$$\mathcal{L}_{\text{distill}} = (1-\alpha)\,\mathrm{CE}\big(y,\ \sigma(z_s)\big) \;+\; \alpha\, T^2\,\mathrm{CE}\big(\sigma(z_t/T),\ \sigma(z_s/T)\big)$$

Equation 1 · Edge AI & Electronics

Shrink It, Train It Small, or Search for It: The Main Strategies for Small Models, Compared

This equation states an equality: the expressions on both sides have the same value under the article’s assumptions.

The oldest of the four strategies starts from an asset that is already paid for: a large model has already been trained, at whatever cost that took, and the cost is sunk. Compression treats the sunk cost as something to copy rather than repeat. Hinton, Vinyals, and Dean set out the core argument in 2015, motivated by a practical deployment problem with large ensembles: “making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users” [ 1 ] . Their proposed fix — train a small “student” network to match a large “teacher” network’s full output distribution, not merely its top label — is usually written…

Meanings in this article

• $y$: the ground-truth label.
• $\sigma$: the softmax.
• $z_s$: the teacher’s and student’s output logits.
• $T^2$: the square of T; the temperature that softens the distribution.
• $z_t$: the teacher’s and student’s output logits.
• $T$: the temperature that softens the distribution.

Equation guide → · Article →