Overview:
The MADUV Challenge 2025 was part of INTERSPEECH 2025, inviting researchers worldwide to develop models that classify mice as either wild-type or ASD models based on high-frequency ultrasound vocalizations. Advanced signal processing techniques were used to handle complex audio data in this globally recognized event.
My Contribution:
- Swin Transformer Implementation: Explored multiple Swin Transformer variants by reviewing recent research papers, implementing the approaches, and conducting thorough experiments.
- Mel-Spectrogram Conversion Pipeline: Developed an end-to-end pipeline to convert raw ultrasound vocalizations into mel-spectrograms for robust feature extraction.
For more details, visit the MADUV Challenge website
Participated in a Statistics Competition organized by the Department of Statistics at Konkuk University in November 2024.
-
Languages:
- C, C++
- JavaScript (Node.js, React)
- Python (Django)
- Java, R
-
Databases:
- MongoDB, MySQL
- Actively solving problems on Baekjoon Online Judge
- Data Structures & Algorithms (Baekjoon)
- React.JS & Node.JS – In-depth study of modern JavaScript frameworks
- Start Date: March 28, 2024
- Resources:
- Deep Learning from Scratch
- CS231n
