The event, hosted by the MSU Museum, featured 10 student researchers from various colleges within MSU, all given the same constraints to present their work.

Abbie Stevens, the CoLab Studio program manager at the MSU Museum and event organizer, said the strict constraints help breed creativity.

"It also just makes for such a dynamic and fast paced evening that's super fun," she said.

Stevens has been the Ignite Talks MSU coordinator for all three events. She said she wants to use them as an opportunity to showcase the wide range of research happening at this university.

"I wanted to impress upon the audience the breadth and depth of the research that students are able to do here," Stevens said.

Preparation was key for an event like this, Stevens said. They held dress rehearsals to make the students more comfortable on stage.

"Presenting in a more entertainment type situation like this is not a common thing for academics to do, so it's a new skill set for the students to learn," she said.

Teaching ChatGPT stereotypes to remove bias

The event began with Michelle Kim, a doctorate student studying computer science, who persuaded the audience to teach stereotypes to language models like ChatGPT.

She explained that language models collect data online, usually large amounts of text, to train themselves. 

"As we all know, people don't always like polite and honest stuff online," Kim said. "Have you ever read a hateful comment from Facebook or Twitter? Well, imagine that comment is given to the language model, and the model learns to speak it. This is bias coming from the training."

These biases can have real-world consequences when people may naively apply language models to solve problems, she said. 

For her research, she wants to remove bias from these language models. However, this begins to create some complications, she said. 

"It is at odds with the way us humans approach bias," Kim said. "How can we remove biases and stereotypes without first identifying and understanding the biases that exist in the training data?"

Additionally, simply removing bias from the models would rid researchers of access to the important information regarding bias — valuable information that teaches them how language models encode the biases while training, she said.

Kim and her colleagues came to the conclusion that they would have to teach stereotypes before removing them. This was done through data sets they compiled and then fed to the model to be scored, Kim explained.

"Our model was not only able to score stereotypes well, but it was able to detect people’s comments and predict toxic content," she said.

Over-utilization of emergency services

Human biology senior Destiny Kanning focused her research on how emergency services are being overwhelmed by unassuming families due to their lack of education surrounding resources.

This over utilization puts a financial burden on the emergency rooms and the clients using them. 

"It's skyrocketing the cost of the emergency department visits, and we're wasting a lot of those resources," Kanning said.

She believes that through educating families about the proper care options available to them, as well as the symptoms a child may be experiencing, overuse can be prevented.

Over the course of two years, she surveyed over 2,000 families who visited the ER, finding that there were over 300 cases of overutilization per month. Of those 300, 80% admitted they didn’t know what they came for wouldn't be classified as an emergency, Kanning said.

"We have the resources in our community," she said. "They just don't know about them."

With the research, Kanning also worked to find what kind of intervention strategies worked best to divert some of those families to other services.

The most effective was a traffic light system, she said. Patients would be given a traffic light and determine whether their issue was red, yellow or green. Red would classify as an emergent issue, that being a broken bone, trouble breathing or high grade fevers. Green would fall under any primary care that was not a direct emergency.

This method saw a 60% success rate in reducing unnecessary visits, Kanning said.

"This is not a topic that a lot of people know about, but it is so real in our community specifically," she said.

Taking a movie of an atom

Stefanie Adams, a doctorate student studying physics, showed audiences it is possible to capture a video of an atom.

She explained that atoms are smaller than the intrinsic limit a light microscope can zoom into, making it unviewable to any microscope. A completely new method must be used to make an atom visible, she said.

To do this, she said her research took inspiration from Braille, a tactile writing system for people with visual impairment.

"We want to use this as an inspiration to find a method to make atoms visible for us without using our eyes directly," Adams said. "In Braille, you scan your fingertip across the dots on the surface to make other letters, put them together into words."

Viewing the dots as atoms on a material surface, Adams said they used an extremely sharp needle, rather than a finger, to scan the atoms in a similar way.

This method is called scanning tunneling microscopy. 

"If our needle is very, very close to a surface, the electrons of the atoms on the surface can randomly decide to jump into our needle, this process is called tunneling," Adams said. "As the needle scans across our material, if the needle is directly over an atom, more electrons will jump into the needle, and so jumping electrons we can measure as a current, and can display this current as a picture."

In order to move from simple images to movies, Adams said they capture repetitions of motions from the atoms such as a rotation or vibration and combine snapshots at different times of motion to create a moving picture. 

The purpose of this research, Adams said, is to improve technological components.

"The materials that we look at are promising candidates to improve applications such as electronic components, energy storage and solar power," she said. "And the goal is to make these devices smaller, faster and more efficient."

Other presentations involved combating sleep deprivation through blue light, the population density of cats, how autism and employment coincide, collective memory in media and more.

The next Ignite Talks event will be held in the spring semester, on Jan. 16, 2024.