Topic: sequential learning

Learning in real life is often sequential: A learner first learns task A, then task B. If the tasks are related, the learner may adapt the previously learned representation instead of generating a new one from scratch. Adaptation may ease learning task B but may also decrease the performance on task A. Such interference has been observed in experimental and machine learning studies. In the latter case, it is mediated by correlations between weight updates for the different tasks. In typical applications, like image classification with feed-forward networks, these correlated weight updates can

In many learning and decision scenarios, especially sequential settings like mazes or games, it is easy to state an objective function but difficult to compute it, for instance because this can require enumerating many possible future trajectories. This, in turn, motivates a variety of more tractable approximations which then raise resource-rationality questions about whether and when an efficient agent should invest time or resources in computing decision variables more accurately. Previous work has used a simple all-or-nothing version of this reasoning as a framework to explain many phenomen