Masahiro Nagamatsu

An emotion processing system based on fuzzy inference and subjective observations

The two theories concerning mixed emotions adopted in this research are briefly explained. Also discussed are the image codes linked to the mixed emotional words, and the way to realize a mechanism using fuzzy inference, in which the internal emotion of a simulated person reacts on a sequence of inputted terms and evokes some emotional change. Next, the relevant part of the theory for a subjective observation model to the application of emotion processing system is briefly explained. The construction methodology of the system, which has two functions, is described. One function is that the evoked emotion is transitive, depending on the observation, and the other is that the aggregated emotion by the several subjective observations settles to the objective emotion. Then, some attractive results obtained by the simulated emotion processing system are shown, namely, what kind of emotions are evoked, and how the emotional transition is done by subjective observations when a poem is used for the input signal. Furthermore, we could obtain several emotional clustering diagrams processed by pair vectors such as [optimism, curiosity], which is a very attractive result from a psychological viewpoint.

Brain-Inspired Emergence of Behaviors Based on Values and Curiosity in Mobile Robots

We propose to introduce the desire for existence, specific curiosity, diversive curiosity, and novelty into reinforcement learning as intrinsic rewards for developing truly autonomous mobile robots capable of behaving without being told what to do. A pursuit-evasion game composed of predators and their prey is selected as a testbed. Simulation experiments and experiments using real mobile robots, WITHs, on a robotic field well demonstrate the effectiveness of introducing intrinsic rewards in addition to external rewards in the conventional reinforcement learning.

Symbolized particles store type neuron model and its application

As a simplified analogy of the function that the real neurons and synapses do by chemical particles, a new type of neuron model is proposed including its simulated synaptic function where the symbolized particles are generated, emitted, stored and transmitted toward the neurons of the next layer. The several kinds of chemical particles have been discovered on the recent scientific knowledge. The simulated system of the model was applied to the analysis of voice signals, and obtained useful results such that avert with a single neuron including the unit simulated synapse, the examined voices demonstrated a high degree of recognition. The reaction mechanism in simulated synaptic region would be shown by some strategy.

Accelerating concurrent fault simulation by parallel pattern emptiness checking of fault lists

In this paper, we propose two new techniques to accelerate concurrent fault simulation. In concurrent fault simulation, if detectable faults are increased, then fault lists of primary outputs become empty for almost all input patterns. In our first technique, we introduce simple calculation instead of exact calculation of the fault lists. In these calculations we check whether the fault lists of the signal lines are empty or not. When emptiness checking of all signal lines is done and the fault lists of all primary outputs are empty, then we conclude that no fault can be detected. Parallel processing of input patterns is possible for this emptiness checking. As a result of emptiness checking, if fault lists of some primary outputs prove to be nonempty, then we must calculate the exact fault lists. Our second technique utilizes the information obtained during emptiness checking to decide the set of lines for which the exact fault lists must be calculated. It is confirmed by the experimental results that these techniques accelerate concurrent fault simulation by about 10 times for the ISACS 85 bench mark circuit.