This section presents the key components of the proofs of our main results. Recall that the goal of the analysis is to understand the rate at which the tail-averaged SGD iterates ¯wS,T approach the risk minimizer w∗. The main error decomposition u...
Our proof closely follows that of Dieuleveut et al. (2017, Theorem 1), with the key differences that
Key to multitask learning is exploiting relationships between different tasks to improve prediction performance. If the relations are linear, regularization approaches can be used successfully. However, in practice assuming the tasks to be linearl...
In this section, we study two special instances of Problem 3, namely variational inequalities and minimization problems. Moreover, for variational inequalities, we prove an additional result, showing that a suitably defined merit function  goes...
Credentials & highlights
- Project Views