Does This Gradient Spark Joy?
This addresses computational inefficiency in reinforcement learning and deep learning training, offering a potential paradigm shift for speculative decoding in training, though it is incremental in its specific application.
The paper tackles the inefficiency of policy gradient methods by introducing a forward-pass signal called 'delight' to screen samples, and the Kondo gate skips most backward passes while retaining nearly all learning quality, with gains increasing as problems get harder.
Policy gradient computes a backward pass for every sample, even though the backward pass is expensive and most samples carry little learning value. The Delightful Policy Gradient (DG) provides a forward-pass signal of learning value: \emph{delight}, the product of advantage and surprisal (negative log-probability). We introduce the \emph{Kondo gate}, which compares delight against a compute price and pays for a backward pass only when the sample is worth it, thereby tracing a quality--cost Pareto frontier. In bandits, zero-price gating preserves useful gradient signal while removing perpendicular noise, and delight is a more reliable screening signal than additive combinations of value and surprise. On MNIST and transformer token reversal, the Kondo gate skips most backward passes while retaining nearly all of DG's learning quality, with gains that grow as problems get harder and backward passes become more expensive. Because the gate tolerates approximate delight, a cheap forward pass can screen samples before expensive backpropagation, suggesting a speculative-decoding-for-training paradigm.