32584 research outputs found
Sort by
Enhancing Code Review Automation with Large Language Models using QLoRA Fine-Tuning and RAGs
No full textIn this technological era where Artificial Intelligence and Machine Learning are revolutionizing various domains, Large Language Models (LLMs) are emerging as a very powerful tool. In the software development lifecycle, it is extremely important to have reliable code reviews to ensure security and maintain code quality. This project aims to survey various existing methodologies to aid creation of efficient code review automation agents and also research on ways to make this process more efficient. Parameter Efficient Fine-Tuning (PEFT) methodologies such as LoRA and QLoRA have been explored with an additional focus on a hybrid model that combines adaptive QLoRA with contrastive RAG - to figure out efficient ways to reduce memory required for fine-tuning also making sure inference quality is not compromised. Context has been utilized from general purpose Meta Llama 3.2 3B model. Experiments show that the hybrid approach reduces memory utilization by nearly 50% while achieving low entropy values. The results also show improved performance over baseline systems both in efficiency and inference stability - highlighting the potential of this hybrid technique for real-world code review automation
Comparative Analysis of Embedding Techniques with Clustering Algorithms for Malware Opcodes
No full textMalware detection and classification remain critical challenges in cybersecurity, especially as malicious software becomes increasingly sophisticated and prevalent. While much of the work involving embeddings has traditionally relied on supervised learning approaches, there is significant potential in leveraging unsupervised learning techniques to discern hidden structures in malware data. By employing embedding techniques to convert malware samples into high-dimensional vector representations, we can capture the subtle and complex patterns inherent in malicious code without relying on pre-labeled data. This unsupervised approach helps categorize malware into predefined malware families, greatly aiding in developing cybersecurity solutions. In contrast to traditional supervised models that depend heavily on historical data and predefined labels, unsupervised learning facilitates the discovery of novel and
previously unseen malware variants. This research investigates a range of embed- ding methods, including Word2Vec, FastText, and Doc2Vec, paired with various
clustering techniques such as DBSCAN, K-Means, Gaussian Mixture, Agglomerative, and BIRCH. The objective is to comprehensively analyze the combined impact of these methods on malware detection in an unsupervised setting. By shifting the focus towards unsupervised learning, this paper highlights the potential to capture malware’s dynamic and evolving nature, ultimately contributing to more adaptive and resilient cybersecurity strategies
FRAMEWORK FOR IDENTITY PRIVACY THROUGH GENDER BASED SKELETONIZATION
No full textThe protection of one’s privacy and sensitive information is becoming increasingly difficult in the modern age full of surveillance and data collection. Through the use of image based object detection machine learning models trained for human and facial recognition, people can be identified and tracked to a terrifyingly accurate degree. On the other hand, the information present in surveillance media can play a key role in security and law enforcement. This presents a problem of how to preserve key information without compromising the privacy of any individuals present in the video. In this research project, Computer Vision techniques and a collection of different machine learning models are used to replace the bodies of people present in a video with a gendered skeleton representation. The Ultralytics YOLO CNN object detection model is used to detect the people in the video. The DeepSORT Deep Learning object tracking model is used to accurately track and assign unique ids to each person. An open source HuggingFace CNN gender detection model is used to assign a gender to each detected person. Finally the Google Mediapipe pose landmark detection model is used to generate a skeleton representation of each detected person. Using this technique, personally identifiable features such as their facial features, skin tone, clothes, etc. can be hidden while preserving gender and movement data
How Theravāda Buddhism Enriches the Language of Thought Hypothesis: Beyond Syntax
Get PDFThe Language of Thought Hypothesis (LOTH) posits that mental states operate through a language-like representational system physically implemented in the brain. However, this purely symbolic approach faces a significant limitation: certain mental phenomena—such as intuitions, emotions, and intentions—lack the syntactically compositional structure required by LOTH and therefore cannot be adequately explained within its symbolic framework. To address this theoretical challenge, we turn to Theravāda Buddhism (TB), an early Buddhist tradition that integrates rigorous philosophical inquiry with meditative practice. Remarkably, TB began to prefigure a solution to LOTH’s limitations over 2,300 years ago. Although LOTH and TB differ in implementation—LOTH through combinatorial syntax and semantics, and TB through meditative practice—they both share a foundational syntactic view of the mind. This distinction opens a pathway to enrich and extend LOTH’s explanatory scope. This paper proposes that TB’s recursive, structured experiential models offer a layered and dynamic foundation for symbolic processes, thereby addressing LOTH’s symbol grounding problem—an infinite regress in which no symbol is intrinsically meaningful, without abandoning formal structure. In doing so, it reframes the apparent tension between LOTH and TB as a syntactic enrichment, wherein structured cognitive architectures accommodate both symbolic representations and experience-grounded processes. Through this enrichment, TB’s meditative approach to cognitive processes lacking syntactic compositionality prefigures a resolution to LOTH’s limitations while preserving its core insights into mental representation
Moral and Ethical Nature in Confucian Liang-Zhi and Islamic Fitra: Comparative Perspectives on Innate Morality
Get PDFThis article offers a comparative philosophical analysis of Confucian liang-zhi (innate moral knowledge) and Islamic fitra(primordial nature), focusing on their accounts of innate morality and moral epistemology. Drawing on classical Confucian sources—especially the praxis-oriented interpretation of Wang Yangming—and foundational Islamic texts and commentary, the study demonstrates that both traditions posit an inborn moral faculty grounding ethical universality and responsibility. Wang Yangming’s liang-zhi is conceived as an innate moral principle that, under the imperative of zhi (to actualize or extend), is continually activated through moral practice. In parallel, fitra is understood as a God-given orientation to truth, cultivated and safeguarded through revelation and ethical discipline. Despite differing theological frameworks, both doctrines converge on an optimistic view of human moral potential and reject the notion of original depravity. This comparison not only clarifies the distinctive metaphysical commitments of each tradition, but also illuminates how non-Western philosophies articulate universalist ethics, offering new perspectives for cross-cultural moral discourse and the renewal of contemporary moral philosophy
A Place for Ancient Philosophy in Axial Age Historiography
Get PDFLong-standing debates over historiographical approaches to the Axial Age have distracted the history of philosophy from its own disciplined inquiry into the breadth and depth of ancient thought beyond the Greeks. The philosopher Karl Jaspers offered a vista for seeing commonalities among ancient innovations and discerning continuities along history to modern times. That dual agenda divided Axial historiography with the question of whether axiality reflects creativities of ancient systems or have modern reflections created images of axiality. A singular chronology for humanity encourages a mode of philosophical history open to providential designs, epochal turns, spiritual evolutions, psychological leaps, or cognitive revolutions. History of philosophy and religion, with the advice of theology and social history herein solicited, can reformulate a stricter and sounder historiography more congenial to a broad scope for ancient philosophy. In particular, arrivals of axiality would appear in distinct stages at different times across separate regions as a matter of responding creatively to changing socio-historical conditions. Twelve candidates for Axial phases across Eurasia during the early Iron Age are accordingly proposed, which include oft-mentioned philosophies and religions as well as overlooked systems that were no less Axial
Resilience in Crisis: How BIPOC Educators Navigated the Challenges Associated with COVID-19, Racial Awakening, and the Move to Online Learning
Get PDFAn unprecedented number of BIPOC teachers exited the profession since June 2020 largely due to the challenges posted by the COVID-19 pandemic (Doan et al., 2024). This study aimed to investigate ways these educators coped with the heavy toll of the converging pandemic(s) by asking what, if anything, sustained them? Participants provided individual and group testimonios and shared snapshots of what it was like to teach and exist during this time via the photovoice methodology and reflected on the challenges and sources of support. BIPOC teachers described the following challenges: difficulty connecting with kids, first-hand and vicarious trauma, increased workload, being asked to lead discussions around race with little resources, and having to compartmentalize trauma. They also described sources of resilience such as having someone who understands the struggle, support from administrators, mentorship, boundaries, and family. As we continue to encounter crises due to climate change and ongoing racial tension in our polarized society, BIPOC teachers and students will continue to be disproportionately impacted. Recommendations for creating race-conscious environments include; creating formal and informal structures for affinity groups, addressing racist language directly, implementing systems for feedback, and providing sufficient and timely instructional resources
Degrees of Freedom: How Project Rebound Facilitates Access to Higher Education for Adjudicated Adults
Get PDFMarginalized students are recommended for suspension and expulsion at higher rates than non dis/abled White peers. Exclusionary discipline contributes to the school-to-prison pipeline and recidivism rates in adults. Educational programs have been effective ways to support adjudicated adults as they transition back to their life after serving their sentences. The CSU program Project Rebound is a highly successful rehabilitation tool for adults seeking higher educational opportunities with an effective 0 percent recidivism rate. To learn about this organization a mixed methods study was conducted. The goal of this study was to capture quantitative and qualitative data about the experiences of formerly incarcerated adults and how higher education created an opportunity for rehabilitation. Participants were provided a survey and then were offered the opportunity to participate in an empathy interview. Program administrators were also interviewed to increase understanding of the program. Seven major themes were identified. Discussion about reducing the carceral press at school and increasing opportunities for support were noted. Implications for K12 and higher education leadership were discussed and ended with recommendations for future research in the field
Spartan Daily, November 19, 2025
No full textVolume 165, Issue 37https://scholarworks.sjsu.edu/spartan_daily_2025/1080/thumbnail.jp
The Effects of College Students’ Online Experiences With Racial/Ethnic Discrimination
No full textTime online involves the risk of direct and vicarious online racial/ethnic discrimination. This study examined the day-to-day associations between online racial/ethnic discrimination and positive and negative affect, somatic symptoms, and anxiety. Participants were 208 fourth-year college students (25% men, 72.1% women, 2.9% not reporting gender; 36.1% Asian, 30.3% White, 17.3% Latinx, 7.7% Multiethnic, 8.7% Other; M age = 22 years). The sample resided in the U.S. Data were collected in 2020. Longitudinal data were collected via online surveys using a daily report approach. The prevalence of online discrimination experiences was generally low but impactful. Main effects analyses showed direct online discrimination was related to negative affect, somatic symptoms, and anxiety. Vicarious discrimination was related to negative affect and anxiety. The significance and strength of associations varied by student race/ethnicity (usually differences between White and non-White students). Findings illustrate how online discrimination impacts adjustment