¶¶Òőapp

By Dave DeFusco

Imagine a world where your watch knows exactly what you’re eating, when you're eating and even how you’re eating—no calorie counting apps, no embarrassing food photos, no scribbled notes. Just a watch on your wrist that quietly observes and helps improve your health.

This isn’t science fiction. It’s a new reality, thanks to DietWatch, a groundbreaking dietary monitoring system developed by a team of computer scientists and engineers. Their goal is to make it easy—and nearly effortless—for people to track their eating habits in real life using just a regular smartwatch.

Unveiled at the 2025 IEEE/ACM CHASE Conference at the ¶¶Òőapp University Museum—and sponsored in part by the Katz School of Science and Health—DietWatch is a smart, low-effort way to monitor diet using a simple smartwatch. No fancy equipment. No awkward steps. Just a device many people already own.

“This is a step toward making nutrition tracking seamless, private and practical for everyday people,” said Dr. Yucheng Xie, the study’s lead researcher and an assistant professor in the Katz School’s Department of Graduate Computer Science and Engineering. “We wanted to design a system that works in the real world—not just in a lab or clinic.”

Dr. Xie and his collaborators Zhen Hou, a student researcher whom Xie mentored, and Dr. Feng Li, a professor at Purdue University, built DietWatch to detect eating behaviors from the motion of your wrist and the sounds of your chewing, but doing that in everyday life turned out to be a massive challenge.

Doctors and scientists agree: the way we eat—when, what and how—has a huge effect on our health. Bad habits can lead to serious problems like obesity, diabetes and heart disease. The World Health Organization warns that unhealthy eating is one of the top causes of these conditions.

But keeping track of everything you eat isn’t easy. Most people rely on memory, journals or apps to log their meals. These tools often fall short—people forget, guess wrong or just give up. Even high-tech gadgets designed to track eating can feel clunky, expensive or intrusive. No one wants to wear a camera around their neck or chew in front of a microphone at lunch.

Most smart systems for tracking eating work best in quiet, controlled settings, but real life is noisy and unpredictable. Think crowded cafeterias, loud music, car rides, kids yelling. People also eat in many ways—using hands, chopsticks or forks, or multitasking while walking. These unpredictable moments are called “open-world scenarios,” and they make tracking much harder.

“Previous systems couldn’t tell the difference between eating and, say, scratching your head or lifting a cup,” said Hou. “We designed DietWatch to understand eating gestures even if it’s never seen them before.”

To do this, the team taught the system to recognize patterns from real eating behavior, like how the wrist moves when bringing food to the mouth, and to ignore distractions like random motions or background noise. They did this using a special kind of artificial intelligence called contrastive learning, which helps the system “learn by comparison.”

“It's like teaching the watch to play a game of ‘what’s different,’” said Dr. Li. “It learns to spot the little things that make eating gestures unique, even if the person is doing something completely new.”

DietWatch uses four main components to decode your dietary behavior:

• Noise-Canceling for Your Wrist and Ears: It filters out both hand-motion “static,” like moving a napkin or tapping your fingers, and background sounds, like clinking dishes or talking, using smart filtering techniques called Conv-TasNet and Bi-GRU. These are types of neural networks, like tiny digital brains, that separate useful signals from noise.
• Gesture Recognition: It doesn’t just look for one eating motion. Instead, it learns what eating generally looks like and can tell it apart from non-eating movements, even new ones it’s never encountered before.
• Timing Is Everything: It groups eating gestures into time windows to figure out when you're eating. This helps uncover patterns like frequent snacking or late-night eating.
• Guess That Food: It listens to the sound of chewing and analyzes your wrist movements to guess what kind of food you’re eating, like crunchy chips vs. soft bread. It does this by combining data from motion and sound through a method called attention-based fusion, which helps the system weigh different types of input intelligently.

In tests involving 23 different activities and 40 food types in noisy, real-world environments, DietWatch identified eating times with nearly 80% accuracy and food categories with 86% accuracy—remarkably high numbers for such a complicated task.

Knowing when and what you eat might seem like a small thing, but it can unlock powerful insights. For example, people often underestimate how much they snack, or they don’t realize how often they eat out of boredom or stress. These hidden habits can add up.

“By tracking dietary patterns in a non-invasive way, we can help people build awareness—and from there, healthier habits,” said Dr. Xie.

The researchers believe this tool could be especially useful for dietitians, weight-loss programs and even doctors managing chronic conditions like diabetes. Instead of relying on what patients say they ate, professionals could see patterns from real-world data. And because the system uses a smartwatch—a tool already worn by millions—there’s a real chance this could scale quickly and easily.

DietWatch is still in the research stage, but its potential is clear. The team has begun integrating the system with commercial smartwatches, including Apple Watch and Android platforms. They are actively seeking industry partnerships and external funding to support further development, with the goal of turning DietWatch into a practical, everyday health companion.

“Technology shouldn’t get in the way of living your life,” he said. “It should quietly help you live it better.”