Persons with amblyopia, especially those with strabismus, are known to exhibit abnormal fixational eye movements. In this paper, we compared six characteristics of fixational eye movements among normal control eyes (n=16), the non-amblyopic fellow eyes and the amblyopic eyes of anisometropic (n=14) and strabismic amblyopes (n=14). These characteristics include the frequency, magnitude of landing errors, amplitude and speed of microsaccades, and the amplitude and speed of slow drifts. Fixational eye movements were recorded using retinal imaging while observers monocularly fixated a 1° cross. Eye position data were recovered using a cross-correlation procedure. We found that in general, the characteristics of fixational eye movements are not significantly different between the fellow eyes of amblyopes and controls, and that the strabismic amblyopic eyes are always different from the other groups. Next, we determined the primary factors that limit fixation stability and visual acuity in amblyopic eyes by examining the relative importance of the different oculomotor characteristics, adding acuity (for fixation stability) or fixation stability (for acuity), and the type of amblyopia, as predictive factors in a multiple linear regression model. We show for the first time that the error magnitude of microsaccades, acuity, amplitude and frequency of microsaccades are primary factors limiting fixation stability; while the error magnitude, fixation stability, amplitude of drifts and amplitude of microsaccades are the primary factors limiting acuity. A mediation analysis showed that the effects of error magnitude and amplitude of microsaccades on acuity could be explained, at least in part, by their effects on fixation stability.

Progress in Machine Learning is often driven by the availability of large datasets, and consistent evaluation metrics for comparing modeling approaches. To this end, we present a repository of conversational datasets consisting of hundreds of millions of examples, and a standardised evaluation procedure for conversational response selection models using 1-of-100 accuracy. The repository contains scripts that allow researchers to reproduce the standard datasets, or to adapt the pre-processing and data filtering steps to their needs. We introduce and evaluate several competitive baselines for conversational response selection, whose implementations are shared in the repository, as well as a neural encoder model that is trained on the entire training set.

Fixation stability in people with macular disease is primarily limited by the amplitude of microsaccades, implying that rehabilitative strategies targeted at reducing the amplitude of microsaccades should improve fixation stability, and may lead to improved visual functions.

No abstract

This paper describes RevUP which deals with automatically generating gap-fill questions. RevUP consists of 3 parts: Sentence Selection, Gap Selection & Multiple Choice Distractor Selection. To select topicallyimportant sentences from texts, we propose a novel sentence ranking method based on topic distributions obtained from topic models. To select gap-phrases from each selected sentence, we collected human annotations, using the Amazon Mechanical Turk, on the relative relevance of candidate gaps. This data is used to train a discriminative classifier to predict the relevance of gaps, achieving an accuracy of 81.0%. Finally, we propose a novel method to choose distractors that are semantically similar to the gap-phrase and have contextual fit to the gap-fill question. By crowdsourcing the evaluation of our method through the Amazon Mechanical Turk, we found that 94% of the distractors selected were good. RevUP fills the semantic gap left open by previous work in this area, and represents a significant step towards automatically generating quality tests for teachers and self-motivated learners.