What is it about?
We worked on a challenge called MuSe-Personalization in 2023, where the aim was to detect human stress levels by analyzing different types of data, like voice and body movements. We made our AI model smarter by teaching it new ways to understand this data. For example, we taught it to recognize certain body poses and to listen to voice patterns in a special way. We also fine-tuned our model by adjusting various settings to get the best results. Our efforts paid off, as our model did a great job, especially in detecting emotional states, and we ranked second in the challenge.
Featured Image
Photo by Elisa Ventur on Unsplash
Why is it important?
Our work stands out because: (1) We enhanced a popular AI model, making it more efficient and avoiding common technical problems. (2) We introduced new ways for the model to understand data, like recognizing specific body movements and using advanced audio analysis. (3) We did not just set our model up and hope for the best. We spent a lot of time adjusting its settings (hyperparameters) and even used a special method to understand which settings were most important. (4) Our results were impressive. We significantly improved upon the baseline results, especially in detecting emotional states, which is a testament to our novel techniques and optimizations.
Perspectives
Combining multiple modalities is challenging, yet often yields better performance than relying on a single one.
Ho-min Park
Ghent University Global Campus
Read the Original
This page is a summary of: MuSe-Personalization 2023: Feature Engineering, Hyperparameter Optimization, and Transformer-Encoder Re-discovery, October 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3606039.3613104.
You can read the full text:
Contributors
The following have contributed to this page
