Citation Context

... for modeling purposes is the problem of active learning, which is a fairly well-studied area in the machine learning literature, and numerous approaches have been proposed to addressing this problem =-=[23]-=-, [24], [36], [58]. We believe that applying active learning methods to address the nonintrusiveness issue constitutes another interesting research opportunity. 3.6 Flexibility Most of the recommendat...

Citation Context

...icates, it is very dif cult to generate ambiguous cases that would help create a highly accurate classi er. Based on this observation, some duplicate detection systems used active learning techniques =-=[24]-=- to automatically locate such ambiguous pairs. Unlike an “ordinary” learner that is trained using a static training set, an “active” learner actively picks subsets of instances from unlabeled data, wh...

Citation Context

... of these cases will lead to improvement of the coverage or accuracy of the rule set. The amount of hand tagging required can be greatly reduced through selective sampling, a form of active learning (=-=Cohn et al. 1994-=-, Lewis and Gale 1994, Dagan and Engelson 1996). When the rule set is empty, the best that can be done is to select instances at random from a reservoir of untagged instances. If the instance space is...

Citation Context

...ing, in which examples are acquired incrementally and the system attempts to use what it has already learned to limit training by selecting only the most informative new examples for the user to rate =-=[8]-=-. Specific techniques for applying this to text categorization have been developed and shown to significantly reduce the quantity of labeled examples required [17, 18]. A slightly different approach i...

Citation Context

...iments such as this, query synthesis may be a promising direction for automated scientific discovery. 2.2 Stream-Based Selective Sampling An alternative to synthesizing queries is selective sampling (=-=Cohn et al., 1990-=-, 1994). The key assumption is that obtaining an unlabeled instance is free (or inexpensive), so it can first be sampled from the actual distribution, and then the learner can decide whether or not to...

Citation Context

...d in active learning have been proven to speed up significantly the learning of machines (e.g., [25]–[31]) and even to allow performance on generalization which are not possible with passive learning =-=[32]-=-. Yet, these techniques were developed for applications in which the mapping to be learned was clean and typically presented as preprocessed well-prepared datasets. They are also typically based on ma...

Citation Context

...f data acquisition costs has seen increasing research attention, both explicitly (e.g., cost-sensitive learning [31], utility-based data mining [19]) and implicitly, as in the case of active learning =-=[5]-=-. Turney [31] provides a short but comprehensive survey of the different sorts of costs that should be considered, including data acquisition costs and labeling costs. Most previous work on cost-sensi...

Citation Context

...kes, and that information can be used on the next iteration of the algorithm. One example of such an interactive clustering system is developed by Cohn et al. (2003). Like active learning approaches (=-=Cohn et al., 1994-=-), where an algorithm requests labels for the items it deems most informative, this system is able to receive more precise guidance towards the desired solution. This can reduce the amount of informat...