Universal Multiclass Transductive Online Learning

In three linesTheoretical paper on universal multiclass transductive online learning with unbounded label space. Characterizes learnability: only two possible optimal rates (bounded or logarithmic). Introduces LCLL tree combinatorial structure and extends results to agnostic and stochastic settings.

Summary generated by Claude — human-verified