Search

This study narratively untangles an African–American student’s experiences in the throes of receiving a scholarship to study computer science and enter a future Science, Technology, Engineering, and Mathematics (STEM) career. Using the “wounded healer” metaphor as an interpretative lens, this work explores challenges the young adult experienced relating to his development, culture, contextualized learning, family interactions, religious beliefs, and self-identity. The student’s stories of teaching-learning, transforming, and healing instantiate the profound impact the grant-supported scholarship had on the youth’s development, life, and career trajectory. Additionally, new connections between narrative and metaphor are forged in ways that strengthen the sense made of teaching-learning, culture, and social interactions in higher education.

Massive textual content has enabled rapid advances in natural language modeling. The use of pre-trained deep neural language models has significantly improved natural language understanding tasks. However, the extent to which these systems can be applied to content generation is unclear. While a few informal studies have claimed that these models can generate `high quality' readable content, there is no prior study on analyzing the generated content from these models based on sampling and fine-tuning hyperparameters. We conduct an in-depth comparison of several language models for open-ended story generation from given prompts. Using a diverse set of automated metrics, we compare the performance of transformer-based generative models - OpenAI's GPT2 (pre-trained and fine-tuned) and Google's pre-trained TransformerXL and XLNet to human-written textual references. Studying inter-metric correlation along with metric ranking reveals interesting insights - the high correlation between the readability scores and word usage in the text. A study of the statistical significance and empirical evaluations between the scores (human and machine-generated) at higher sampling hyperparameter combinations (t = {0.75, 1.0}, k = {100, 150, 250}) reveal that the top pre-trained and fine-tuned models generated samples condition well on the prompt with an increased occurrence of unique and difficult words. The GPT2-medium model fine-tuned on the 1024 Byte-pair Encoding (BPE) tokenized version of the dataset along with pre-trained Transformer-XL models generated samples close to human written content on three metrics: prompt-based overlap, coherence, and variation in sentence length. A study of overall model stability and performance shows that fine-tuned GPT2 language models have the least deviation in metric scores from human performance.

We perform an in-depth, systematic benchmarking study and evaluation of phishing features on diverse and extensive datasets. We propose a new taxonomy of features based on the interpretation and purpose of each feature. Next, we propose a benchmarking framework called `PhishBench,' which enables us to evaluate and compare the existing features for phishing detection systematically and thoroughly under identical experimental conditions, i.e., unified system specification, datasets, classifiers, and evaluation metrics. PhishBench is a first in the field of benchmarking phishing related research and incorporates thorough and systematic evaluation and feature comparison. We use PhishBench to test methods published in the phishing literature on new and diverse datasets to check their robustness and scalability. We study how dataset characteristics, e.g., varying legitimate to phishing ratios and increasing the size of imbalanced datasets, affect classification performance. Our results show that the imbalanced nature of phishing attacks affects the detection systems' performance and researchers should take this into account when proposing a new method. We also found that retraining alone is not enough to defeat new attacks. New features and techniques are required to stop attackers from fooling detection systems.