Curricula, sampling schedules, pseudo-labeling and multi-task mixing, with the standard training recipe as the control.
Use the envelope's phase transition to choose whether clipping should primarily control update energy or preserve the raw gradient and reduce clipping bias. In the energy-dominated regime, regulate the retained update energy; in the bias-dominated regime, regulate the removed-gradient residual and monitor rare outliers explicitly.
Train one functional flow-matching network against conditional velocity targets formed from randomly varying finite-rank reconstructions, including sensor sets that are not nested across training examples. Decode predictions from two sensor layouts into a common function representation and add a cross-layout consistency penalty. The paper's convergence result predicts that this remains statistically valid as reconstruction error decreases, unlike methods that implicitly rely on changing grids…
Train masked predictors with an explicit mixture of high-visibility masks, low-visibility masks, and a small atom at the fully masked input. High-visibility masks preserve ordinary denoising quality, while low-visibility and fully masked examples force the network to learn global mode frequencies that are invisible when nearly all context is shown. Tune the low-visibility mass using unconditional-mode recovery as an auxiliary validation metric.
Replace ordinary low-precision multiply-add accumulation in selected neural-network reductions with a two-word floating-point accumulator updated by the paper's branch-free DW-FMA network. The high word retains the main sum and the low word stores the rounding residual, improving cancellation behavior without the control-flow divergence of conditional compensated summation.
Use a small set of known or trusted pairwise costs to estimate the effective entropic temperature of a Sinkhorn attention or mixture-of-experts routing layer directly from its observed transport plan. This provides a calibration controller that can detect over-concentrated routing and adjust epsilon without backpropagating through a costly temperature search.
Construct training and evaluation examples whose every (d-1)-variable marginal is exactly independent, but whose full d-variable distribution contains a parity interaction. This isolates genuine high-order reasoning from shortcuts based on pairwise or lower-order statistics and can expose whether attention or MLP architectures learn the intended interaction.
Use the shared-marginal KL projection to turn token-to-expert routing into a low-rank, exactly balanced assignment rather than relying only on an auxiliary load-balancing penalty. Tokens retain normalized routing distributions while the shared latent marginal enforces consistent aggregate usage across two independently learned routing factors.
Use the paper's attribution converse to calibrate watermark strength and sequence length for a registry of N users, rather than tuning detection and attribution thresholds independently. A dual controller allocates a per-token information and KL budget so that the learned key information approaches the minimum required for reliable attribution, avoiding both underpowered marks and unnecessarily visible perturbations.
Adapt the slope of each spiking neuron's surrogate derivative using the normalized entropy of its block's attention distribution. High centered entropy uncertainty increases the slope, while low uncertainty decreases it, and a dead zone holds the default slope fixed for ordinary fluctuations. The adaptation exists only in backpropagation, so the forward spike function, parameter count, and inference cost remain unchanged.
Represent selected activations, weights, or optimizer accumulators as four floating-point limbs and evaluate products through tensor-core matrix multiplications encoding limb convolution. Retain the convolution components during reductions and renormalize only at block boundaries, avoiding branch-heavy multi-double arithmetic inside every multiply-add.