ML: Training dynamics

Machine-learning ideas tagged Training dynamics in the ML taxonomy of the Math2NN corpus.

Failed on benchmark 2026

Risk-budgeted MoE capacity reservations

Replace the single global MoE capacity factor with expert-specific capacity reservations chosen from a small reliability menu. Experts with highly variable or operationally important token loads receive larger robust buffers, while predictable experts run closer to their mean load. This should reduce token dropping and padding waste simultaneously, especially under distribution shift or bursty routing.

Useful7/10
Difficulty5/10
Novelty5/10
Paper: A Robust Chance Constrained Approach to Surgery Scheduling arXiv:2608.03931
Mechanism failed 2026

Reset-Integral Sliding Optimizer

Add a scalar integral/sliding variable and a resettable auxiliary state to parameter optimization. The sliding controller rejects bounded gradient perturbations, while resetting the auxiliary state prevents accumulated momentum or integral windup; the reset mechanism is designed not to alter the reaching dynamics of the sliding surface.

Useful7/10
Difficulty5/10
Novelty8/10
Paper: Input-to-State Stability of Reset-Integral Sliding Mode Control for Linear Systems arXiv:2608.03802
Failed on benchmark 2026

Chemotactic Feature Transport Layer

Replace part of a CNN or continuous-depth feature block with two coupled feature fields. One field is transported up gradients of the other through a conservative cross-gradient flux, creating adaptive spatial organization that ordinary diffusion or symmetric convolution cannot produce. The coupling strength and dominant wavelength are controlled by a directly testable linear-instability boundary.

Useful7/10
Difficulty6/10
Novelty8/10
Paper: Pattern formation: reactivity is not necessary for chemotaxis--driven instabilities arXiv:2608.03685
Mechanism failed 2026

Partitioned Gain-Phase Stable Neural Feedback

Build a recurrent or equilibrium network as a feedback interconnection of heterogeneous blocks, certifying some blocks through induced-gain bounds and others through phase or sector bounds. This avoids imposing a uniformly small Lipschitz constant on all blocks: dissipative or strongly contractive blocks use gain certificates, while approximately skew, oscillatory, attention-like, or state-space blocks use phase certificates. The network is accepted only when the local certificates satisfy the…

Useful7/10
Difficulty6/10
Novelty7/10
Paper: Partitioned Mixed Small Gain-Phase Decentralized Stability Criterion for Power Systems arXiv:2608.03641
Failed on benchmark 2026

Braess-Aware Graph Rewiring

Use the paper's saddle-node sensitivity mechanism to decide which message-passing edges should be added, strengthened, or rejected. In a graph neural ODE, neural consensus layer, or recurrent graph block, estimate the critical coupling at which node representations become phase-locked or contractive, then prefer candidate edges whose predicted sensitivity lowers that threshold. This avoids the assumption that more connectivity always improves propagation and gives a topology-aware alternative…

Useful7/10
Difficulty7/10
Novelty8/10
Paper: Predicting the occurrence of Braess paradox in the synchronization threshold of coupled oscillator systems arXiv:2608.03594
✓✓ Beats tuned baseline 2026

Collective-Mode De-Gennes Optimizer

Replace a single global learning rate with mode-dependent rates determined by the static correlation structure of recent parameter updates or hidden-state updates. Correlated modes are treated as collective diffusive modes: their effective relaxation rate is reduced in proportion to their structure-factor amplitude, so the optimizer accelerates weakly correlated modes while damping collective slow modes. The method also supplies a diagnostic for when the Markovian approximation is invalid and…

Useful7/10
Difficulty6/10
Novelty7/10
Paper: The intermediate scattering function of an interacting adlayer as a characteristic function: a closed-form theory of Ising lattice-gas surface diffusion arXiv:2608.03398
✓✓ Beats tuned baseline 2026

Escape-Threshold Learning-Rate Controller

Use bounded-noise escape as a measurable stability transition to adapt the learning rate or recurrent integration step before catastrophic loss of confinement. Periodically estimate the disturbance radius at which the current training dynamics exits its stable region, then adjust the step size to maintain a fixed safety margin.

Useful7/10
Difficulty6/10
Novelty9/10
Paper: From Flows to Maps: Sampling Laws for Attractor Intensity and Bounded-Noise Escape arXiv:2608.02933
Mechanism failed 2026

Energy-Adaptive Inertial Optimizer

Replace constant friction in a second-order neural-network optimizer by a scalar damping coefficient that grows as a power of the current parameter energy plus velocity energy. This should selectively damp large oscillations and unstable excursions while preserving lower friction during small, potentially useful movements.

Useful7/10
Difficulty4/10
Novelty7/10
Paper: Dynamics of Fractional Wave Equations with Nonlocal Damping arXiv:2608.02842
✓✓ Beats tuned baseline 2026

Two-Channel Fractal Renormalization Network

Replace ordinary depth-wise feature propagation by a ternary hierarchical block that recursively aggregates three child representations while maintaining separate neutral and defect channels. The block is initialized from the Sierpinski six-vertex recursion, then optionally learns a bounded correction. The neutral channel preserves the paper's cubic mixing law, while the defect channel provides a controlled route for long-range and nonlocal interactions.

Useful7/10
Difficulty6/10
Novelty7/10
Paper: Fractal deconfinement and confinement in Sierpinski ice arXiv:2608.02741
Failed on benchmark 2026

Relevant-Noise RG Curriculum

Inject weak, unpostselected stochastic perturbations into activations, attention links, or recurrent transitions, but scale their strength according to effective computational size. The schedule is designed so that noise is initially a weak perturbation and becomes dominant only beyond a controlled depth or sequence length, producing a measurable crossover rather than uncalibrated constant dropout.

Useful7/10
Difficulty4/10
Novelty6/10
Paper: Universal crossovers in weakly-monitored quantum critical states arXiv:2608.02716
Mechanism confirmed, baseline not beaten 2026

Forward-Intersection Spectral Latent Dynamics

Replace the raw transition matrix of a Koopman-inspired latent model or linear state-space model by its restriction to a data-derived forward-compatible subspace. The subspace is obtained by repeatedly intersecting the current latent dictionary with its image under the learned dynamics, suppressing directions that generate spurious or unsupported eigenmodes while retaining nonzero Koopman modes represented by the dictionary.

Useful7/10
Difficulty5/10
Novelty7/10
Paper: Beyond Invariant Dictionary: Data-Driven Koopman Spectral Recovery with Filtered Extended Dynamic Mode Decomposition arXiv:2608.02661
Mechanism confirmed, baseline not beaten 2026

Van der Corput progressive expert scheduler

Replace random or greedy one-expert-at-a-time activation with a deterministic binary van der Corput sequence. At each training or inference step, the schedule chooses an expert whose cumulative usage remains close to its proportional target, while recursively balancing nested expert groups rather than only balancing individual experts.

Useful7/10
Difficulty5/10
Novelty7/10
Paper: House-monotone multi-level apportionment has logarithmic quota discrepancy arXiv:2608.02559
Failed on benchmark 2026

Smooth Barrier Tube Controller

Treat undesirable neural-network states as obstacles and steer training or inference away from them with a smooth distance barrier while preserving a nominal loss descent direction. The barrier can protect against exploding activations, excessive attention concentration, unsafe controller outputs, or leaving a certified representation region without introducing discontinuous gradient clipping.

Useful7/10
Difficulty4/10
Novelty5/10
Paper: Safe and robust tube-based path-following for robot navigation arXiv:2608.02530
Mechanism confirmed, baseline not beaten 2026

Reachset-Conformance Noise Calibration

Calibrate process and observation uncertainty bounds by requiring a learned neural dynamical model to contain calibration trajectories in its reachable sets, instead of fitting a Gaussian noise model. The resulting bounds can control an uncertainty-aware loss, trigger teacher forcing or re-observation, and identify latent coordinates whose dynamics are not adequately modeled. This transfers the paper's conformance principle into a falsifiable training monitor and adaptive rollout schedule.

Useful7/10
Difficulty5/10
Novelty8/10
Paper: A General Set-Based Framework for Cognitive State Estimation: Theory and Application to Conditionally Automated Driving arXiv:2608.02308
Mechanism confirmed, baseline not beaten 2026

Komuro Time-Warp Expansivity Regularizer

Apply Komuro-style expansivity to a continuous-time neural latent flow by requiring distinct latent trajectories to separate even when the second trajectory is allowed an arbitrary increasing time reparametrization. This targets neural ODE world models and irregularly sampled sequence models, where ordinary pointwise separation can mistake clock-speed differences for different states.

Useful7/10
Difficulty6/10
Novelty8/10
Paper: Komuro Expansivity and Periodic Orbit Growth for Multi-Singular Hyperbolic Flows arXiv:2608.02186
Failed on benchmark 2026

Smooth Spectral Muon

Replace the exact matrix-polar normalization in Muon with the smoothed feedback \(h_\epsilon(M)=M(M^\top M+\epsilon I)^{-1/2}\). This retains singular-vector-aware updates and approximately unit-normalizes dominant spectral modes, but avoids unstable behavior when the momentum matrix is rank deficient or has tiny singular values.

Useful7/10
Difficulty5/10
Novelty4/10
Paper: A Continuous-Time Analysis of Smoothed Matrix-Polar Spectral Gradient Flows for Muon-Type Optimization arXiv:2608.01911
Failed on benchmark 2026

PIPO-PITO bounded recurrent gain

Build a positive continuous-depth RNN or state-space layer in which a nonnegative recurrent-input gain is generated by a PITO controller. If sustained large gain produces sustained large hidden-state output through a PIPO plant, the controller automatically decreases the gain, preventing runaway recurrent dynamics without requiring a globally tiny fixed gain. The construction predicts a quantitative attenuation threshold and exponential decay rate when the hidden output stays above that…

Useful7/10
Difficulty5/10
Novelty7/10
Paper: On input-output persistency and the interconnection of positive nonlinear systems arXiv:2608.01699
Mechanism failed 2026

Analytic Markov-Routing Lyapunov Controller

Use a finite-state Markov router to select recurrent or expert Jacobians, and regularize or optimize the router through the top Lyapunov exponent computed from state-conditioned projective statistics. The paper's mechanism predicts that this exponent varies smoothly with routing probabilities when the transition matrix is primitive and the dominant exponent is simple, while loss of primitivity, resonance, or exponent collision marks a detectable boundary where routing gradients may become…

Useful7/10
Difficulty6/10
Novelty7/10
Paper: Analyticity of Lyapunov Exponents for Mixed Markov Quasi-Periodic Cocycles arXiv:2608.01569
Mechanism confirmed, baseline not beaten 2026

BDD-Certified Modular Equilibrium Network

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
Paper: Decentralized Control Synthesis in IBR-Dominated Power Systems: A Block Diagonal Dominance Based Approach arXiv:2608.01236
Failed on benchmark 2026

Hyperbolic Shadowing RNN

Constrain a recurrent transition so that its dynamically relevant invariant subspaces have no eigenvalues near the unit circle, separating contracting memory directions from expanding prediction directions. Add a pseudo-orbit consistency loss so that trajectories generated with bounded transition perturbations remain close to clean trajectories, as expected from hyperbolic shadowing.

Useful7/10
Difficulty6/10
Novelty6/10
Paper: Shadowing Endomorphisms of Compact Groups arXiv:2608.00955
Failed on benchmark 2026

Weakly Normally Hyperbolic Cyclic Optimizer

Augment an optimizer with a periodic phase and deliberately use a cyclic learning-rate or momentum forcing whose averaged dynamics have an attracting low-dimensional set. Treat the resulting periodic parameter orbit as an invariant torus and tune the schedule so transverse contraction dominates tangential sensitivity and minibatch perturbations. The goal is a robust, phase-locked training orbit that explores parameter space without losing attraction toward a useful solution manifold.

Useful7/10
Difficulty5/10
Novelty7/10
Paper: Weakly Normally Hyperbolic Invariant Tori: Persistence and an Averaging Principle arXiv:2608.00812
Mechanism failed 2026

Residual-to-Symbolic Neural Pruning

Use a two-stage residual augmentation loop: first let a residual network explain model mismatch, then project its learned vector field onto a physically constrained candidate library and replace the flexible residual with the accepted sparse terms. This turns an unconstrained neural correction into a low-complexity dynamical law that is easier to roll out over long horizons and can expose unsupported hidden-state explanations.

Useful7/10
Difficulty5/10
Novelty6/10
Paper: SPIRAL-PO: Symbolic Identification of Partially Observed Nonlinear Dynamics with Application to Rotating Machinery arXiv:2608.00466
Mechanism failed 2026

Onsager–Casimir Response Regularizer

Train a sequence model so that measured perturbation responses and spontaneous hidden-state correlations satisfy the paper's off-diagonal fluctuation–response identity. This discourages arbitrary non-reciprocal dynamics while preserving a controlled antisymmetric response that can encode directional temporal dependencies.

Useful7/10
Difficulty6/10
Novelty8/10
Paper: Memory with Onsager-Casimir symmetry: Rotating particle in a viscoelastic fluid arXiv:2608.00344
Mechanism failed 2026

Critical-Gain Covariance Controller

Track the covariance of a small recurrent population state and regulate its effective gain before finite-size fluctuations diverge. The controller uses the covariance Jacobian eigenvalues from the paper, making the distance to criticality an explicit adaptive regularization signal for recurrent or state-space neural networks.

Useful7/10
Difficulty5/10
Novelty6/10
Paper: Mechanistic bridges from receptors to whole-brain dynamics: mean-field reductions, validity domains, and computational trade-offs arXiv:2608.00306