Leo Schwartz Laidlaw Blog Post

My first summer research project underwent a significant transformation as I navigated various challenges and learned more about what is a niche and complex research field. Initially, I intended on investigating whether electroencephalography (EEG) evidence could reveal which facial recognition processing stage contributes to longer response times in older adults. However, I revised this to examine whether EEG evidence can reveal if older adults demonstrate prioritised neural processing of social threats during perceptual decision-making tasks. To investigate this, I designed an experiment where participants were exposed to threatening and non-threatening faces while their neural activity was recorded using an EEG. By analysing the subsequent neural responses during these tasks, I aim to reveal how social threats are processed differently in older adults (aged 65+) when compared to younger adults (aged 18-35). Recent literature suggests that the neural encoding of threat-related information occurs earlier (around 100-200 ms) than for non-emotional dimensions, such as colour (which occurs around 260 ms), supporting the idea that social threats hold motivational value in guiding behaviour during perceptual decision-making (El Zein et al., 2024). This project is ongoing, and I plan to continue it through my third year, with hopes to publish the results. 

One major challenge arose early on. The Radboud Faces Database I intended to use for stimuli was under construction. This was a problem because the code I planned to use relied on this database. I had two choices: 

1. Wait for the database to become available 

2. Create my own face repository from scratch. 

Initially, I believed it would be best to wait for this problem to resolve itself. However, as time passed and the database remained under construction, I realised that waiting was no longer an option, forcing me to create my own face repository for this project. This was a time-intensive process, but it was unequivocally necessary to move forward with my research.

To create the face repository, I first obtained access to the FACES database, a database of videos of faces morphing from neutral to various emotions (angry, happy, shocked, etc.). Once I downloaded it, I utilised Python to convert each video into seven unique photos at specific time points, creating a gradient of emotional intensity for each video. To further alter the images, I utilised deep learning techniques to track facial landmarks, which allowed me to precisely crop over a thousand faces into the desired shapes conducive to non-noisy EEG data collection. This was an extremely meticulous process and consumed a significant amount of my time, but ensuring that the stimuli met the necessary standards for my experiment was crucial, especially as my supervisor and I plan on publishing this work.

After creating my repository, I had to validate it, as the way in which I created the gradient of emotional intensity is not necessarily ecologically valid. By this, I mean the timestamps in which I extracted photos from videos have no psychological or logical validity and represent no accurate metric of emotional intensity. Thus, to validate my repository, I created a website which allowed users to rate faces on a scale of emotional intensity to receive the perceived emotional intensity of each face. A set of 1776 faces ranging from neutral to full emotional expression were rated by ten people (fellow Laidlaw Scholars and close friends of mine). During the experiment, the faces were split into ten blocks; during each block, faces flashed on the screen for 500ms, and participants then had 2 seconds to make a decision on a scale of 1-7 on how intense they perceived that emotion (each block took only 5-6 minutes).

Table 1

Example of face morph ranging from neutral to angry

After validating the face repository, I faced another obstacle. I had initially planned to use open-source MATLAB code to run my experiment. However, I soon discovered that the code I intended to use was unavailable. Faced with this setback, I had no choice but to code the task myself in MATLAB, a new programming language for me. Initially, the task felt overwhelming and daunting, as I was not only entirely new to coding in MATLAB but also found its functions to be chalked with higher-order mathematics and tricky syntax. What I thought would be a straightforward process became a monumental challenge, as this was not just a matter of writing a few lines of code—I ended up writing over 15,000 lines from scratch. To put this into perspective, coding an entire neuroscience experiment from the ground up is typically not expected, even of Master's students in neuroscience, let alone a second-year undergraduate psychology student. Building everything from the ground up pushed me far beyond what is usually required at this stage, making the entire process both mentally and technically exhausting. Despite the challenges and sharp learning curve, this experience enhanced my programming skills and greatly deepened my understanding of what it takes to study neuroscience in academia, a career path I am considering pursuing.

What I have outlined above is what I achieved in my summer one research project. As mentioned, I plan on continuing this research throughout my third year. While the setbacks I discussed have affected the process and length of this project in inconvenient ways, I have come to appreciate the value of these processes individually, and in their own terms. For example, despite having little experience before this summer, I have learned advanced coding techniques and how to put them to use, something that I had not initially prepared or expected to do while planning. More generally speaking, this has taught me the value of pushing through complications, even if this means landing in uncharted territories. Retrospectively, I feel confident in the decisions I made to propel the project forward, as the first face database I intended to use is still unavailable. 

Table 3

A Screenshot of the Homepage of the Radboud Faces Database Taken on September 15th 

Albeit simple, if there is anything I have learned this summer –  beyond coding – it is that if you want something done right, you must do it yourself. As I look back, I realise I was in an admittedly tricky spot whereby I could begin working on creating my face repository, with the prospect of the Radboud one being available that afternoon. Thus, I was at risk of finding myself in a position where all the work I was doing could have essentially been worthless. If the face database had become available, my problems would have all been solved. Ultimately, I am glad that after a short break from waiting for the Radboud Faces Database to finish construction, I started working on my repository as the database remains down in mid-September.