Machine Learning models are vulnerable to adversarial attacks that rely on perturbing the input data. This work proposes a novel strategy using Autoencoder Deep Neural Networks to defend a machine learning model against two gradient-based attacks: The Fast Gradient Sign attack and Fast Gradient attack. First we use an autoencoder to denoise the test data, which is trained with both clean and corrupted data. Then, we reduce the dimension of the denoised data using the hidden layer representation of another autoencoder. We perform this experiment for multiple values of the bound of adversarial perturbations, and consider different numbers of reduced dimensions. When the test data is preprocessed using this cascaded pipeline, the tested deep neural network classifier yields a much higher accuracy, thus mitigating the effect of the adversarial perturbation.

Simulating times of high geomagnetic activity are an important part of the continuing efforts to understand space weather and its impacts on humanity. Recent improvements to the Ring-current Atmosphere interactions Model with Self-Consistent magnetic field (RAM-SCB) have been undertaken, with the purpose of expanding the accuracy and robustness of the model during these highly active times. The improvements include a number of changes to the functionality of both the RAM model and the self-consistent magnetic field calculation. In addition, the models have been modernized and rewritten in an effort to make them more user friendly and understandable. The effect of these changes is shown by simulating the day of 17 March 2013, which saw Sym-H drop below −100 nT and included a magnetosphere push-in within geosynchronous orbit. The comparison between the previous model configuration and this new configuration is investigated by calculating a series of metrics for the relation between the observed and modeled Sym-H, as well as the >10-keV electron flux measured by the HOPE and MAGEIS instruments aboard RBSP-B and the simulated electron flux. These metrics show that the improvements to the model have increased the accuracy of the model for the given simulation. Kinetic ring current models have been in use for over 30 years (Harel et al., 1981; Wolf et al., 1982) and continue to be used and developed extensively (e.g.,

No abstract

Audio captioning quality metrics which are typically borrowed from the machine translation and image captioning areas measure the degree of overlap between predicted tokens and gold reference tokens. In this work, we consider a metric measuring semantic similarities between predicted and reference captions instead of measuring exact word overlap. We first evaluate its ability to capture similarities among captions corresponding to the same audio file and compare it to other established metrics. We then propose a fine-tuning method to directly optimize the metric by backpropagating through a sentence embedding extractor and audio captioning network. Such fine-tuning results in an improvement in predicted captions as measured by both traditional metrics and the proposed semantic similarity captioning metric.