Pitoyo Hartono

Learning from imperfect data

For a supervised learning method, the quality of the training data or the training supervisor is very important in generating reliable neural networks. However, for real world problems, it is not always easy to obtain high quality training data sets. In this research, we propose a learning method for a neural network ensemble model that can be trained with an imperfect training data set, which is a data set containing erroneous training samples. With a competitive training mechanism, the ensemble is able to exclude erroneous samples from the training process, thus generating a reliable neural network. Through the experiment, we show that the proposed model is able to tolerate the existence of erroneous training samples in generating a reliable neural network. The ability of the neural network to tolerate the existence of erroneous samples in the training data lessens the costly task of analyzing and arranging the training data, thus increasing the usability of the neural networks for real world problems.

Top-Down Control of Learning in Biological Self-Organizing Maps

This paper discusses biological aspects of self-organising maps (SOMs) which includes a brief review of neurophysiological findings and classical models of neurophysiological SOMs. We then discuss some simulation studies on the role of topographic map representation for training mapping networks and on top-down control of map plasticity.

Fast reinforcement learning for simple physical robots

In the past few years, the field of autonomous robot has been rigorously studied and non-industrial applications of robotics are rapidly emerging. One of the most interesting aspects of this field is the development of the learning ability which enables robots to autonomously adapt to given environments without human guidance. As opposed to the conventional methods of robots’ control, where human logically design the behavior of a robot, the ability to acquire action strategies through some learning processes will not only significantly reduce the production costs of robots but also improves the applicability of robots in wider tasks and environments. However, learning algorithms usually require large calculation cost, which make them unsuitable for robots with limited resources. In this study, we propose a simple two-layered neural network that implements a novel and fast Reinforcement Learning. The proposed learning method requires significantly less calculation resources, hence is applicable to small physical robots running in the real world environments. For this study, we built several simple robots and implemented the proposed learning mechanism to them. In the experiments, to evaluate the efficacy of the proposed learning mechanism, several robots were simultaneously trained to acquire obstacle avoidance strategies in the same environment, thus, forming a dynamic environment where the learning task is substantially harder than in the case of learning in a static environment and promising result was obtained.

Embedded Neural Network for Swarm Learning of Physical Robots

In this study we ran real time learning of multiple physical autonomous robots situated in a real dynamic environment. Each robot has an onboard micro controller where a simple neural network is embedded. The neural network was built with the consideration of the power and calculation resources limitation which is a general characteristic of simple robots. In the experiments, several autonomous robots were placed in one environment, where each of them was given a specific task which was expressed as the evaluation function for the robots neural network. The learning processes of the robots were started simultaneously from their randomized initial conditions. The presence of several robots consequently formed a dynamic environment, in which an action of one robot affected the learning process of others. We demonstrated the efficiency of the embedded learning mechanism with respect to different environmental factors.

Learning with ensemble of linear perceptrons

In this paper we introduce a model of ensemble of linear perceptrons. The objective of the ensemble is to automatically divide the feature space into several regions and assign one ensemble member into each region and training the member to develop an expertise within the region. Utilizing the proposed ensemble model, the learning difficulty of each member can be reduced, thus achieving faster learning while guaranteeing the overall performance.

An Interpretable Neural Network Ensemble

The objective of this study is to build a model of neural network classifier that is not only reliable but also, as opposed to most of the presently available neural networks, logically interpretable in a human-plausible manner. Presently, most of the studies of rule extraction from trained neural networks focus on extracting rule from existing neural network models that were designed without the consideration of rule extraction, hence after the training process they are meant to be used as a kind black box. Consequently, this makes rule extraction a hard task. In this study we construct a model of neural network ensemble with the consideration of rule extraction. The function of the ensemble can be easily interpreted to generate logical rules that are understandable for human. We believe that the interpretability of neural networks contributes to the improvement of the reliability and the usability of neural networks when applied to critical real world problems.

Selective attention improves self-organization of cortical maps with multiple inputs

Models of self-organizing cortical maps have focused on demonstrations with single objects in the environment. Recently, the validity of a traditional biological model has been questioned for the case of multiple simultaneous input sources. Here we show that the standard model is able to self-organize with multiple inputs. However, we also show that the ability to self-organization can be enhanced considerably by including top-down attention as well as some noise. The model is also used to simulate the development of tuning curves.

Class-proximity SOM and its applications in classification

In this study, we propose a model of self-organizing map (SOM) capable of mapping high dimensional data into a low dimension space by preserving not only the feature-proximity of the original data but also their class-proximity. A conventional SOM is known to map original high dimensional data with similar features into points located close to each other in the low dimensional map in a so called competitive layer. In addition to this feature, the proposed SOM is also able to map high dimensional data belonging to a same class in each others proximities. These characteristics retains the ability of the map to be used as a visualization tool of high dimensional data while also support the execution of high quality pattern classifications in the low dimensional map. In the experiments the classification performance of the proposed SOM is compared to that of MLP with regards to wide varieties of problems.

Modular robot with adaptive connection topology

In this study, we physically built hardware modules which enable us to freely construct robots with various morphologies. As opposed to the existing studies of modular robotics where the connection topology among the modules has to be hand-designed, our modules are able to adaptively modify their connection topology which enables them to generate an overall behavior as one robot. We ran several physical experiments where robots with various morphologies are assembled from the proposed modules to acquire several target behaviors.

Learning intialized by topologically correct representation

In this research, we proposed a model of a hierarchical three-layered perceptron, in which the middle layer contains a two dimensional map where the topological relationship of the high dimensional input data (external world) are internally represented. The proposed model executes a two-phase learning algorithm where the supervised learning of the output layer is proceeded by a self-organization unsupervised learning of the hidden layer. The objective of this study is to build a simple neural network model which is more biologically realistic than the standard Multilayer Perceptron model and that can form an internal representation that supports its learning potential. The characteristics of the proposed model are demonstrated using several benchmark classification problems.