Using machine learning to measure the ‘tensile strength’ of memories

24 Feb 2021

Lorin Sweeney. Image: Insight

PhD student Lorin Sweeney is using computer science to explore how we can create more meaningful and memorable content.

With a BSc in computer applications and software engineering, you might not expect Lorin Sweeney to be examining the intricacies of how our memory works. It’s a topic he first encountered during his undergraduate studies and he is now delving into it as a PhD student in computer science at Dublin City University (DCU).

Conducting his research at Science Foundation Ireland’s Insight research centre for data analytics, Sweeney aims to design machine learning models that can predict how memorable a piece of content will be.

“I believe that the study of memory and memorability is uniquely important due to our lack of metacognitive insight into what we will remember or forget,” he said. “It would be far stranger for the subject not to be studied given that our memories make us who we are.”

Unfortunately, the pandemic has restricted Sweeney’s ability to access the neurophysiological data required for this research. He told about these challenges as well as his concerns around public engagement with science.

‘I like to think of memories as the tethering threads that tie us to the world, and memorability as the measure of their tensile strength’

What inspired you to become a researcher?

I can’t say that I experienced a single defining moment of inspiration. I was a relentlessly curious child, always asking questions and never quite fully satisfied with the incompleteness of the answers I received. My innate inclination towards inquisitiveness and my learning-induced excitement makes me feel like my path towards research was quite natural and somewhat inevitable.

I believe that knowledge acquisition and the act of sharing knowledge with others is a uniquely beautiful activity that has the power to preserve the spark in our childlike hearts. It’s certainly a source of great joy in my own life.

Can you tell us about the research you’re currently working on?

My research is focused on media memorability – understanding what makes media in its many modalities memorable, and designing machine learning models to predict how memorable given media will be.

I was first introduced to the topic during the final year of my undergraduate degree. One of my PhD supervisors, Prof Alan Smeaton, told me about memorability when we were discussing dissertation projects. I was instantly enamoured and did my project on the subject.

After finishing my project, Alan informed me that I had the opportunity to continue studying media memorability if I wanted to pursue a PhD. Naturally, I jumped at the opportunity.

Going into my PhD I was keen to explore the neural correlates of visual memorability, so I spent quite a lot of time learning about EEGs and preparing an experiment with my other supervisor, Dr Graham Healy. Unfortunately, due to the advent of Covid-19, we had to postpone the experiment and have yet to gather data.

In the meantime, I shifted my focus to examining the influence of audio on video recognition memorability and ended up finding evidence to suggest that audio primarily plays a contextualising role, with the potential to act as a signal or trigger that aids recognition depending on the extent of its high-level features such as arousal, causal uncertainty, imageability and familiarity. I am currently working on formulating a more robust measure of memorability which can account for the moment of recognition, and better reflect real-world remembering.

In your opinion, why is your research important?

I like to think of memories as the tethering threads that tie us to the world, and memorability as the measure of their tensile strength. The threads of memory are spun from fibres of many modalities, obscuring the contribution of a single fibre to a thread’s overall tensile strength. Unfurling these fibres is the key to understanding the nature of their interaction, and how we can ultimately create more meaningful and memorable media content.

What commercial applications do you foresee for your research?

The study of computational memorability is relatively new, so I think that the current commercial applications are quite narrow in scope and limited in practical utility. However, with richer, more precise measures of memorability, I believe that the possibilities are endless, from educational content curation to potential memory therapy for individuals with cognitive disorders.

I believe that the application most within reach is memorability as a feedback mechanism in a classroom context, where automatically generated memorability scores will allow educators to refine their materials.

What are some of the biggest challenges you face as a researcher in your field?

One of the biggest challenges at the moment is gathering data. Covid-19 has made it impossible to gather neurophysiological data that requires in-person participants.

Are there any common misconceptions about this area of research?

The term memorability itself can cause quite a bit of confusion since its meaning changes depending on the context of its usage.

The two most prominent measures of memorability are recognition memorability and recall memorability.

Recognition memorability is where, amidst content presentation, participants indicate which items they feel they have previously perceived. Recall memorability is where participants recount as much information as they can concerning previously presented content. Ideally we would want to find a way to unify these two measures to prevent confusion, or at least take into account the kernel of what they capture.

What are some of the areas of research you’d like to see tackled in the years ahead?

Rather than a specific area of research, I would like to see changes to the research environment and ecosystem. One of the most disturbing trends of our time is the dismissal of scientific facts, empirical evidence and the very concept of objective truth. Coupled with the inexorable rise of social media, this tech-fuelled trend has led to the rampant propagation of disinformation and left our society vulnerable to malignant entities sowing specious narratives and exploiting collective dialectic tensions for political gain.

I believe that this trend is in part due to the failure of research communities to communicate with the public, the politicisation of science and the lack of visible integration of research with infrastructure that benefits people’s daily lives. A greater effort should be made to implement cutting-edge research in ways that are visibly positive, and to create and disseminate narratives founded in facts, promoting research initiatives.

Want stories like this and more direct to your inbox? Sign up for Tech Trends, Silicon Republic’s weekly digest of need-to-know tech news.