No abstract

Temporal di erence (TD) methods constitute a class of methods for learning predictions in multi-step prediction problems, parameterized by a recency factor . Currently the most important application of these methods is to temporal credit assignment in reinforcement learning. Well known reinforcement learning algorithms, such as AHC or Q-learning, may be viewed as instances of TD learning. This paper examines the issues of the e cient and general implementation of TD( ) for arbitrary , for use with reinforcement learning algorithms optimizing the discounted sum of rewards. The traditional approach, based on eligibility traces, is argued to su er from both ine ciency and lack of generality. The TTD (Truncated Temporal Di erences) procedure is proposed as an alternative, that indeed only approximates TD( ), but requires very little computation per action and can be used with arbitrary function representation methods. The idea from which it is derived is fairly simple and not new, but probably unexplored so far. Encouraging experimental results are presented, suggesting that using > 0 with the TTD procedure allows one to obtain a signi cant learning speedup at essentially the same cost as usual TD(0) learning.

Despite the rapid growth of other types of social media, Internet discussion forums remain a highly popular communication channel and a useful source of text data for analyzing user interests and sentiments. Being suited to richer, deeper, and longer discussions than microblogging services, they particularly well reflect topics of long-term, persisting involvement and areas of specialized knowledge or experience. Discovering and characterizing such topics and areas by text mining algorithms is therefore an interesting and useful research direction. This work presents a case study in which selected classification algorithms are applied to posts from a Polish discussion forum devoted to psychoactive substances received from home-grown plants, such as hashish or marijuana. The utility of two different vector text representations is examined: the simple bag of words representation and the more refined embedded global vectors one. While the former is found to work well for the multinomial naive Bayes algorithm, the latter turns out more useful for other classification algorithms: logistic regression, SVMs, and random forests. The obtained results suggest that post-classification can be applied for measuring publication intensity of particular topics and, in the case of forums related to psychoactive substances, for monitoring the risk of drug-related crime.