Indianapolis – As a master’s degree in Purdue University in electrical and computer engineering in West Lafayette, Karen d’ouza was impressed by the depth and extent of the research offered by the University’s IT department. So much so that Souza was convinced to return to Purdue to finish his doctorate in artificial intelligence and in automatic learning in the educational space.
Since the fall of 2021, from Souza has been a vital member of the research group led by Snehasis Mukhopadhyay, professor of computer science and information in Indianapolis who has become his doctoral advisor.
“It seemed to me to be a good adjustment to dive into advanced research,” said Souza, who obtained a baccalaureate in computer engineering in Bengaluru, India. Souza was particularly attracted by the Purdue project which, thanks to the use of AI, aimed to stimulate the participation of chemistry students from all walks of life. “Overall, the Purdue University provided me with an excellent research environment to thrive in the doctoral program in computer science,” she adds.
Flourishing. While interviewing the research position with the team led by Mukhopadhyay, the teacher of education in chemistry and STEM Pratibha Varma-Lelson and the computer teacher at the University of Indiana Shiaofen Fang, of Souza learned more deeply the AI project in the group’s AI education. This has also bitten the interest of Souza since his objective was to build a unique educational solution while working in close collaboration with the Department of Chemistry and an existing online class.
The team’s research also adapts to Souza’s interests to develop solutions in the field of education, in particular for multimodal learning and human AI in the loop. Here, humans actively participate in the training, evaluation and operation of ML models, providing precious advice and comments. It has also been struck by the interdisciplinary nature of the research of the program, combining IT (AI), education in chemistry, psychology and even sociology.
“At Purdue University in Indianapolis, I applied to the AI project in education by a deep curiosity on how and why machines can be designed to model human cognition and learning processes,” says Souza. “I explored each of these areas in depth, finally developing an integrated framework for educational AI.”
Respond to the need for AI tools for educators
As data from several text methods, images, audio and video have increased – accelerated by the transition to hybrid and distance learning during the COVID -19 world pandemic – Souza says that a clear demand emerged for AI tools which could provide contextual comments to educators.
“Multimodal AI offers a holistic approach to a complex data representation challenge, while a human approach in the loop guarantees that I have been able to make context corrections by including humans with each iteration,” she says.
Souza’s research focuses on an online undergraduate chemistry course that uses an educational tool called Cyber Peer-Diled Team Learning (CPLTL). In this format, small groups engage in the learning of peer groups, approaching complex subjects without traditional instructor. His work models have an AI to assess various dimensions of peer learning, including critical reasoning, teamwork and problem solving. But there was a hitch. Faced with a lack of sets of data on public education due to confidentiality problems, it explored generative AI solutions using large languages.
“One of the highlights of my research was to compare the analysis of AI from models of large languages with a human evaluation. It was precious because it was never explored before using the generator, “she says.
While focusing on her studies and research aspirations as a boilermaker, Souza jumped at the opportunity to develop the skills necessary outside the class and the laboratory – in case she chooses a collaborative career as an agademic with industry:
- She participated in the semester Safety Purdue Ai Of course, an initiative managed by students, where from Souza has acquired an overview of mechanistic interpretability and its essential role in the development of safety systems before deployment.
- She actively participated in the weekly learning of peers with the Purdue Machine Learning Club, remaining up to date with the latest advances in the field.
- And she joined Purdue’s Society of Women EngineersNetworking with industry and obtaining advice to develop as a leader.
When Mukhopadhyay encouraged Souza to consider teaching, she spent a year as a speaker for a purely cycle of undergraduate – an experience that she found more rewarding than she expected. “I quickly learned that class discussions triggered new ideas for helping me to articulate more effectively complex subjects, even in my research. It was one of the most precious experiences of my time in Purdue, “says Souza.
Press learning activities beyond the class
Beyond the classroom, she obtained internships with Dell, IBM Research and the Pacific Northwest National Laboratory, obtaining precious perspectives of the industry that have enriched her studies and strengthened her research point. “These experiences played a decisive role in my thesis because they allowed me to apply my knowledge to research directly,” she said.
Mukhopadhyay says that Souza’s research, which has focused on the development of essential algorithms to model a multimodal and human machine learning solution from several data channels, will have applications in STEM collaborative education.
“His work is the first of its kind in STEM education, where the integration of traditional automatic learning and LLM (large languages models) using a human approach in a loop has not been explored previously. Research in these fields requires a strong anniversary of theoretical concepts”, explains Mukhopadhyay. “And Karen’s academic competence, perseverance, self-motivation and open-mindedness to discover new ideas have helped to advance the project.”
Souza is also working on the Mukhopadhyay health care project, which uses AI to predict the possibility of sinusitis surgery. She says that the skills drawn from her thesis objective have been transferred transparently to health care, because education and health care are faced with similar risks of confidentiality and security. The two areas also lack data accessible to the public.
This collaborative research effort received an R01 R01 subsidy of $ 3.3 million over five years. The portion of Mukhopadhyay Mukhopadhya represents $ 1.4 million in this total. The predictive models of the AI that he and his team helped to develop could eliminate cases where there is no medical advantage for sinus surgery, relieving high financial charges, as well as significant physical and mental costs for millions of Americans.
“Although these case studies are often perceived as megadroned problems, they can actually be discussed as small data problems,” said Souza. “The project itself is a unique opportunity to develop an advanced AI solution which attacks both multimodal data and incorporates experts in the matter to improve precision and results.”
While continuing his doctorate, Souza also works as a computer scientist in a generative AI at the Idaho National Laboratory, offering him a unique opportunity to serve as a postdoctoral researcher in modeling probabilistic risks. In the laboratory, it takes advantage of its expertise to develop advanced AI applications for nuclear science and technology.
“Dr. Mukhopadhyay was monumental in my success in Purdue,” said Souza. “I attribute my success to his advice and the dynamic research environment he favored. He encouraged me to collaborate with experts in various fields, including computer visualization, multimedia, education and psychology, expanding my perspective and strengthening my research.