Description

Project Overview

For human-in-the-loop (HITL) systems, it is important to understand and quantify users’ perceptions to make the next predictions in line with the user’s intention. HITL systems that employ visual recognition often require inferring similarity between a percieved object and a mental targed. This is often difficult to determine when it comes to complex stimuli like faces. While HITL systems traditionally rely on explicit user input, implicit EEG responses can support the decision-making process effortlessly.

Project Goals

In this project, we will be exploring brain signals (more accurately, ERP components) in similariy prediction for visual stimuli.

• Experiment Design: Participants will be shown a target image followed by a number of non-target images. For every image pair (target & target), they will need to determine similarity on a scale. During the experiment, they will provide keyboard input, eye tracking data and EEG data.
• Data Analysis: Preprocess the received data and analyze ERP components.
• Model Training: Train a model on the collected data that predicts different levels of similarity between images.

You will

• Generate visual stimuli for the experiment (via Midjourney or similar) OR explore previously used datasets of images from similar studies
• Integrate the stimuli in a given system
• Conduct a user study with EEG and Eye Tracking
• Collect and analyze the collected data, focusing on finding correlations between assessed similarity and ERP components
• Summarize findings in a thesis and present them to an audience
• (Optional) Co-write a research paper based on the results

You need

• Strong communication skills in English
• Knowledge of machine learning and data analysis (e.g., Python)

References

• De la Torre Ortiz, The P3 indexes the distance between perceived and target image https://helda.helsinki.fi/bitstream/10138/572292/1/main_1.pdf
• De la Torre Ortiz, Cross-Subject EEG Feedback for Implicit Image Generation https://helda.helsinki.fi/server/api/core/bitstreams/d2a54f39-7d87-4311-9b08-0d71578a36ca/content