient tracking with a low-dimensional minimax search over the exact scalar-mode pole radius. The optimizer chooses the step size that minimizes the worst predicted contraction over the observed graph spectrum and an estimated curvature interval, rather than relying only on conservative global bounds.
Useful7/10
Difficulty5/10
Novelty6/10
✓✓ Beats tuned baseline
2026
Replace a jointly optimized scalar electrostatic potential in a neural PDE solver with a dual flux represented by a Hodge curl correction. The resulting inner problem is a positive quadratic minimization with the divergence constraint satisfied exactly, avoiding unstable primal-dual training dynamics.
Useful7/10
Difficulty6/10
Novelty8/10
✓✓ Beats tuned baseline
2026
Replace raw spatial coordinates supplied to a neural field or PINN by a learnable monotone radial coordinate generated from a positive neural density. The density is trained through the PDE energy or residual after solving for the network weights, allowing the warp to discover where resolution is needed without singularity labels or an analytic interior solution. Near a singular point, a factor s^(q-1) gives a controlled regularity gain, while a positive learned correction redistributes…
Useful7/10
Difficulty5/10
Novelty7/10
✗ Failed on benchmark
2026
Use a learned quasipotential barrier as feedback for optimizer noise and restart control. Increase stochasticity when training is trapped in a high-loss metastable basin and reduce it near a desirable basin, with switching thresholds determined by the estimated barrier rather than by a fixed patience schedule.
Useful7/10
Difficulty6/10
Novelty7/10
✗ Failed on benchmark
2026
Replace independently restarted proximal-gradient or quasi-Newton solves for a composite neural objective with a curvature-recycling Douglas–Rachford loop. The previous proximal state, residual, and limited-memory BFGS curvature pairs are transported to the next proximal center, reducing expensive loss and gradient evaluations while retaining the cheap nonsmooth proximal operation.
Useful7/10
Difficulty5/10
Novelty7/10
✓✓ Beats tuned baseline
2026
Replace a fixed SGD learning rate with a per-update step selected from the positive curvature observed along the proposed direction. The controller estimates the directional Taylor remainder using one or two function evaluations, increases the step when the observed direction is benign, and backtracks only when the update fails a sufficient-decrease test.
Useful7/10
Difficulty5/10
Novelty6/10
✗ Failed on benchmark
2026
Replace a single optimizer trajectory by N parameter particles and optimize the time until the first particle reaches a target loss or reward threshold. Use distinct interaction regimes: bounded normalized interactions should provide only the usual logarithmic extreme-search improvement, whereas unnormalized coherent force accumulation and stochastic pairwise kicks should produce distinct 1/N and 1/(N ln N) first-hit laws.
Useful7/10
Difficulty5/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Replace independently parameterized scalar, vector, and higher-order neural outputs with consecutive spaces of ReLU-power differential forms linked by an exact exterior-derivative layer. The network can then produce curl-free, divergence-free, or more general closed fields by construction, while the complex prevents artificial null-space modes that commonly appear when differential constraints are enforced only through sampled residual losses.
Useful7/10
Difficulty5/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Use dephasing as a depth- or time-dependent control variable rather than a fixed regularizer: early layers retain coherent transport for feature discrimination, while later layers increase dephasing to eliminate unstable high-frequency oscillations. The schedule is selected from an observable spectral or correlation ratio, giving a falsifiable switch point instead of tuning noise blindly.
Useful7/10
Difficulty5/10
Novelty8/10
✓✓ Beats tuned baseline
2026
Represent hidden features with several local orthogonal or unitary frames, and let group-valued transition maps align neighboring experts. Instead of forcing every expert to learn a globally coherent coordinate system, train local experts independently and impose a patching constraint that produces a globally consistent feature field. This is especially suitable for graph neural networks on data with heterogeneous local geometries or mixture-of-experts models whose experts specialize by region.
Useful7/10
Difficulty5/10
Novelty7/10
✗ Failed on benchmark
2026
Use the recent history of generator outputs as a controllable training window instead of fixing the replay-memory depth globally. Estimate how quickly each fitness level improves as more same-level examples enter the window, and increase memory only when the measured escape probability improves enough to justify the extra stale data.
Useful7/10
Difficulty5/10
Novelty6/10
✗ Failed on benchmark
2026
Train a neural policy against task cost while penalizing its induced drift mismatch from a reference policy or offline-data dynamics model. Unlike action-space behavior cloning, the penalty weights deviations by the inverse diffusion covariance, so deviations in highly noisy directions are cheap and deviations in predictable directions are expensive. This gives a principled interpolation between reference preservation and task optimization.
Useful7/10
Difficulty5/10
Novelty6/10
✓✓ Beats tuned baseline
2026
Train a mixture-of-experts router by solving its regularized nonnegative simplex least-squares subproblem with a matrix-free active-set conjugate-gradient method instead of projected gradient or Adam. The router coefficients remain exactly nonnegative and sum to one, while CG rapidly solves each free-set quadratic and the active-set pivots identify sparse expert assignments.
Useful7/10
Difficulty5/10
Novelty7/10