Democratizing Generative AI Models
We’re trying to democratize AI models by compressing them and fine-tuning them into smaller sizes that can be used by small and medium-sized enterprises, universities, and even personal use.
–Associate Professor of Computer Science, Reza Rawassizadeh
In the world of artificial intelligence (AI), generative AI modeling is a rapidly growing field with many career opportunities. Technologies in this area can be used to solve real-world problems that have a significant impact on people's lives, including large language models (e.g., ChatGPT), self-driving vehicles, drug discovery, medical image diagnoses, and manufacturing, among others.
Unfortunately, this technology is usually developed by large corporations with huge hardware infrastructures, and many individuals, including universities, cannot afford such complex systems to train AI models.
To address this problem, Associate Professor of Computer Science in the Metropolitan College Reza Rawassizadeh received Accelerating Classroom Transformation (ACT) grant funding to purchase a desktop machine with two graphics processing units (GPU) that could handle the large data processing required for AI models.
According to one student enrolled in the course, Sandeep Yerra:
The Generative AI course was at the cutting edge, incorporating state-of-the-art models in its course content. Most of the latest artificial intelligence models rely on computation capacity, hence the GPU helped in both understanding finer details of algorithms and using the algorithms to hypothesize and build research projects. One of these research projects was making deep learning algorithms more resource-efficient, thus reducing the carbon footprint overall in the world.
The Metropolitan College recently started offering several courses that cover AI modeling, including MET CS 767 Advanced Machine Learning and Neural Networks and MET CS 788 Generative AI. Given the recent technological breakthroughs in generative AI and the growing demand for expertise in this area, more courses are needed in related areas like deep learning models and applications.
Rawassizadeh has been able to incorporate this technology into a MET special topics course that has impacted 60 students so far, with more expected to benefit. In this class, students were able to experiment with the algorithms and models they were learning about and acquire hands-on experience. Through this, they built interesting AI models that ran locally on their machines and increased response times for real-world applications while maintaining their privacy.
Rawassizadeh is in the process of developing new courses that will incorporate and be supported by the computer technology he purchased through this grant, and hopes to build a concentration around machine learning and AI in computer science.
“With this technology, our students can implement what they have learned in the class and enable them to move away from pure theory, which is the main obstacle that demotivates our students,” Rawassizadeh said. “By being able to experiment with and deploy these complex models, students will also gain the confidence needed to be successful on the job market as well.”
Conferences, News & Publications
-
Zhang, X., Kedri, K., & Rawassizadeh, R. (2024). Can LLMs substitute SQL? Comparing Resource Utilization of Querying LLMs versus Traditional Relational Databases. arXiv, doi.org/10.48550/arXiv.2404.08727
Project Lead
Reza Rawassizadeh has a PhD in computer science from the University of Vienna, Austria. He serves as an associate professor in the Department of Computer Science, Metropolitan College. His research primarily focuses on ubiquitous technologies, which include wearables, mobile devices, and robots, with a significant…
