Attention Structural Change During CLIP Fine-Tuning: A Comparative Study of Full Fine-Tuning, LoRA, and Regularization
A structural analysis of how CLIP attention changes under full fine-tuning, LoRA, and regularization.
A portfolio of research, engineering, and community work
ARTIFICIAL MINDS,
HUMAN VALUES,
AND
SOCIAL IMPACT
Our Mission
This site investigates how artificial intelligence can be designed, governed, and taught in ways that protect human dignity, strengthen public trust, and expand social opportunity.
Inspired by the interdisciplinary spirit of technology, philosophy, and civic engagement, the project asks a simple but demanding question: How do we build intelligent systems without losing sight of the people they are meant to serve?
The answer lives in research, essays, real-world experience, and community-facing work. In other words: fewer black boxes, more human responsibility.
Research
Real-Time On-Device Diffusion: Practical Acceleration via Fused Low-Bit Kernels
A systems paper on accelerating diffusion inference with fused low-bit kernels and cache-update fusion.
Text2Sign: A Practical Single-GPU Diffusion Architecture for Text-to-Sign Language Video Generation
A single-GPU diffusion system for generating short sign-language videos from text prompts.
Our Concerns
When Bias Becomes Infrastructure
A closer look at how AI systems can formalize unequal treatment, with the health-care risk algorithm study as a concrete warning.
•
2 min read
Read more
Generative AI and the Reorganization of Work
The labor issue is not only whether jobs disappear, but how AI changes bargaining power, task control, and whose work becomes more precarious.
•
2 min read
Read more
The Material Cost of AI
AI is often described as pure software, but its growth depends on data centres, power systems, cooling, and public infrastructure.
•
2 min read
Read more
Experience
NYLF Engineering Scholar
A residential engineering leadership program at Georgia Tech focused on design thinking, robotics, communication, and collaborative problem solving.
•
1 min read
Read more
Blog
Gradient Descent
Gradient Descent Gradient Descent is an iterative algorithm used to optimize objective functions, widely applied in machine learning and deep learning to train models by minimizing …
•
3 min read
Read more
Machine Learning Basics
Machine Learning Basics This article is based on Chapter 5 of Machine Learning Yearning and introduces some basic concepts and methods of machine learning. Supervised Learning This …
•
2 min read
Read more
Deep Feedforward Neural Networks
Deep Feedforward Neural Networks This article is a study note for Chapter 6 of the book Deep Learning. Introduction Deep Feedforward Networks, also known as Multilayer Perceptrons …
•
1 min read
Read more
NFLS AI Club
Meeting #3: From Sketch Recognition to AIGC — Industrial AI in Practice
The third NFLS AI Club meeting featured an invited guest lecture on how computer vision, multimodal understanding, and large language models are deployed in real industrial …
•
2 min read
Read more
Meeting #2: Deep Learning Workshop — From Math to Code
The second meeting was a 35-minute technical workshop tracing a single idea from visual intuition through rigorous mathematics to working PyTorch code: how neural networks solve …
•
4 min read
Read more
Meeting #1: Ice Breaking — Welcome to the NFLS AI Club
The first formal meeting brought new and returning members together for an ice-breaking session designed to surface motivations, share curiosity, and establish the club as a real …
•
2 min read
Read more
NFLS AI Club at the School Club Fair
The NFLS AI Club officially launched at the annual school club fair, opening recruitment and introducing AI to students across the school.
•
1 min read
Read more
🚀Xia Ruize
Student Researcher in AI & Society
Nanjing Foreign Language School
Artificial Minds, Human Values
About the Founder
Artificial Minds, Human Values
I’m Ruize Xia, a student researcher interested in the meeting point between artificial intelligence, ethics, and public life.
My work begins with a conviction: technical progress is not automatically human progress. Powerful systems can help people learn, communicate, and solve hard problems, but only when they are built with care for fairness, accessibility, accountability, and human dignity.
That belief shapes the work collected on this site:
- Research on value-aligned evaluation, accessible AI, and decision support.
- Writing that connects machine learning to education, civic trust, and moral responsibility.
- Service and public-facing experiments that turn abstract principles into workshops, tools, and community-facing practice.
The goal is not to romanticize technology or reject it. The goal is to ask better questions, design better systems, and keep human values legible even when the machines grow more capable.
Focus Areas
AI Ethics and Governance
Human-Centered Machine Learning
Accessibility and Inclusive Design
Civic Technology
Education and AI Literacy
Interdisciplinary Research