32584 research outputs found
Sort by
Moving Target Defense with Quantized Morphence: Defense Quantification Against Common Adversaries in Image and Time Series Problems
No full textIn recent years, the vulnerability of deep learning models to adversarial attacks has emerged as a serious threat, particularly in domains where reliability and robustness are critical. This project builds upon the Morphence framework, a Moving Target Defense (MTD) strategy designed to counter adversarial threats by maintaining a dynamic pool of models and introducing randomness at inference time. While Morphence was originally developed for image classification tasks, this work not only reproduces the original architecture using MNIST and CIFAR-10 datasets but also extends the core principles to an entirely new domain: time series forecasting. The project proposes a unified defense pipeline that begins with clean model training, followed by the generation of perturbed student models, adversarial training using FGSM, BIM, and PGD attacks, and post-training quantization to support memory-efficient deployment. The approach is validated through extensive experiments on both image and time series data — including electricity load forecasting using a Transformer-based model. Across both domains, the Morphence-inspired defense shows significant improvements in adversarial robustness while maintaining high accuracy and low memory footprint, even after 8-bit and 16-bit quantization. By adapting MTD strategies to regression-based forecasting, this project demonstrates the broader applicability of Morphence and provides a concrete step toward building lightweight, attack-resilient machine learning models suitable for real-world deployment
Reinforcement Learning-based End-to-End Monitoring Path Selection in Multi-domain Optical Networks
No full textThis paper presents a novel approach for optimizing network monitoring in optical communication systems using Reinforcement Learning (RL). Assuming a multi-domain architecture with limited domain visibility, we simulate multiple optical connections using an optical communications simulation software, GNPy, obtaining key network metrics to model the system. We developed two RL agents: the first agent selects near-optimal monitoring paths based on network states, and the second agent dynamically adapts its selected paths in response to state changes, such as fiber failures or issues with ROADMs. This adaptive approach allows for continuous improvement of network monitoring, ensuring resilience and efficient fault detection. Experimental results demonstrate the RL agents’ ability to select a set of monitoring paths performing within single-digit percentages of the optimal set
riboMoE: An Application of Mixture of Experts on Artificial Riboswitch Classification
No full textUrban water sewage is a potential health concern due to its possibility to spread contagious RNA viruses such as Coxsackievirus B3. However, detection of viral particles remains challenging because of low viral concentrations in wastewater and high mutation rates of the RNA virus. To address this, this study proposes a novel viral detection method using synthetic riboswitches that bind to the target virus and trigger a reporter gene, amplifying the detection signals. To support the design of effective riboswitches, we present a machine learning model for classifying riboswitch performance, integrating RNA sequence data with secondary structural features. This model used a sparsely gated Mixture of Expert (MoE) layer to route mixed input to specialized experts, achieving excellent generalization performance. Future work includes improving the biological relevance and interpretability of the MoE model
Human vs. LLM: Interpreting Emotion from Neutral Faces in Contextually Charged Backgrounds
No full textIn this study, we investigate the intersection of cultural context, visual perception, and implicit bias in interpreting neutral human expressions. Specifically, we explore how emotionally charged backgrounds can shape viewers interpretations of neutral facial expressions. Using an experimental setup, participants are shown neutral human portraits paired with varying background types ranging from emotionally neutral or pleasant to evocative scenes. Participants are tasked with selecting which background best matches the emotional state of the person depicted in the portraits. Eye-tracking data is collected to analyze visual attention patterns and cognitive processing. In parallel, we also evaluate how large language models (LLMs) respond to the same stimuli, identifying whether similar biases appears in AI driven interpretations. Our preliminary results reveal that both human participants and LLMs tend to project emotional meaning onto neutral faces based on background context, indicating some susceptibility to contextual and cultural bias. These findings contribute to a deeper understanding of human machine perception and also pave way for developing fairer, culturally sensitive AI systems
Link Failure Localization in Hierarchical, Multidomain Optical Networks
No full textOptical networks transport data encoded on light signals over optical fiber cables. Individual optical networks are managed by domain administrators, such as service providers, vendors, or regional bodies. Multidomain optical networking explores the possibility of enabling seamless data transmission across domain boundaries. In this project, we explore how we can perform network monitoring in a hierarchical multidomain optical network while preserving domain security and autonomy. This is done by allocating dedicated monitoring trails across various broker abstractions of our network topologies. We propose three heuristic functions that expedite trail selection. Of the three heuristics, our least monitored algorithm performs the best, achieving results within 4% of the theoretical optimum on average for small topologies. We also propose an LSTM model to detect soft link failures based on OSNR data simulated using GNPy. The model achieves an accuracy of 91% and an F1 score of 63%
What is the Impact of an Education Program Designed to Remove the Healthcare Professional as a Barrier to Care for the Transgender and Gender Diverse Community?
Get PDFHealthcare disparities disproportionately affect transgender and gender-diverse (TGD) individuals, often due to implicit biases among healthcare providers. This study examines the impact of an educational intervention designed to address these biases and improve care for TGD patients. Using the Biased Care Model as a guiding framework, the intervention targeted three critical phases of bias: pre-visit assumptions, in-visit interactions, and post-visit decisions. Conducted in the intensive care unit of a mid-sized hospital in Northern California, the study employed a pretest-posttest quasi-experimental design involving multidisciplinary healthcare workers. Participants completed a modified Jefferson Scale of Empathy survey to assess changes in cognitive and affective empathy levels before and after the training. The results of the educational intervention revealed significant improvements in healthcare providers\u27 awareness of implicit biases and their ability to demonstrate empathy toward transgender and TGD patients. Participants reported increased confidence in addressing the unique needs of TGD individuals, with marked improvements in both cognitive and affective empathy as measured by the Jefferson Scale of Empathy. These findings suggest that targeted bias-reduction training can positively influence provider attitudes and behaviors, fostering a more inclusive and equitable healthcare environment. However, challenges such as institutional barriers and skepticism about the authenticity of allyship underscore the need for ongoing efforts to sustain these gains and ensure meaningful, lasting change in the care of TGD patients
Reassessing the Deontological Undercurrents in Mo Zi’s Ethical Thought
Get PDFThis article reexamines Mo Zi’s ethical philosophy, challenging the dominant interpretation of Mohism as a purely consequentialist or proto-utilitarian system. While Mo Zi’s emphasis on social utility—such as material welfare, population growth, and aversion to warfare—aligns with outcome-oriented reasoning, I argue that his invocation of the will of Heaven introduces a deontological dimension. Heaven’s will serves not merely as a heuristic for maximizing benefits but as an absolute moral standard, demanding adherence to duties like universal love and non-aggression, irrespective of immediate practical gains. Through a close reading of the Mo-Zi corpus, this study reveals a hybrid ethical framework that synthesizes consequentialist and deontological principles. Mo Zi’s moral imperatives are justified both by their practical outcomes and their alignment with Heaven’s commands, creating a system where duty and utility are mutually reinforcing. The article also explores the tension between intent and consequence in Mohist ethics, arguing that while actions are judged by their results, intent is essential for ensuring moral agency and sustainable outcomes. By critiquing Eurocentric interpretations that reduce Mo Zi to a proto-utilitarian, this reassessment highlights the complexity of his thought, situating it within its unique historical context. Ultimately, the article offers a nuanced understanding of Mo Zi’s ethics, demonstrating its relevance to contemporary debates on moral theory and cross-cultural philosophy
Book Review on Epistemology of Modality and Philosophical Methodology (edited by Anand Jayprakash Vaidya and Duško Prelević)
Get PDFEarly Care and Education Systems in the United States of America: Voices of the Past, Present, and Future
Get PDFThis mixed-methods study centers the voices, knowledge, and leadership of Black and Indigenous women working in Early Care and Education (ECE) while tracing how racism, sexism, and classism have shaped the field from colonization to the present day. Twenty in-depth interviews with sixteen Black and Indigenous educators across the United States are paired with secondary analyses of national workforce data to create a multilayered portrait of their professional realities. Grounded in Black feminist thought and Indigenous desire-based inquiry, the analysis surfaces six inter-locking themes: (a) enduring structural barriers that depress wages and limit advancement, (b) the day-to-day resilience that sustains educators in the face of those barriers, (c) culturally responsive pedagogies rooted in community, (d) blocked pathways to professional growth and leadership, (e) the emotional and physical toll of under-resourced work, and (f) participant visions for a liberated, justice-oriented ECE system. Collectively, these findings demonstrate that Black and Indigenous women are— and have long been—central architects of high-quality early learning, yet their contributions are undervalued by policy and practice. The study offers evidence-based recommendations for dismantling oppressive structures, reimagining compensation and career ladders, and building inclusive environments that honor the expertise of the workforce on which young children and families most depend