Physics + Information Processing = Intelligent computation
Machine learning is like the magic mirror from Snow White—except it's real. Computers today are capable of predicting our future by constructing logical relationships between the data and deciphering undiscovered rules behind them. Tanaka-Ohzeki Laboratory strives to make this modern magic accessible to everyone.
Deep learning (end-to-end machine learning based on raw features), for example, automates complex computation tasks that would otherwise take humans thousands of years to solve. What really empower this machine learning process are the underlying technologies that consume a large amount of quality data. Collecting a sufficient amount of data for computers to classify them and discover undefined rules, however, is often difficult. Here we promote sparse modeling, a technique that efficiently extracts the most essential parts of hidden data structure, even from a small dataset. With this modern information processing method, we increase the efficiency of data collection through optimized experiments and dataset evaluations.
The foundation of these technologies is a mathematical problem called optimization problem. They are often hidden in our quotidian activities and appear in different forms. As the competition for developing the next generation of computing technology intensifies across the globe, solving these optimization problems that pursue maximized efficiency has become a constant endeavor. At Tanaka-Ohzeki Lab, we incorporate physical principles into these computational algorithms—one of which is called quantum annealing.
Quantum annealing is an optimization method that utilizes quantum effects. It performs much faster than traditional optimization methods in some cases, but there are many other kinds of optimization problems that quantum annealing finds difficult to solve with adequate efficiency. To tackle these strenuous optimization problems, we strive for improving the performance of quantum annealing by integrating the method that was originally developed in the field of information science.
The demand for handling big data is also growing. We can discover implications from data—whether they are images, sounds, or texts—only when its blocks of information comprised of individual elements correlate with one another. As such, establishing the correlations across datasets is a significant challenge when processing large-scale data, such as 4K and 8K high-definition images. Tanaka-Ohzeki Laboratory proposes the use of a graphical model that presents the relationship between the data in the form of graph structures. Based on this graphical modeling concept, we combine the latest technologies like Bayesian statistics and statistical machine learning to construct a data processing framework.
In collaboration with enterprises and institutions across industries, Tanaka-Ohzeki Laboratory is at the forefront of educating both machines and our society through the world's leading research and engineering techniques. We invite you to join our adventurous research and development of this evolving, cutting-edge science that seeks the essence of learning.