Search for: All records

Contemporary views on what students should learn increasingly emphasize that students need to acquire more than a base of knowledge; they need to acquire the skills and abilities to use such knowledge in dynamic and flexible ways. To be most effective, learning environments need assessments that are aligned to these perspectives. Using a principled design framework can help guide assessment development toward such targets. Even when using a framework, however, thorny design challenges may arise. Technology-enhanced assessments offer opportunities to overcome such challenges but are not a solution in and of themselves and can also introduce new challenges. In this paper, we describe three challenges (conflict between multiple dimensions of science proficiency, authentic data, and grade-appropriate graphing tools) that we faced when designing for a specific Next Generation Science Standard, and the theoretical and design principles that guided us as we ideated design solutions. Through these designs we maintained alignment to our multidimensional assessment targets, a critical component of our larger assessment validity argument. 

Synthetic aperture radar (SAR) image classification is a challenging problem due to the complex imaging mechanism as well as the random speckle noise, which affects radar image interpretation. Recently, convolutional neural networks (CNNs) have been shown to outperform previous state-of-the-art techniques in computer vision tasks owing to their ability to learn relevant features from the data. However, CNNs in particular and neural networks, in general, lack uncertainty quantification and can be easily deceived by adversarial attacks. This paper proposes Bayes-SAR Net, a Bayesian CNN that can perform robust SAR image classification while quantifying the uncertainty or confidence of the network in its decision. Bayes-SAR Net propagates the first two moments (mean and covariance) of the approximate posterior distribution of the network parameters given the data and obtains a predictive mean and covariance of the classification output. Experiments, using the benchmark datasets Flevoland and Oberpfaffenhofen, show superior performance and robustness to Gaussian noise and adversarial attacks, as compared to the SAR-Net homologue. Bayes-SAR Net achieves a test accuracy that is around 10% higher in the case of adversarial perturbation (levels > 0.05).