There are plenty of strategies to train your dog, but is there a particularly effective method to get your pet pal to pay attention to you? A team of scientists believes the most successful technique likely involves combining two tried-and-true signals--and they gathered data from canines strapped with eye-tracking headgear to back up their theory. Dog owners frequently try communicating with their pets by looking or pointing directly at an object, but a team at the University of Veterinary Medicine Vienna recently wondered if either method (or a combination of the two) worked best. Led by comparative cognition postdoctoral candidate Christoph Völter, researchers introduced various communication scenarios to dogs to learn the answer. To evaluate the best human-to-dog strategy, a researcher first sat on their knees with a bowl on either side of them, only one of which contained a concealed treat.

Pet dogs know when you intend to give them a treat, even if you drop it where they can't get to it Dogs can understand when humans mean well, even if they don't get what they want from us. Prior to this work, the ability to distinguish between a human being unwilling or unable to perform a task had only been found in non-human primates. The close social bond between humans and canines is well established, but researchers have a limited understanding of if and how dogs comprehend human intent. To see if pet dogs can distinguish between intentional and accidental actions by strangers, Christoph Völter at the University of Veterinary Medicine Vienna in Austria and his colleagues ran tests with humans offering dogs food while the animals' body movements were tracked using eight cameras. Each dog and human were separated by a transparent plastic panel with holes that a slice of sausage could be passed through.

When 3D animated balls on a computer screen defy certain laws of physics, dogs act in a way that suggests they feel like their eyes are deceiving them. Pet dogs stare for longer and their pupils widen if virtual balls start rolling on their own rather than being set in motion by a collision with another ball. This suggests that the animals are surprised that the balls didn't move the way they had expected them to, says Christoph Völter at the University of Veterinary Medicine, Vienna. "This is the starting point for learning," says Völter. "You have expectations about the environment – regularities in your environment that are connected to physics – and then something happens that doesn't fit. And now you pay attention. And now you try to see what's going on."

The goal of this approach is to improve classication accuracies produced by learning algorithms by improving the quality of the training data. Our approach uses a set of learning algorithms to create classiers that serve as noise lters for the training data. We evaluate single algorithm, majority vote and consensus lters on ve datasets that are prone to labeling errors. Our experiments illustrate that ltering signicantly improves classication accuracy for noise levels up to 30%. An analytical and empirical evaluation of the precision of our approach shows that consensus lters are conservative at throwing away good data at the expense of retaining bad data and that majority lters are better at detecting bad data at the expense of throwing away good data. This suggests that for situations in which there is a paucity of data, consensus lters are preferable, whereas majority vote lters are preferable for situations with an abundance of data. 1. Introducti The maximum accuracy achievable depends on the quality of the data and on the appropriateness of the chosen learning algorithm for the data. The work described here focuses on improving the quality of training data by identifying and eliminating mislabeled instances prior to applying the chosen learning algorithm, thereby increasing classication accuracy. Labeling error can occur for several reasons including subjectivity, data-entry error, or inadequacy of the information used to label each object. Subjectivity may arise when observations need to be ranked in some way such as disease severity or when the information used to label an object is dierent from the information to which the learning algorithm will have access. For example, when labeling pixels in image data, the analyst typically uses visual input rather than the numeric values of the feature vector corresponding to the observation. Domains in which experts disagree are natural places for subjective labeling errors (Smyth, 1996). A third cause of labeling error arises when the information used to label each observation is inadequate. For example, in the medical domain it may not be possible to perform the tests necessary to guarantee that a diagnosis is 100% accurate. For domains in which labeling errors occur, an automated method of eliminating or correcting mislabeled observations will improve the predictive accuracy of the classier formed from the training data. In this article we address the problem of identifying training instances that are mislabeled.