The computational intelligence research group is trying to completely exploit the intellectual potentials of computers to enrich our lifestyle. We believe the key to maximize the capabilities of computers is "self-learning" rather than teaching everything by hand. In order to realize self-learning, we have been investigating:
- Evolutionary computation
- Statistical machine learning
- Human robot interaction
We have applied algorithms to robotics and web mining, such as image search and recommender systems. New potential members are free to explore other domains as well.
Keywords: evolutionary computation, neural networks, reinforcement learning, statistical machine learning, natural language processing, evolutionary robotics, modular robotics
Automatic Image Annotation
Haven't you had trouble searching your storage filled with image files to find one particular image that you wanted? Since cellphones with cameras and digital cameras become more and more popular, the amount of image files increases drastically. Therefore,the importance of searching the image files that you need is becoming greater, too. One of the ways to realize this / solve this issues is to use a system that annotates images automatically. This system takes into account the words related to the image files and the relationships between the words. In this way, it helps humans to find an image file more quickly and more precisely compared to systems, which only use the words. In our laboratory, in order to generate the caption, which includes the information showing "what" and "how" the contents of the image file are, we use a technique based on generic object recognition and one based on similarity-based image retrieval.
Keywords: generic object recognition, similarity-based image retrieval, machine learning, natural language processing
Achieving computer intelligence that is as smart as human intelligence has been one of the ultimate aims in the field of computer science. One approach towards this aim is to utilize neural networks, which model the human brain nerve system as a mapping from a state to an action. In our laboratory, we especially focus on the technology of neuroevolution, a technology to evolve the structure and the parameters of neural networks by using evolutionary computation. We try to apply neuroevolution in fields such as automatic controlsystems(?) of robots, artificial intelligence of games, etc. to develop an application program that is usable in everyday life. In other words, we try to improve everyday life through neuroevolution.
- Hirotaka Moriguchi (D3)
- Shingo Horiuchi (M2)
- Yuki Inoue (M1)
- Shengbo Xu (M1)
- H. Moriguchi and S. Honiden, CMA-TWEANN: Efficient Optimization of Neural Networks via Self-adaptation and Seamless Augmentation, In Proceedings of Genetic and Evolutionary Computation Conference (GECCO 2012), 2012.
- H. Moriguchi and H. Lipson, Learning Symbolic Forward Models for Robotic Motion Planning and Control, In Proceedings of European Conference of Artificial Life (ECAL 2011), pp. 558-564, MIT Press,2011.
- H. Moriguchi and S. Honiden, Sustaining Behavioral Diversity in NEAT, In Proceedings of Genetic and Evolutionary Computation Conference (GECCO 2010), pp. 611-618, ACM, 2010.