✗ Mechanism failed
2026
Cluster recurrent modules or MoE experts by the geometry of their observed finite-horizon input-output behaviors rather than by parameter distance. Train one shared optimizer/controller or low-rank adapter per cluster while retaining module-specific parameters and routing. This should reduce control and optimizer overhead without merging modules whose temporal responses are dynamically incompatible.
Useful8/10
Difficulty5/10
Novelty8/10
△ Mechanism confirmed, baseline not beaten
2026
Replace a standard ReLU surrogate with an input convex neural network whose hidden-to-hidden weights are constrained to be nonnegative. The network remains piecewise linear and expressive, but its convexity allows downstream minimization to use continuous ReLU epigraph constraints instead of binary activation variables, potentially eliminating the integrality bottleneck of neural optimization.
Useful8/10
Difficulty5/10
Novelty5/10
✗ Failed on benchmark
2026
Replace an instantaneous diagonal optimizer with a causal convolution of recent gradients, where cross-layer or cross-module gradient correlations define a finite-memory Onsager response matrix. Estimate the response at several parameter-block pairs and lags, integrate it to obtain a finite-time transport matrix, and use its regularized inverse or symmetric part to precondition the update. This targets optimization regimes in which gradients propagate between blocks with measurable delay, such…
Useful8/10
Difficulty6/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Compress the matrix gradient or momentum before applying Muon's polar LMO, and maintain an error residual in the uncompressed gradient space. The residual prevents systematic sign quantization bias from accumulating, unlike error feedback applied after the nonlinear polar/sign operation. This is suitable for distributed training because workers communicate one sign bit per matrix entry while the server still applies a matrix-aware Muon step.
Useful8/10
Difficulty5/10
Novelty6/10
✓✓ Beats tuned baseline
2026
Use a consensus-coupled optimizer for replicated model parameters, but construct every communication perturbation so that the all-ones consensus direction remains in the Laplacian null space. This prevents topology noise, pruning, or heterogeneous communication weights from changing the common parameter trajectory while still allowing disagreement modes to be damped.
Useful8/10
Difficulty5/10
Novelty6/10
✗ Failed on benchmark
2026
Group W consecutive diffusion or flow-model loss terms and approximate every intermediate parameter Jacobian by a time-weighted interpolation of the Jacobians at the group’s two endpoints. Sum the intermediate upstream signals into two endpoint cotangents, then perform only two full DiT backward passes instead of W. Add a cosine-similarity gate comparing predicted and actual intermediate velocity changes so that groups violating the local-linearity assumption use exact backpropagation.
Useful8/10
Difficulty6/10
Novelty7/10
✗ Failed on benchmark
2026
Treat neural modules as interconnected dynamical subsystems and estimate the gain from every module input to every neighboring module output. Replace an expensive global Jacobian spectral-radius calculation by decentralized directed-cycle tests inside clusters and path-gain tests between clusters. Penalizing violations during training should prevent exploding recurrent trajectories while retaining less conservative behavior than constraining every individual block independently.
Useful8/10
Difficulty6/10
Novelty6/10
△ Mechanism confirmed, baseline not beaten
2026
Replace a deep feed-forward block by the fixed point z=phi(Wz+Vx+b), with the recurrent weight W constrained so that the fixed point is unique for every input. The same condition makes forward fixed-point iteration stable and makes implicit differentiation well-conditioned, allowing depth-independent memory usage while providing a measurable spectral failure boundary.
Useful8/10
Difficulty5/10
Novelty4/10
✓✓ Beats tuned baseline
2026
Represent communicating layers, experts, or distributed workers as nodes of a weighted graph and apply strong corrective updates only to a small pinned subset. Select pins by the increase they produce in the grounded Laplacian smallest eigenvalue, because this spectral gap predicts the decay rate of representation disagreement.
Useful7/10
Difficulty6/10
Novelty6/10
✗ Failed on benchmark
2026
Partition neural-network parameters into competing blocks, such as LoRA adapters, mixture-of-experts heads, or task-specific heads, and update each block by minimizing its local quadratic model while holding the other blocks fixed. Use the exact Jacobi coupling spectral radius to decide whether simultaneous updates are stable; near the boundary, apply damping or fall back to sequential Gauss-Seidel updates.
Useful7/10
Difficulty5/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Add sparse directed coupling between parallel neural modules, recurrent states, or distributed replicas so that each module is driven toward a common trajectory without forcing an undirected or balanced communication graph. Select n-1 directed paths per strongly connected component and assign gains using the estimated Lipschitz bound of the uncoupled module; activate the coupling only when its graph-certified strength exceeds the predicted synchronization threshold.
Useful7/10
Difficulty5/10
Novelty7/10
✓✓ Beats tuned baseline
2026
Replace the final layer of a neural predictor with Bayesian linear regression over deterministic trigonometric features, retaining a computable posterior variance and a high-probability confidence envelope over the full bounded input domain. Use this envelope to reject unsafe actions, downweight uncertain training targets, or restrict optimizer updates in regions where the network is extrapolating.
Useful7/10
Difficulty5/10
Novelty6/10
✗ Failed on benchmark
2026
Replace a diagonal learning-rate or preconditioner matrix with a small full block matrix and communicate a worker's updated gradient or parameter only when its local state has drifted sufficiently from the last communicated state. Jointly select the block preconditioner and the largest safe trigger threshold using robust Lyapunov inequalities over several empirical Hessian or Gauss-Newton matrices. The expected gain is fewer synchronization events without the instability normally caused by…
Useful7/10
Difficulty7/10
Novelty7/10
✗ Failed on benchmark
2026
Run multiple optimizer workers, neural-network branches, or expert replicas with delayed parameter messages, using diffusive coupling for agreement and a separately slowed local gradient vector field. The delay should preserve the collective descent direction to first order while multiplying its evolution speed by a predictable factor, allowing communication-delay robustness to be tested independently from ordinary stale-gradient behavior.
Useful7/10
Difficulty6/10
Novelty7/10
✓✓ Beats tuned baseline
2026
Compress only the difference between the current client gradient and a persistent control variate, rather than compressing the full gradient. As the control variate tracks the client gradient, the residual shrinks and the same communication budget produces less compression noise than direct gradient quantization.
Useful7/10
Difficulty4/10
Novelty5/10
△ Mechanism confirmed, baseline not beaten
2026
Replace the standard diagonal or identity preconditioner used when solving an implicit neural layer with a coarse/fine Schur-complement preconditioner. The hidden state is decomposed into a low-dimensional coarse subspace and its orthogonal complement; the coarse interaction is solved accurately, while the fine block receives a damped approximate inverse. The method is especially suitable for deep equilibrium models, implicit MLPs, and Newton or quasi-Newton training of residual dynamics.
Useful7/10
Difficulty6/10
Novelty7/10
✗ Failed on benchmark
2026
Add a response-spectrum monitor to recurrent, state-space, or deep-equilibrium networks by treating products of hidden features as composite observables. Estimate the full susceptibility and a bare susceptibility, reconstruct an irreducible interaction vertex, and damp the state update whenever the leading Bethe–Salpeter eigenvalue approaches one. This targets collective failure modes that ordinary single-feature Jacobian checks can miss.
Useful7/10
Difficulty5/10
Novelty8/10
△ Mechanism confirmed, baseline not beaten
2026
Partition a neural network into N interacting modules and constrain the Jacobian of its implicit residual map to be block diagonally dominant. Each module can compute its update locally while cross-module coupling is monitored through a normalized block-row margin. The certificate guarantees local nonsingularity of the equilibrium equations and predicts a sharp loss of robustness when the largest BDD ratio approaches one.
Useful7/10
Difficulty6/10
Novelty7/10
✗ Failed on benchmark
2026
Keep the empirically effective post-LMO sign update, but reject it whenever a fresh minibatch estimates that it is poorly aligned with the gradient. Fall back to the gradient-side error-feedback candidate in those cases. This converts the paper's constructive divergence warning into an inexpensive runtime safeguard rather than assuming that any sign placement is universally safe.
Useful7/10
Difficulty4/10
Novelty7/10
✗ Failed on benchmark
2026
Design distributed training workers so checkpoint service is anonymous: every active writer receives a throughput determined only by the current number of active writers, not by worker identity, age, or phase. For identical compute periods and checkpoint durations shorter than the period, this removes pairwise phase attraction and prevents deterministic checkpoint synchronization; controlled timing jitter can then be added when rapid phase mixing is desired.
Useful7/10
Difficulty5/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Use measured local Jacobian growth to set the variance of dropout, feature noise, or stochastic-depth perturbations, implementing the paper's fluctuation-response idea that multiplicative noise is tied to the positive scrambling or Lyapunov rate. The controller maintains a target growth regime instead of applying a fixed noise schedule throughout training. It predicts a stability transition when the estimated growth rate crosses zero and a variance-growth proportionality that can be tested…
Useful7/10
Difficulty4/10
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
✗ Failed on benchmark
2026
Replace ordinary graph propagation by diffusion with a positive node-dependent mass matrix \(\mathbf V\), so high-volume nodes update slowly and low-volume nodes update rapidly. Use node volumes as fixed metadata, a function of degree, or learned positive gates; this makes the architecture sensitive to dynamical localization that degree-normalized GCNs cannot represent.
Useful7/10
Difficulty5/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Use small-gain diagnostics to jointly learn module normalization and a communication partition rather than imposing a fixed global spectral constraint. Clusters should be formed around high-gain feedback loops, because grouping weakly related modules cannot improve the certificate and only adds bookkeeping.
Useful7/10
Difficulty7/10
Novelty7/10