First-order meta-learning capability added to ML platform

AI Impact Summary

This change introduces support for first-order meta-learning algorithms within the platform's ML training capabilities. By enabling first-order gradient approximations, training across tasks can run with reduced computational overhead, accelerating meta-learning experiments and rapid adaptation to new tasks. Teams should anticipate updates to training scripts to opt into first-order mode and validate convergence and accuracy, as removing second-order terms can alter optimization dynamics. Existing pipelines relying on second-order gradients may require benchmarking and potential tuning of hyperparameters to maintain performance.