Africazine:
A Kookmin University research team has made strides in artificial intelligence by presenting a groundbreaking paper in Morocco.
The team showcased their work at the 29th Annual Conference on Artificial Intelligence and Statistics in Tangier, Morocco, from May 2-5. Their paper, titled “SQuaT: Self-Supervised Knowledge Distillation via Student-Aware Quantized Teacher Features,” introduces a novel method to enhance the performance of quantized models without the need for training labels.
Kookmin University Advances AI Research in Morocco
The Kookmin University team, guided by assistant professor Kim Jang-ho, included master’s and undergraduate students from the department of AI, big data and management. Their research focuses on improving quantized models, which are essential for efficient machine learning applications.
By addressing the limitations of conventional label-free quantization-aware training methods, the team developed a student-aware projection method. This innovation allows for better alignment between high-precision teacher models and low-bit student models, enhancing the stability of quantization-aware training.
Kookmin University: Key figures on AI Research
- AISTATS 2026 took place from May 2-5 in Tangier, Morocco
- The paper is titled “SQuaT: Self-Supervised Knowledge Distillation via Student-Aware Quantized Teacher Features”
Future Directions for AI Model Development
- The team plans to continue research on efficient AI models.
- They have publicly released the SQuaT code for further studies.
The Kookmin University team’s research marks a significant step forward in the field of artificial intelligence, particularly in enhancing model efficiency.
