Toshihiko Yanase

Evolutionary motion design for humanoid robots

We propose a new approach to generating the motion of humanoid robots intuitively by means of Interactive Evolutionary Computation (IEC). In our system, novice users are able to design effective motions through the subjective evaluation of displayed individuals, even if they do not have any technical knowledge. The motions evolved by the IEC system are not necessarily stable nor feasible in real environments. Thus, appropriate adjustments are required to revise the motions. For this purpose, we use a real-valued GA in a dynamic simulator. We empirically show the effectiveness of our approach by designing a kick motion for a humanoid robot.

Binary encoding for prototype tree of probabilistic model building GP

In recent years, program evolution algorithms based on the estimation of distribution algorithm (EDA) have been proposed to improve search ability of genetic programming (GP) and to overcome GP-hard problems. One such method is the probabilistic prototype tree (PPT) based algorithm. The PPT based method explores the optimal tree structure by using the full tree whose number of child nodes is maximum among possible trees. This algorithm, however, suffers from problems arising from function nodes having different number of child nodes. These function nodes cause intron nodes, which do not affect the fitness function. Moreover, the function nodes having many child nodes increase the search space and the number of samples necessary for properly constructing the probabilistic model. In order to solve this problem, we propose binary encoding for PPT. Here, we convert each function node to a subtree of binary nodes where the converted tree is correct in grammar. Our method reduces ineffectual search space, and the binary encoded tree is able to express the same tree structures as the original method. The effectiveness of the proposed method is demonstrated through the use of two computational experiments.

Evolutionary Multi-Objective Optimization for Biped Walking

We introduce an application of Evolutionary Multi- Objective Optimization on multi-layered robot control system. Recent robot control systems consist of many simple function modules. The parameter settings for most of these modules were manually adjusted in previous research. Our goal is to develop an automatic parameter adjustment method for the robot control system. In this paper, we focused on three modules as the experiment environment: whole-body motion generator, footstep planner and path planner. At first the features of these three modules are examined. Then we discuss the trade-off relationship between the requirements of each module. Finally, we examined an application of Evolutionary Multi-Objective Optimization on this problem.

End-to-end Argument Generation System in Debating

We introduce an argument generation system in debating, one that is based on sentence retrieval. Users can specify a motion such as This house should ban gambling, and a stance on whether the system agrees or disagrees with the motion. Then the system outputs three argument paragraphs based on “values” automatically decided by the system. The “value” indicates a topic that is considered as a positive or negative for people or communities, such as health and education. Each paragraph is related to one value and composed of about seven sentences. An evaluation over 50 motions from a popular debate website showed that the generated arguments are understandable in 64 paragraphs out of 150.

bunji at SemEval-2016 Task 5: Neural and Syntactic Models of Entity-Attribute Relationship for Aspect-based Sentiment Analysis.

This paper describes a sentiment analysis system developed by the bunji team in SemEval2016 Task 5. In this task, we estimate the sentimental polarity of a given entity-attribute (E#A) pair in a sentence. Our approach is to estimate the relationship between target entities and sentimental expressions. We use two different methods to estimate the relationship. The first one is based on a neural attention model that learns relations between tokens and E#A pairs through backpropagation. The second one is based on a rule-based system that examines several verb-centric relations related to E#A pairs. We confirmed the effectiveness of the proposed methods in a target estimation task and a polarity estimation task in the restaurant domain, while our overall ranks were modest.

Neural Attention Model for Classification of Sentences that Support Promoting/Suppressing Relationship

Evidences that support a claim “a subject phrase promotes or suppresses a value” help in making a rational decision. We aim to construct a model that can classify if a particular evidence supports a claim of a promoting/suppressing relationship given an arbitrary subject-value pair. In this paper, we propose a recurrent neural network (RNN) with an attention model to classify such evidences. We incorporated a word embedding technique in an attention model such that our method generalizes for never-encountered subjects and value phrases. Benchmarks showed that the method outperforms conventional methods in evidence classification tasks.

Analysis and Learning Frameworks for Large-Scale Data Mining

The first framework is for analysis phase, in which we find out how to utilize business data through trial and error. The proposed framework stores tree-structured data using vertical partitioning technique, and uses Hadoop MapReduce for distributed computing. These methods enable to reduce disk I/O load, and to avoid computationally-intensive processing, such as grouping and combining of records.

Evolutionary Multi-Objective Optimization for Biped Walking of Humanoid Robot

The recent remarkable progress of robotics research makes advanced skills for robots to solve complex tasks. The divide-and-conquer approach is an intuitive and efficient method when we encounter complex problems. Being a divide-and-conquer approach, the multilayered system decomposes the problem into a set of levels and each level implements a single task-achieving behaviour. Many researchers employ this approach for robot control system, dividing a complex behaviour into several simple behaviours. For example, Lie et al. developed the Evolutionary Subsumption Architecture, which enables for heterogeneous robots to acquire the cooperative object transferring task (Liu & Iba, 2003). The autonomous locomotion of humanoid robots consists of following modules: global path planning using given geometrical information, local path planning based on the observation of environment, footstep planning, and whole-body motion generator. Since these modules mainly exchange the information with their neighbours, we can observe that they are hierarchically arranged from the aspect of communication. The parameter settings among these modules are necessary to adapt the system to the targeting environment. The problem involves a number of conflicting objectives such as stability of the robot motion and speed of locomotion. In this paper, we present a parameter tuning method for multi-layered robot control system by means of Evolutionary Multi-Objective Optimization (EMO). We explore the set of parameters for modules to adapt various kinds of environments. Switching these parameter sets enables us to operate the robots effectively. We developed three modules as the experiment environment: walking pattern generator, footstep planner and path planner. In the experiment, we focused on the footstep planner shown in Fig. 1, which realizes collisionfree walking. The parameter setting was manually adjusted in previous researches (Kuffner et al., 2003; Chestnutt et al., 2005). We discuss the conflicting objectives for the optimization footstep planner, and introduce a parameter setting method using EMO. Then we propose a simple rule to use parameter sets obtained by EMO to adapt the footstep planner to both crowded and sparse fields. The rest of the paper is organized as follows: Section 2 describes our robot control system, Section 3 shows an experiment of the parameter setting of the footstep planner, and Section 4 shows an application using the parameter setting obtained from Section 3 and a 8