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.
Paper: Fair Dynamic Operating Envelopes using Distributed Multi-Period Optimal Power Flow and Jain Index for Active Distribution NetworksarXiv:2608.23444
Use adaptive performance specifications to prevent a neural controller or policy from demanding output changes that exceed bounded actuator amplitude or action-rate limits. The target error envelope tightens when the policy has control authority and relaxes when saturation or rate clipping persists, instead of allowing the controller to destabilize while chasing an infeasible target. This converts actuator clipping into an explicit slow state that can be used by reinforcement-learning policies…
Replace uniform or purely loss-driven update allocation with a scheduler that targets both the mean update rate and the temporal variance of updates for each parameter group, task, or expert. At every training step, assign the available minibatch slots or accelerator workers to groups with the largest weighted deficits, preventing starvation while avoiding highly bursty update streams that can produce optimizer oscillations.
Train the critic on the action that the environment actually received after safety filtering, not only on the actor's nominal action. Prioritize transitions whose estimation residual, barrier proximity, or filter intervention is large, so replay concentrates on the distribution shift introduced by the safety controller instead of repeatedly sampling benign nominal behavior.
Use the mode-wise instability condition as a controller for a learned cross-channel transport gain. During training or inference, estimate the linearized feature dynamics and adjust the chemotactic strength to remain below a stability margin for robust processing, or deliberately cross the threshold during a controlled pattern-forming stage. This replaces blind gain tuning with a measurable dynamical criterion.
When training a neural state-space model, SSM, or recurrent world model from trajectories, constrain the data-generation policy or augmentation process to satisfy both a Hankel-rank condition and a task-weighted frequency-coverage condition. The rank condition prevents unidentifiable dynamics, while the frequency condition concentrates samples at frequencies that affect the target prediction horizon, tracking objective, or closed-loop controller instead of merely producing broadband-looking…
Add a causal gate to a neural safe-RL controller that distinguishes between evidence observed before a potentially irreversible action and evidence generated by that action itself. The policy may switch from a conservative controller to a model-specific aggressive controller only when the precommitment likelihood ratio against every dangerous alternative exceeds a threshold; otherwise it must choose an action with a verified safe continuation.
Use the external field H as a slowly swept control variable for a neural module, loss coefficient, or optimizer gain, and deliberately retain the resulting branch memory instead of replacing it with an instantaneous equilibrium update. Forward and backward sweeps produce a hysteresis loop whose shape diagnoses first-order-like training transitions, while controlled disorder changes the loop area and can suppress harmful branch dependence. This supplies a continuation-based curriculum or…
Retain the iteration at which each state enters each modal winning set and use that integer as a dense training target for a neural critic. The policy is additionally encouraged to choose transitions that decrease every finite modal distance, supplying progress information even when the environment reward is sparse.
Use the two-species phase diagram as a training controller: gradually increase one-way mutation \(\lambda\) and mutant suppression \(\mu\) while monitoring whether feature activity approaches the absorbing boundary. This creates a measurable curriculum from independent learning to controlled competition instead of turning on strong destructive interactions at initialization.