I'm looking for some references for online multi-class classification problem where the number of classes grows over time. Concretely, the data at time-step $t$ comes in the form $\left(\mathbf{x}_t, y_t\right) \,:\, \mathbf{x}_t\in\mathbb{R}^n,\, y_t\in\left\{1,\dots,K_t\right\}$. Any a new time-step can introduce a new class to the dataset, \ie possibly $t'>t$ could introduce a $K_{t}+1$ class by giving a training sample of the form $\left(\mathbf{x}_{t'}, y_{t'}=K_t+1\right) \,:\, \mathbf{x}_{t'}\in\mathbb{R}^n$.
Any help/pointers would be appreciated :)
