
Breakthroughs & Insights
Some of the latest advancements and transformative research findings from Columbia Engineering faculty and collaborators.
A new way to detect AI-generated content
Generative artificial intelligence systems are already being used across the world for nefarious purposes that range from financial fraud to political propaganda. As these systems have grown more sophisticated, it’s become increasingly difficult for humans to detect whether content was created by humans or by a machine.
As such, researchers are racing to develop computational methods for detecting AI-generated content.
At Columbia Engineering, computer science professor Junfeng Yang and YM Associate Professor of Computer Science Carl Vondrick are leading the charge by developing a framework, called Raidar, that exploits the tendency of AI models to prefer content created by other AI models.
Raidar detects text written by a large language model by measuring the number of edits another LLM makes when revising a given piece of text, comparing the original text with the rewritten text. Many edits mean the text was likely written by humans, while fewer modifications indicate the text was likely machine generated.
Raidar’s remarkable accuracy—it surpassed previous methods of detection by up to 29%—inspired the researchers to turn their attention to video. Their video-detection system, called DIVID, extends a method called diffusion reconstruction error to compare frames of a given video against those of an AI reconstruction. As with Raidar, DIVID measures how many changes an AI system makes as it optimizes a video.
The system achieved a groundbreaking detection accuracy of up to 93.7% for videos in the researchers’ benchmark data set of diffusion-generated videos from Stable Vision Diffusion, Sora, Pika, and Gen-2.
An electric bandage enables faster healing
Chronic wounds, such as sores resulting from diabetes, are extremely serious — in some cases, leading to amputation and even death. Unfortunately, the treatments currently available are expensive and often ineffective.
To help remedy that, a team of researchers developed a cheap, more effective treatment that uses electricity to promote healing.
“These lightweight bandages healed wounds faster than the control, at a similar rate as bulkier and more expensive wound treatment,” says Sam Sia, a professor of biomedical engineering at Columbia Engineering and a member of the team behind the technology.
In animal tests, wounds dressed with the new bandage healed about 30% faster than those dressed with a typical bandage.
Designed to be simple enough for patients to use at home, the bandages consist of a bio-safe battery and flexible electrodes that conform to the shape of the wound. When activated with a drop of water, the battery creates an electric field that lasts for several hours. “We found that the electrical stimulation from the device sped up the rate of wound closure, promoted new blood vessel formation, and reduced inflammation, all of which point to overall improved wound healing,” says Maggie Jakus, co-first author of the study and Sia’s PhD student.
A new tool for designing high-performance batteries

Developing new battery technology is essential in efforts to reduce society’s reliance on fossil fuels and meet our carbon emission targets. Researchers have spent 40 years developing lithium metal batteries, which stand to hold significantly more energy than now-common lithium-ion batteries.
The difference between the two designs — lithium-ion batteries have an anode made of graphite while lithium metal batteries use lithium for that component — could lead to electric vehicles that can travel on a charge as far as gaspowered vehicles can travel on a tank of gas.
Unfortunately, the lithium anode is so chemically reactive that designing safe, long-lasting lithium metal batteries has proved an elusive goal.
Lauren Marbella, associate professor of chemical engineering at Columbia Engineering, is pioneering new methods for peering inside batteries so researchers and developers can better understand what’s happening on the anode and design components that resist damage.
The technique her lab uses, called nuclear magnetic resonance spectroscopy, allows researchers to both measure how quickly lithium ions are moving and determine the exact chemical and physical structure of the problem-causing defects.
“Once we know what structural changes are occurring, then we can intentionally engineer these in and design lithium metal batteries that meet the performance metrics required for commercialization,” Marbella says.
Python teeth inspire innovation that doubles the strength of rotator cuff repairs

More than 17 million Americans injure a rotator cuff every year. These injuries aren’t just debilitating — they’re also hard to treat with existing techniques. While a younger patient with a minor tear can expect surgery to succeed about 80% of the time, the success rate for older patients with significant injuries is just 6%.
To improve these outcomes, a multidisciplinary team led by Stavros Thomopoulos, a professor of orthopedics at Columbia University’s Vagelos College of Physicians and Surgeons and a professor of biomedical engineering at Columbia Engineering, looked to the animal kingdom for inspiration.
We designed it specifically so that surgeons won’t need to abandon their current approach. They can simply add the device and increase the strength of their repair.
Iden Kurtaliaj
Pythons are known for using their bodies to constrict prey before eating it whole, but that’s not the full story. The snake uses its backward-facing teeth to grasp prey so they can’t wriggle away.
As it turns out, this design is perfectly suited for grasping soft tissue— such as a torn rotator cuff— without inflicting too much damage. Thomopoulos and student Iden Kurtaliaj collaborated with orthopedic surgeon William Levine and mechanical engineering professor Guy Genin to apply the technique to rotator cuff repair.
Their device attaches to a healing rotator cuff using an array of small, backward-facing teeth. By protecting the tissue from the sutures installed during surgery, the device prevents the most common route of surgical failure.
“We designed it specifically so that surgeons won’t need to abandon their current approach,” Kurtaliaj says. “They can simply add the device and increase the strength of their repair.”
A groundbreaking microwave chip

Electronic devices like GPS units, autonomous vehicles, and wireless communication systems need components that provide a source of stable microwave signals. These high-frequency metronomes serve as clocks and carry information. Random variation in that signal, called noise, limits the device’s overall performance.
Within the past decade, researchers have developed a new way to generate microwave signals with very little noise. However, the technique — called optical frequency division — requires a hardware setup the size of a tabletop, making it impractical for use in smaller devices.
Researchers at Columbia Engineering have developed a device that performs this technique on a chip that could balance on the point of a sharp pencil.
“We have realized a device that is able to perform optical frequency division entirely on a chip in an area as small as one to two millimeters using only a single laser,” says Alexander Gaeta, the David M. Rickey Professor of Applied Physics and Materials Science and a professor of electrical engineering at Columbia Engineering.
“We demonstrate for the first time the process of optical frequency division without the need for electronics, greatly simplifying the device design.”
Developed in collaboration with Michal Lipson, the Eugene Higgins Professor of Electrical Engineering and a professor of applied physics at Columbia Engineering, the chip is designed and fabricated to produce a completely optical device that uses just one laser to generate a 16-GHz microwave signal with the lowest frequency noise that has ever been achieved in an integrated chip platform. The device uses two photonically coupled micro-resonators to produce the extremely pure signal.
“Eventually, this type of all-optical frequency division will lead to new designs of future telecommunication devices,” Gaeta says. “It could also improve the precision of microwave radars used for autonomous vehicles.”
A new method for tracking atoms

The materials used to make devices like computer chips, batteries, and chemicals place a ceiling on how efficient, reliable, and robust those technologies can be.
For example, important fertilizers are created by reacting gases on the surface of a catalyst. Over time, the catalyst degrades as the atoms that comprise it shift and cause deformations, slowing the process and requiring more energy. Understanding these atomic changes is key to optimizing fertilizer production and securing the global food supply.
Until recently, researchers developing advanced materials faced a roadblock: The atoms they were working with moved faster than they could track using existing technology.
Just as a camera with a slow shutter speed can’t capture an extremely quick event, the X-ray pulses that researchers used to study materials were too slow to capture images of atomic behavior on the scale of a picosecond. (In one second, light can travel around the earth seven and a half times. But in one picosecond, light can travel only one-third of a millimeter.)
A team of researchers at Columbia Engineering and the Department of Energy’s Brookhaven National Laboratory finally cracked the problem.
For the first time, the researchers employed a technique for analyzing materials — ultrafast atomic pair distribution function — using an X-ray free-electron laser. The breakthrough required careful coordination across multiple teams of physicists and engineers.
“I was simply blown away by how well it worked,” says Simon Billinge, professor of materials science and of applied physics and applied mathematics at Columbia Engineering.