The Perceptual Intelligence Lab (PIL)
From Human Hypotheses to Machine Representations — Bridging Human Perception and Machine Intelligence with Mathematical Formulation
Mission and Vision
Our mission is to understand how machines learn and what machine intelligence reveals about intelligence itself. Modern artificial intelligence has achieved remarkable success, yet many fundamental questions remain unanswered. Why do neural networks learn meaningful representations? How do architectural assumptions influence optimization? When biological knowledge, physical laws, or structural priors are encoded into a model, does the model truly learn through these mechanisms—or does it arrive at entirely different solutions?
At PIL, we study machine learning not merely as an engineering tool for prediction, but as a scientific discipline for understanding intelligence. This perspective connects machine learning with biology, physics, mathematics, and engineering, where models are valued not only for their predictive performance, but also for their ability to reveal and explain the underlying mechanisms of complex systems.
Our research bridges human perception, mathematical modeling, machine learning, and machine perception, transforming scientific hypotheses into computational representations that machines can optimize while investigating what they actually learn.
Research Philosophy
Human-designed architectures encode assumptions about how a system should learn.
Optimization transforms these assumptions into learned representations.
Our role as researchers is to understand the relationship between designed structures and learned representations—and to uncover what emerges in between.
Research Environment
PIL emphasizes both scientific rigor and technical excellence.
Students are encouraged to develop deep foundations in
- Machine Learning
- Mathematics
- Optimization
- Probability
- Scientific Computing
- Software Engineering
Our laboratory maintains modern computational infrastructure for reproducible AI research, including Linux-based GPU servers and scalable computing platforms.
Students gain practical experience with technologies such as
- Python
- JAX
- PyTorch
- C++
- Linux
- GPU Computing
- Distributed Training
- Cloud-native Infrastructure
- Open-source Scientific Software
We believe that excellent researchers should understand both the algorithms they develop and the computational systems that enable them.
Laboratory Culture
PIL values curiosity, intellectual independence, and depth of understanding.
We aim to build an environment where students are free to explore ideas, test hypotheses, and learn through experimentation. We believe that meaningful innovation often emerges from exploration, and that allowing space for trial and error is essential for developing original scientific insight.
Rather than optimizing for immediate correctness, we encourage a culture of safe experimentation, where mistakes are treated as part of the research process rather than failures. This environment is designed to help students develop both confidence and creativity in tackling open-ended scientific problems. We also emphasize long-term thinking over short-term performance, and depth. Our goal is to cultivate researchers who can ask bold questions, explore freely, and build deep understanding through iterative discovery.