What does this equation mean?

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 with a softened distribution and a temperature parameter: $\mathcal{L}_{\text{distill}} = (1-\alpha)\,\mathrm{CE}\big(y,\ \sigma(z_s)\big) \;+\; \alpha\, T^2\,\mathrm{CE}\big(\sigma(z_t/T),\ \sigma(z_s/T)\big)$. where $z_t$ and $z_s$ are the teacher’s and student’s output logits, $\sigma$ a softmax, T a temperature that softens the distribution, and y the ground-truth label. The second term is the entire point of the method: it transfers the relative probability the teacher assigns to every wrong answer, not just which answer was right, a far richer training signal per example than a raw label alone provides. That is the rationale, stated plainly by the paper that introduced it.