RESEARCH | SOGABE Laboratory, The University of Electro-Communications AI Artificial Intelligence Quantum

Our lab focuses on integrating cutting-edge AI techniques into three key research areas: energy optimization, device design, and quantum computing. The primary objective is to create automatic control systems and design schemes that tackle tasks too complex for human knowledge alone.

 

 

Our two core computing technologies are: Physical Simulation, using ab initio methods for applications from atomic simulations to global weather modeling, and AI Simulations, powered by the ReNom framework, which includes deep learning, topological data analysis, and reinforcement learning for tasks like image analysis and robotics.

 

 

In AI-based smart grid optimization, we have successfully developed an ensemble deep reinforcement learning algorithm. This algorithm trains multiple learning agents to handle a wide range of weather scenarios, simulated through a global weather forecasting package. The trained agents not only learn optimal policies but also provide risk evaluations for decision-making processes.
(Tomah Sogabe, et al., Attention and masking embedded ensemble reinforcement learning for smart energy optimization and risk evaluation under uncertainties. J. Renewable Sustainable Energy 1 July 2022; 14 (4): 045501. https://doi.org/10.1063/5.0097344)

 

 

In device design, we optimized transparent solar cells for efficiency, transmittance, and lifespan. Traditional trial-and-error methods cannot handle the complexity of over 30 parameters. Using AI, time-dependent DFT atomic simulation, and reinforcement learning, we optimized molecular structures for specific absorption profiles, achieving maximum light absorption and visible light transmittance.
(Tomah Sogabe, et al., Ultrafast inverse design of quantum dot optical spectra via a joint TD-DFT learning scheme and deep reinforcement learning. AIP Advances 1 November 2022; 12 (11): 115316. https://doi.org/10.1063/5.0127546)

 

 

We have demonstrated the applicability of AI technology in quantum circuit design, which is considered one of the most challenging aspects of advanced quantum computing. Quantum circuits exhibit intricate features such as high dimensionality and non-linearity, aligning well with AI’s strengths. The integration of AI and quantum computing has proven highly synergistic, facilitating the generation of target quantum states and the creation of robust and reliable ansatz states with minimal molecular energy.
(Tomah Sogabe, et al., Quantum circuit architectures via quantum observable Markov decision process planning. Journal of Physics Communications vol.6(7) 075006 (2022) DOI 10.1088/2399-6528/ac7d39)