Projects | Elizabeth M. Siefert

Memory errors in behavior, brains, and machines

Sat, 10 Dec 2022 00:00:00 +0000

My Role: As part of my PhD work I designed and programmed the experiments, helped secure grant funding, created the stimuli, collected online data online, wrangled/analyzed data in R, and developed neuroimaging analysis pipelines.

Summary: Information differs in how generalizable or specific it is across experiences. For example, as we encounter birds we can learn that there are shared features that link them together (birds can fly and lay eggs). But we can also learn that there are unique features that set them apart (flamingos are pink and have long legs). We tested whether or not people misremember information differently depending on whether or not it is shared or unique.

In short, to do this, we developed online memory games where participants memorized the colors of cartoon satellite categories. We then tested how people’s memory for colors were distorted based on whether or not that satellite part was shared or unique across satellites. We used color because it gives us a tightly-controlled and accessible way to measure how memories are being distorted. We hypothesized that if a part was shared across satellites people might misremember its color as being more similar to the color of the other satellites, a kind of with this blending of colors reflecting a memory error.

In humans, we find that memory biases are strongest for information that is shared across experiences (shared features). In a neural network model trained on the same satellites, we find a strikingly similar effect where the model’s internal hidden layer representations are also distorting shared features the most.

This shows that both humans and neural network models—both of which are remarkable at learning patterns—show memory errors that emerge from learning these patterns. This makes sense! If we learn that certain things in the world are related, in our mind we might represent those things similarly, making it more likely we will mix up the details . I have recently run research to look at these errors but in human brain by having people play the same memory games, but while lying inside an MRI scanner.

I was fortunate to have the opportunity to present this work as a talk at Neuroscience '22 in San Diego, one of the biggest science conferences in the world.

I also presented a neuroimaging (fMRI) version of this research at an international conference, Cognitive Neuroscience Society, in Toronto, Canada.

Impacts of indoor air quality on cognitive performance

Sun, 27 Nov 2022 00:00:00 +0000

My Role: I developed, programmed, and administered 6 cognitive tasks that provide different snapshots of human cognition (e.g. attention, memory, decision-making, etc.). I worked closely with engineers and chemists to design these experiments according to their constraints while also advocating to them for human/user psychology (and how cognition works more broadly).

Summary: Poor air quality is related to many health issues. But usually air quality has been studied outdoors and these harmful effects emerge across years. Because we spend so much time inside (such as at work), we were interested in how everyday exposure to air pollutants indoors negatively impact cognitive performance in real-time.

With a cross-functional team of engineers, chemists, psychologists, and neuroscientists, we conducted systematic literature reviews and also collected in-lab experimental evidence to test how indoor air quality impacts cognition. In lab, we carefully manipulated air quality while participants were performing different cognitive tasks that I designed. We found evidence that both CO2 and lemon essential oils (from a diffuser) have detrimental effects on decision-making. Our work has been cited by engineers, psychologists, educators, and architects. As it shows why ventilation is critical for building design, such as in workplaces, schools, and this is something studied too in high-stakes environment (like submarines and spaceships), as decision-making may otherwise be impaired.

I also presented earlier versions of this work at a XSEED conference (Toronto, Canada), a conference meant for students to present work that tackles complex, multidisciplinary challenges.

When is the best time to learn?

Thu, 27 Oct 2022 00:00:00 +0000

My role: As part of my Master’s thesis, I wrangled, analyzed, and visualized data from 7 experiments. I also wrote up the paper and made the figures.

Summary: It is one thing to remember information. But it is another to take what you learn and transfer it to new scenarios, such as applying a math concept to solve a new problem or improvising in live jazz. Because of known differences in brain activity throughout the day, I was interested in how time of day affects this ability to learn and generalize information. Across 7 experiments, I brought in 331 undergraduates into the lab either in the morning (8am) or evening (8pm) and had them learn new concepts, in which they had to later generalize what they learned to new scenarios. I found that people generalized much better in the morning than in the evening. This is consistent with neuroscientific findings showing that the brains has reduced inhibitory control in the morning. I speculate that this state of low inhibition can help people flexibly transfer what they learn to new scenarios (generalize) because of more spreading activation across memories, facilitating the ability to connect the dots and see new patterns.