Match reused work to changing inputs and the outcome you need

A sparse update, a robot action, and a caption choice all invite reuse of earlier work. Yet the next input may change what is useful, and a faster calculation or a higher benchmark score may miss the outcome the user needs.

These studies suggest a bounded rule: identify what persists as the workload changes, then measure the intended outcome in that workload. Their evidence concerns computation time, task success, and candidate selection. It does not establish a reduction in stored memory.

Suppose a changing input requires a matrix calculation at every step. If only a few input rows change, recalculating the full result seems wasteful.

A sparse matrix update study instead examines work tied to those changed rows, which it calls the dirty set. It also asks whether a prepared schedule can be reused when that set moves.

This makes the practical problem sharper: saving preparation time helps only if the prepared schedule still suits the next update and its execution cost.

A robot faces a different moving target. While it computes an action, an object on a conveyor may leave the position where the robot last saw it.

RealtimeWAM studies the delay from observation to action in models that process images and predict actions. It reuses some previously computed observation features and adjusts later computation to the predicted movement.

Here the reader needs to ask about both the time saved and whether the robot still completes its task.

A caption selector presents a third version of the problem. It can score several descriptions of an image and choose one, but a good score on a set of preference questions might say little about the descriptions its own generator produces.

The reward-modeling study retains a large vision and language model while training a small scoring component. Across all three cases, the shared question is when retained work helps under changing inputs, as judged by the outcome each system is meant to deliver.