Rochester Institute of Technology

This study explores the intersection of gender identity and sexual orientation in shaping the experiences of threatening behaviors and sexual abuse on university campuses. It highlights how cisgender women and LGBTQ+ individuals report significantly higher rates of verbal threats, sexual harassment, stalking, and sexual violence compared to their male or heterosexual peers. The findings emphasize the compounded vulnerabilities faced by LGBTQ+ individuals, especially LGBTQ+ women, who experience heightened risks of severe violence due to both gender-based and sexual orientation-based marginalization. These results support the importance of an intersectional approach in understanding victimization and advocate for tailored sexual violence prevention strategies in university settings. The study calls for comprehensive policies that address the unique needs of marginalized groups, with a focus on creating inclusive and supportive environments for all students. Keywords: college students, gender identity, sexual orientation, sexual violence, cisgender women, LGBTQ+ individuals, verbal threats, sexual harassment, sexual abuse, stalking, victimization, intersectional approach

Photonic chips are powerful tools for measuring and analyzing light, but most compact spectrometers face a fundamental trade-off: improving resolution usually requires larger devices or sacrifices in signal quality. Here, we introduce a chip-scale architecture that overcomes this limitation by extracting phase information corresponding to the hidden timing of light waves using only simple intensity measurements. Our method generalizes earlier minimum phase designs to allow sparse and non-sequential optical delays, enabling accurate phase reconstruction on a single circuit. By engineering these delays, the device can determine the wavelength of an unknown laser with sub-picometer precision, all while using just one input and one output. This single-stream design reduces loss, improves robustness, and avoids the complexity of traditional spectrometers. The result is a compact, scalable platform that enables high-accuracy wavelength metrology and opens possibilities for on-chip sensing and computational spectroscopy.

Bank telemarketing campaigns often experience low subscription rates due to customer heterogeneity and severe class imbalance, which pose challenges for reliable predictive modeling. This study investigates a data-driven approach that integrates synthetic minority oversampling and cost-sensitive learning to improve the prediction of telemarketing outcomes. Experiments are conducted using the Portuguese Bank Marketing dataset, comprising 41,188 instances with a positive response rate of 11.3%. Eight machine learning models are evaluated under a unified preprocessing pipeline and five-fold stratified cross-validation, including Logistic Regression, Decision Tree, Random Forest, and Ensemble methods. The results show that Ensemble models, particularly CatBoost, XGBoost, and LightGBM, achieve improved performance compared with traditional baselines, with notable gains in minority-class recall and overall discrimination ability. The best-performing model attains an F1-score of 0.540, a recall of 0.812 for the positive class, and a ROC–AUC of 0.908. To enhance interpretability, SHAP-based analysis is applied to quantify feature contributions, identifying campaign duration, previous contact outcomes, and selected macroeconomic indicators as key predictors. These findings indicate that combining resampling strategies with cost-sensitive optimization provides a robust and transparent approach for learning from imbalanced telemarketing data, thereby supporting reproducible and data-driven financial decision-making by explicitly addressing difficulty in minority-class identification under imbalance and class imbalance under cross-entropy training in imbalanced banking data.

Titanium matrix composites (TMCs) offer low density, high specific strength and good thermal stability, yet their deployment in certification-critical aerospace, automotive, energy and biomedical components remain constrained by incomplete understanding of strength–toughness–durability trade-offs. This work provides a mechanistic, meta-analytical review of reinforced TMCs, spanning ceramic, intermetallic, metallic, nanocarbon and laminate or hybrid architectures, with emphasis on how reinforcement chemistry, interface design and hierarchical microstructure jointly control static and time-dependent performance. A quantitative property landscape is compiled for major reinforcement families, enabling side-by-side comparison of tensile strength, stiffness, ductility, fracture toughness, fatigue strength, creep resistance and environmental stability, and revealing how ceramic-rich systems occupy a strong-but-brittle corner of design space, whereas laminate, hybrid and nanocarbon-reinforced systems offer more balanced toughness and durability windows. Building on this analysis, the review introduces a tri-coupled framework that treats interface engineering, hierarchical reinforcement architecture and multiscale validation as interdependent design levers for next-generation TMCs. Within this framework, advanced characterization, digital-twin concepts and machine-learning-assisted models are integrated with mechanical testing to generate more predictive process–structure–property maps and to identify microstructural signatures of long-term damage tolerance. The review further outlines key translation barriers—including microstructural reproducibility in advanced processing routes, the lack of standardized multi-environment fatigue and creep protocols, data limitations for mechanistic and data-driven modeling, and sustainability constraints—and proposes an agenda in which targeted durability testing, hetero-deformation-based strengthening models and data-rich digital workflows are systematically combined to accelerate qualification of titanium-based composite systems for reliable, certifiable service in demanding structural applications.

Gradient Boosting Decision Trees (GBDTs) are widely used in industry and academia for their high accuracy and efficiency, particularly on structured data. However, the subject of watermarking GBDT models remains underexplored, especially compared to neural networks. In this work, we present the first robust watermarking framework tailored to GBDT models, utilizing in-place fine-tuning to embed imperceptible and resilient watermarks. We propose four embedding strategies, each designed to minimize impact on model accuracy while ensuring watermark robustness. Through experiments across diverse datasets, we demonstrate that our methods achieve high watermark embedding rates, low accuracy degradation, and strong resistance to post-deployment fine-tuning.

Multimodal large language models (MLLMs), which integrate language and visual cues for problem-solving, are crucial for advancing artificial general intelligence (AGI). However, current benchmarks for measuring the intelligence of MLLMs suffer from limited scale, narrow coverage, and unstructured knowledge, offering only static and undifferentiated evaluations. To bridge this gap, we introduce MDK12-Bench, a large-scale multidisciplinary benchmark built from real-world K–12 exams spanning six disciplines with 141K instances and 6,225 knowledge points organized in a six-layer taxonomy. Covering five question formats with difficulty and year annotations, it enables comprehensive evaluation to capture the extent to which MLLMs perform over four dimensions: 1) difficulty levels, 2) temporal (cross-year) shifts, 3) contextual shifts, and 4) knowledge-driven reasoning. We propose a novel dynamic evaluation framework that introduces unfamiliar visual, textual, and question form shifts to challenge model generalization while improving benchmark objectivity and longevity by mitigating data contamination. We further evaluate knowledge-point reference-augmented generation (KP-RAG) to examine the role of knowledge in reasoning. Key findings reveal limitations in current MLLMs in multiple aspects and provide guidance for enhancing model reasoning, robustness, and AI-assisted education.

Simplified sea level modelling approaches are developed to efficiently explore future sea level rise and associated uncertainties. Sea level emulators (SLEs) are mostly calibrated against the responses of process-based complex models, they can be run on multi-century timescales and feed into regionalisation efforts, integrated assessment and coastal risk modelling. Here, we introduce the Sea Level Emulator Intercomparison Project (SLEIP) to systematically assess available sea level emulators and identify future research needs to maximise the utility of this modelling approach. SLEIP covers 13 datasets from the participating models BRICK (with DOECLIM and SNEASY climate forcing), FACTS (7 individual emulator workflows), FRISIA, MAGICC, ProFSea and SURFER. All of the participating SLEs produce projections out to the year 2300 for the main sea level drivers thermal expansion, glacier mass loss, Greenland and Antarctic ice sheet mass loss, and land water storage. Participating SLEs differ in whether and how they account for low-confidence, high-impact processes of poorly known likelihood, such as marine ice-cliff instability (MICI). The SLE components with the largest response range are the Greenland and Antarctic ice sheet, with the Antarctic ice sheet becoming the most uncertain sea level driver in 2300. With identical MAGICC climate forcing input, 2300 median global mean sea level rise estimates range from 0.46 m to 1.71 m (outer 17th-83rd percentile range: 0.32-3.20 m) under very low emissions (SSP1-1.9), 0.67 m to 2.01 m (0.47-3.56 m) under low emissions (SSP1-2.6), 1.64 m to 4.07 m (1.15-10.53 m) under moderate emissions (SSP2-4.5), 2.35 m to 9.33 m (1.68-14.39 m) under high emissions (SSP3-7.0), and 2.44 m to 11.16 m (1.74-15.79 m) under very high emissions (SSP5-8.5), all relative to 1995-2014. SLEIP also allows investigating the sea level response under overshoot. Under the overshoot scenario SSP5-3.4-OS (peak GMT: 2.3 °C, 2100 GMT: 1.9 °C), median projections range from 0.45 m to 0.86 m (0.36-1.31 m) in 2100 and 0.80 m to 2.30 m (0.56-9.82 m) in 2300.

When the resources for experimentation are limited, experimenters usually turn to the class of 2-level fractional factorial designs. Resolution III fractional factorial designs are the smallest available designs, but they alias main effects and 2-factor interactions. The class of Resolution IV designs avoids this and provides clear estimates of the main effects, assuming that 3-factor and higher-order interactions are not active. Meanwhile, some two-factor interactions remain aliased with each other. Resolution V designs have no aliasing of main effects and two-factor interactions, assuming that all higher-order interactions are inactive. However, they are often too large for situations with six or more factors of interest. For example, with six factors, the only design capable of estimating all main effects and 2-factor interactions has 32 runs. Consequently, resource restrictions often require experimenters to use smaller designs of lower resolution, typically Resolution IV. The aliasing of effects often requires additional follow-up experimentation to de-alias all active effects. However, there are situations in which follow-up experiments are impossible to perform due to the unavailability of certain test resources. An alternative to using a 16-run Resolution IV design and a follow-up experiment is to use a design with more than 16 runs as the initial experiment. We investigated a strategy for initially augmenting a class of 16-run Resolution IV designs with either 4 or 8 runs. We use a simulation study to show that this augmentation strategy improves the ability to estimate active factors when standard analysis methods are employed. The analysis methods used in this study are Stepwise, LASSO, and Dantzig.