Neural information retrieval: at the end of the early years

A recent “third wave” of neural network (NN) approaches now delivers state-of-the-art performance in many machine learning tasks, spanning speech recognition, computer vision, and natural language processing. Because these modern NNs often comprise multiple interconnected layers, work in this area is often referred to as deep learning. Recent years have witnessed an explosive growth of research into NN-based approaches to information retrieval (IR). A significant body of work has now been created. In this paper, we survey the current landscape of Neural IR research, paying special attention to the use of learned distributed representations of textual units. We highlight the successes of neural IR thus far, catalog obstacles to its wider adoption, and suggest potentially promising directions for future research.     

How do elite ski jumpers handle the dynamic conditions in imitation jumps?

We examined the effect of boundary conditions in imitation ski jumping on movement dynamics and coordination. We compared imitation ski jumps with – and without – the possibility to generate shear propulsion forces. Six elite ski jumpers performed imitation jumps by jumping from a fixed surface and from a rolling platform. The ground reaction force vector, kinematics of body segments, and EMG of eight lower limb muscles were recorded. Net joint dynamics were calculated using inverse dynamics. The two imitation jumps differed considerably from each other with regard to the dynamics (moments, forces), whereas the kinematics were very similar. Knee power was higher and hip power was lower on the rolling platform than on the fixed surface. Mean EMG levels were very similar for both conditions, but differences in the development of muscle activity were indicated for seven of eight muscles. These differences are reflected in a subtle difference of the alignment of ground reaction force with centre of mass: the ground reaction force runs continuously close to but behind the centre of mass on the rolling platform and fluctuates around it on the fixed surface. This likely reflects a different strategy for controlling angular momentum.     

Varieties of work related learning

Based on 30 interviews with and observations of professionals, 5 prototypical metaphors of learning are described and related to existing theories about learning. Two of the metaphors were previously described by Sfard: the acquisition and participation metaphors. As well, three new metaphors were found: the discovery metaphor, the apperception metaphor and the exercising metaphor. In an initial study, a newly devised questionnaire was tested with 713 elementary teachers, 92 principals and 33 support people. The five learning metaphors could be distinguished and proved to be reliable, although improvements were necessary. The correlations between the scales were as expected. Meaningful relations were found with the variables of the educational system, age, experience and educational level. In a second study, a revised questionnaire was tested with 556 professionals from 12 different organisations. The five learning metaphors proved to be more reliable than in the first study. The correlations between the scales were in line with expected patterns of correlations: implicit, explicit and social preferences correlated more with each other than with the other preferences. The expected differences between organisations, professions and groups were apparent and could be interpreted meaningfully. The discussion describes the practical use of the metaphors and suggestions for further research.     

Anatomy Teaching,a “Model” Answer? Evaluating “Geoff”, a Painted Anatomical Horse,as a Tool for Enhancing Topographical Anatomy Learning

Development of new methods for anatomy teaching is increasingly important as we look to modernize and supplement traditional teaching methods. In this study, a life-sized equine model, “Geoff,” was painted with surface and deep anatomical structures with the aim of improving students’ ability to convert theoretical knowledge into improved topographical anatomy knowledge on the live horse. Third and fourth year veterinary medicine students (n = 45) were randomly allocated into experimental (used “Geoff”) and control (used textbook) groups. The efficacy of the model was evaluated through a structured oral exam using a live horse. Questionnaires gathered information on student confidence and enjoyment of the task. There was no significant difference in the performance of experimental and control groups either immediately (44±20% vs. 40±21%; P = 0.504) or 9 weeks after the learning intervention (55±17% vs. 55±20%; P = 0.980). There were however specific questions on which the experimental group performed better than controls, and for which gender effects were apparent. The students using “Geoff” showed a transient gain in confidence following the session (Likert scale 2.7 to 3.6) however the initial increase was no longer present at the second test. There was a significant influence of gender on confidence with greater confidence gains in females in the Experimental group. The students found the model to be extremely useful and both groups found the sessions enjoyable. The model will be of benefit as a complementary learning tool for students.