Research Subjects of Takashi Kanamaru
In this page, I briefly summarize my recent researches.
Smartphones / Single board computers / Image processing / Reinforcement learning for biped robots
robot For these themes, please see the following pages.
Dynamical Behaviors of the Brain
Recently, in the physiological experiments of the brain, dynamical behaviors of neurons, such as "oscillation and synchronization", "correlation of pulses", and "chaos" are reported by many researchers. Such dynamics are typically observed in the visual cortex and the hippocampus, and they might be related to the bindings of the visual information and the control of the synaptic plasticity. These experimental observations suggest that the brain might have richer and more complex dynamics than the conventional artificial neural network. I try to understand these dynamical behaviors using the theories of "nonlinear dynamics" and "statistical physics".

Usually, "oscillation and synchronization", "correlation of pulses", and "chaos" are considered to be the different behaviors, and they are treated separately. However, I regard them as three sections of the same high-dimensional dynamics of the brain, and I think that they should be treated together.

Recently, I am studying pulse neural networks with excitatory and inhibitory connections, and they show the above three behaviors. Thus, they might be effective to analyze the behavior of the brain from the pulse level.

ft Moreover, properties of neurons consisting of the brain, and structures of networks are not uniform. To analyze such systems, the probabilistic representation is crucial to understand their dynamics.

JeJi For example, let us consider a system consisting of 200 neurons as shown in the right figure. Because noise is injected to the system, each neuron fires randomly (shown with red points). Although it is difficult to analyze such stochastic systems directly, we can obtain the Fokker-Planck equation that governs the time-development of the probability density of the system in the limit of large number of neurons.

By integrating the Fokker-Planck equation numerically, we can obtain a smooth trajectory of a solution as shown in the left figure. We can perform precise analyses because the Fokker-Planck equation is a deterministic (partial) differential equation.

I am concerned with understanding the brain using the probabilistic methods.

Chaos and nonlinear dynamics
As for this subject, please see the following website.
Learning the control of the biped robot
bipedal1 Let us consider movements of human, such as biped walking. Although there are several conditions for walking, such as "walking on the flat ground", "walking on the graveled ground", "ascending/descending the stairs", "going up/down the slope", and so on, we can walk stably regardless of the conditions.

This is because we have been learning how to walk adapting to the environments for a long period. We have been learning how to walk since we were about 1 year old.

Can't robots move adapting to the environment like human? Can't robots learn how to move and how to process the information like human?

The theories of neural networks and machine learning treat such problems. In our laboratory, we use a method called reinforcement learning, and we try to construct a model which learn how to walk with trial and error like human.

For further information about the reinforcement learning, please see:
Image Recognition based on Wavelet Transformation
wavelet Unlike the Fourier transformation, the wavelet transformation is suitable for spatially and temporally localized signals such as images.

You can apply the wavelet transform to some images using the following Java simulators. After applying the wavelet transformation in the pre-processing, we can use several techniques for the image recognition such as machine learnings and neural networks.

Back to Takashi Kanamaru's page