ANYTIME NEURAL NETWORK A VERSATILE TRADEOFF BETWEEN COMPUTATION AND ACCURACY

Anytime prediction:

• An anytime predictor produces a coarse result quickly and then continues to refine it until the test-time computational budget is depleted.

Gap:

• Thus, it can be difficult for such applications to choose between slow predictors with high accuracy and fast predictors with low accuracy.
• At any test-time budget, the anytime predictor has a valid result for the sample, and the more budget is spent, the better the prediction is.

This paper weighted the sum the loss generated by each sub-network composed previous L proceeding layer.

---

• ← Previous Post
• Next Post →