91ÇàÇà²Ý students to present research at Symposium of Student Scholars

Students: Lauren Lee, Skylar Colman 
Research Mentor: Weiwei Chen  

Description: Our project aims to address the disparities in health insurance coverage between cardiac and glucose monitoring devices. With heart disease being the leading cause of death and nearly 12% of the US population affected by diabetes, monitoring medical devices are crucial in managing these conditions. Access to these devices is essential for effective care, yet insurance coverage for them is often unequal. By focusing on cardiac and glucose monitoring devices, which are critical in healthcare, we seek to highlight the need for more comprehensive insurance coverage, as one disease can often lead to another. Through our research, we hope to identify which medical devices require better coverage and which ones currently receive sufficient support, ultimately advocating for more equitable healthcare policies. (Colman and Lee) 
  
Inspiration: My inspiration for conducting this research comes from my family history of cardiovascular diseases and diabetes. After losing my grandfather in Jamaica from Type II Diabetes back in 2018, I noticed how important monitoring medical devices were. He did not have a wearable continuous glucose monitor (CGM) due to costs and limited access to healthcare, so he would have to manually check his blood sugar. Having a CGM would’ve been a lot more manageable for him instead of frequent painful finger sticks. Also, from my experience working as a patient care technician on the cardiovascular unit, conducting research on devices relevant to patients I’ve encountered frequently is very fulfilling. (Colman) 
  
Outcomes: By identifying gaps in insurance coverage, we aim to help advocate for better reimbursement policies and ensure that patients in need of these medical devices can access them. Our research helps raise awareness for situations where healthcare providers may assume that their patient isn’t complying with treatment; however, lack of insurance coverage hinders access to these devices. (Colman) 
  
Lessons: Throughout this process, I’ve learned that research is far from easy. It requires significant time and effort to find reliable resources, determine a clear direction, and execute the study effectively. (Lee) 

Students: Audrey Beilharz, Alondra Aguilar-Rodriguez, Kayme Chavez, Lota Iwuagwu, Linda Guevara 
Research Mentor: Chloe West-Jacobs 
 
Description: This research aims to investigate how attention driven by the value of stimuli affects the ability to detect changes in visual scenes, specifically by comparing neutral versus value-driven stimuli using the change blindness flicker paradigm. In this paradigm, images alternate rapidly with brief interruptions, testing participants' ability to notice changes. By examining responses to neutral and value-driven stimuli, the study aims to determine whether emotionally significant objects capture more attention and reduce change blindness, compared to neutral stimuli that may not draw the same level of attention. Ultimately, this research addresses how personal or emotional significance influences visual perception and susceptibility to change blindness. 
 
Inspiration: We were inspired when taking our perception class and learning about the phenomenon change blindness. This phenomenon is so interesting because it reveals how flawed human perception of the world is, as we might miss a big change if our attention isn't focused in the right spot. Change blindness exposes the limits in human attention, and this is just a fascinating concept, as most humans think they take in everything around them at once when in reality we only process small parts of the scene at a time. We were also inspired by the value-driven attentional control, which is a concept in psychology where attentional control is influenced by stimuli people find valuable. That is why we came up with the value-driven stimulus. We were interested to see if our peers detected a change with something considered valuable that was changed between the photographs. 
 
Outcomes: There are a few practical outcomes for our research, including marketing, forensics, and enhancing safety. In marketing, our research can help identify what catches people's eye so marketers know what to make stands out in advertisements. In the forensic field, this research can further the understanding of eye-witness testimonies and can be used to caution against people's assumptions that eyewitnesses remember everything in a scene. Finally, in enhancing safety in aviation, medicine, or military operators can miss critical visual changes. Change blindness studies inform how to design alert systems (e.g., blinking lights, audio cues) that effectively capture attention when something important changes.  
 
Lessons: One unexpected lesson that I took from this project is how strongly emotional or personally charged stimuli can influence our attention. While I expected that emotionally charged objects would capture more attention, I was surprised by how much this attention seemed to facilitate more effective change detection, even in situations where the changes were subtle. It highlighted how our brain prioritizes emotionally significant stimuli, which can both enhance and possibly distort our perception of the surrounding environment, depending on how much attention these stimuli divert from other neutral elements in the scene. 

Students: Aster Cheung, Jessica Mambo, Samson Samuel, Linh Luong 
Research Mentor: Paul Lee  
  
Description: This study seeks to address the limitations of current methods for assessing mitochondrial capacity in aging skeletal muscle due to their high cost and the need for multiple occlusions, which can be uncomfortable for older adults. (Luong) 
  
Inspiration: I've been interested in the medical field ever since I've been a kid, and I have full intention on going to medical school. Disease, disability, and research regarding health have always been a passion of mine, so doing this research for muscular aging was a great opportunity to involve myself in this field of research. (Samuel) 
 
Outcomes: The practical outcome of this research suggests that using DCS and NIRS simultaneously to access mitochondrial capacity is feasible without requiring arterial occlusion. This method can be implemented for home-based monitoring of older adults. (Cheung) 
  
Challenges: One unexpected lesson I learned is that conducting research with human subjects is not easy. Before participating in the research, I thought we only needed to perform a literature review, develop a methodology, collect data, analyze results, and form a conclusion. However, working with human subjects makes the process much more complex. As researchers, we need to obtain certifications from the CITI Program about human ethics (offered by 91ÇàÇà²Ý) and apply for IRB approval before conducting studies on human subjects. Developing a conclusion requires significant time and effort. Additionally, collecting data using optical sensors comes with challenges. External factors, such as muscle fiber motion artifacts, can increase the difficulty of analyzing results. The First-Year Scholars Program and Dr. Lee have given me valuable insight into conducting proper research and provided me with hands-on experience in data collection. (Luong)  

Students: Rohit Malik, Manisha Kumari 
Research Mentor: Md Abdullah Al Hafiz Khan 
 
Description: This research seeks to address several key challenges in food image and recipe retrieval by improving the semantic alignment between visual and textual modalities. The major issues it tackles include: 

• Bridging the semantic gap between images and text
  • Traditional models struggle to link food images with their corresponding recipes due to the complexity and diversity of food appearances and textual descriptions.   
• Handling variability in food presentation
  • Dishes can appear visually different due to variations in ingredients, cooking styles, lighting conditions, and angles.
• Improving feature representation for images and text
  • Earlier methods relied on handcrafted features or shallow models, which failed to capture deep semantic relationships.
• Addressing data challenges (class imbalance & noisy data)
  • Many food datasets suffer from uneven class distributions, where some cuisines or dishes have fewer samples, affecting retrieval performance. 
• Enhancing multi-modal learning with joint embeddings
  • Prior models struggled to learn shared representations that effectively encode both images and recipes into a common embedding space.  

By addressing these issues, the proposed model aims to significantly improve food image-to-recipe retrieval accuracy, making it more robust, scalable, and semantically meaningful. 
 
Inspiration: The inspiration for conducting this research stems from several key motivations: 

• Bridging the gap between visual and textual food data 
• Enhancing user experience in food search & discovery
• Addressing challenges in noisy and imbalanced datasets
• Revolutionizing nutrition & health tracking
• Advancing AI in culinary and e-commerce applications
• Learning from recent breakthroughs in AI & transformers.  

Outcomes: The practical outcomes of this research extend to various real-world applications, improving both food image retrieval and recipe understanding. Some key outcomes include:  

• Improved food image-to-recipe retrieval
• Enhanced recipe recommendation systems
• Automated food logging and nutrition analysis
• Cross-modal learning for food industry & e-commerce
• Advancements in AI for multi-modal learning
• Improved handling of noisy and imbalanced datasets
• Smart restaurant and culinary AI integration  

Lessons: What is one unexpected lesson you learned from this project?  
One unexpected lesson from this project was how complex and ambiguous food data can be. Unlike traditional classification tasks where objects have clear, distinct categories (e.g., cats vs. dogs), food images and recipes often lack a strict one-to-one mapping. 
   
This insight forced us to rethink traditional machine learning approaches. Ultimately, this project highlighted the nuances of food AI, teaching us that understanding food is not just about vision or text but its deep, contextual interplay.