[FILL: workshop paper title], explained

12 June 2026

This is a stub post. Replace everything on this page with the real text. The equation and code block below exist only to verify that math and syntax highlighting render correctly — delete them too.

The paper asks a simple question: once a model has seen enough unlabelled tokamak signal, how much labelled data do you actually need for the tasks people care about?

Inline math renders like $x_t \in \mathbb{R}^d$, and display math like:

Code renders through Shiki:

def patchify(x, patch_len: int):
    """Split a multichannel signal into non-overlapping patches."""
    T = x.shape[-1] - x.shape[-1] % patch_len
    return x[..., :T].reshape(*x.shape[:-1], -1, patch_len)