I'm currently working on a paper on continual learning in deep learning and am reading a novel paper on a method called "Dark Experience Replay". In the first equation where they describe the function that needs to be optimized, a lowercased curved L is used within an expected value.
In my interpretation the function describes a Laplace Transform (which can also be described as an expected value, as far as I know), therefore the character would indicate the $\mathrm{e}^y$, and $f(x)$, but I'm not sure if this checks out...
Can anybody help me out?
Edit: For further clarification, should have explained the rest of the formula: D_t is a i.i.d. distribution of tasks that consist of input samples x and ground-truth labels y. f is the function that is used for classification with parameters theta and input x
In machine learning, $\mathcal{l}$ usually means loss function.
It measures how much error is incurred when we use $f_\theta(x)$ to predict $y$.
You are trying to find the parameter $\theta$ that minimizes the expected loss.