Tatsushi Matsubayashi

Topigraphy: visualization for large-scale tag clouds

This paper proposes a new method for displaying large-scale tag clouds. We use a topographical image that helps users to grasp the relationship among tags intuitively as a background to the tag clouds. We apply this interface to a blog navigation system and show that the proposed method enables users to find the desired tags easily even if the tag clouds are very large, 5,000 and above tags. Our approach is also effective for understanding the overall structure of a large amount of tagged documents.

Initial Positioning Method for Online and Real-Time Dynamic Graph Drawing of Time Varying Data

Microblogging services generate huge histories that suit animated visualization based on graph drawing, but response speeds are insufficient. This paper proposes an effective updating method for the real-time visualization of time varying data. We propose an initial positioning method by combining Additional Edge Resizing (AER) with Sorted Sequential Barycenter Merging (SSBM). AER resizes edges between existing nodes when new edges are added before updating the visualization. SSBM initially positions multiple new nodes sequentially through priority based on the degree of connections to the existing graph. The proposed method prevents a decrease in readability at data update and achieves fast convergence with high accuracy. Quantitative and qualitative evaluations using the energy function and RMSE are detailed along with some visualization results. A graph drawing tool based on HTML5 is also introduced as an implementation of the proposed method and present some use cases for twitter data.

Regular behavior measure for location based services

We introduce a method that can measure the degree of regularity or irregularity of the behavior for enhancing the performance of location-based services (LBSs) such as check-in. It is still challenging for LBSs to determine the places to recommend that best suits the users needs. Our aim is to identify the users status (regular or irregular) of each check-in. Most previous studies approached this problem by acquiring usual locations (e.g., home or office) or assessing check-in frequency. We propose more effective measure by using a multinomial-distribution-based method that considers the periodic check-ins of the user on various time-scales. Our method can accurately identify irregular check-ins even in usual locations and we find that the users tend to continue irregular check-ins in a certain range of time.

Mobile topigraphy: large-scale tag cloud visualization for mobiles

We introduce a new mobile topigraphy system that uses the contour map metaphor to display large-scale tag clouds. We introduce the technical issues for topigraphy, and recent requirements for and developments in mobile interfaces. We also present some applications for our mobile topigraphy system and describe the assessment on two initial applications.

Multi Agent Flow Estimation Based on Bayesian Optimization with Time Delay and Low Dimensional Parameter Conversion

Forming comprehensive security plans is essential to ensure safety at large events. The Multi Agent Simulator (MAS) is widely used for preparing security plans that will guide responses to accidents at large events. For forming security plans, it is necessary that we simulate crowd behaviors that reflect real world observations. However, crowd behavior simulations require OD information (departure time, place of Origin, and Destination) of each agent. Moreover, from the viewpoint of protecting personal information, it is difficult to observe the complete and detailed trajectories of all pedestrians. Therefore, the OD information should be estimated from the data observed at several points, usually the number of people passing fixed points. In this paper, we propose an accurate method for estimating OD information; it has three features. First, by using Bayesian optimization (BO) which is widely used to find optimal hyper parameters in the machine learning fields, the OD information is efficiently and accurately estimated using fewer parameter searches. Second, by dividing the time window and ignoring the identity of the observed people, the parameter dimension of the OD information is reduced to yield a solvable search space. Third, by considering the time delay created by the physical separation of the observation points, we develop a more accurate objective function. Experiments evaluate the proposed method using the data collected at three events (University festival, projection-mapping event, and music live), and the accuracy with which reproduction MAS can reproduce the people flows is assessed. We also show an example of the MAS-based process used in making guidance plans to reduce crowd congestion.

Improving Route Traffic Estimation by Considering Staying Population

Estimating the number of people who travel by a particular route (route traffic) is an important task for multi-agent simulations in the transportation field. Previous studies have used the traffic count to estimate the route traffic. We propose a new method that utilizes the staying population (stay count) in addition to the traffic count. With experiments using synthetic data, we demonstrate that the proposed method achieves a \(19.85 \%\) smaller error rate than the conventional method when the traffic count’s observation is incomplete. In addition, we analyze real-world data.

Generalized Inverse Reinforcement Learning with Linearly Solvable MDP

In this paper, we consider a generalized variant of inverse reinforcement learning (IRL) that estimates both a cost (negative reward) function and a transition probability from observed optimal behavior. In theoretical studies of standard IRL, which estimates only the cost function, it is well known that IRL involves a non-identifiable problem, i.e., the cost function cannot be determined uniquely. This problem has been solved by using a new class of Markov decision process (MDP) called a linearly solvable MDP (LMDP). In this paper, we investigate whether a non-identifiable problem occurs in the generalized variant of IRL (gIRL) using the framework of LMDP and construct a new gIRL method. The contributions of this study are summarized as follows: (i) We point out that gIRL with LMDP suffers from a non-identifiable problem. (ii) We propose a Bayesian method to escape the non-identifiable problem. (iii) We validate the proposed method by performing an experiment on synthetic data and real car probe data.

Non-negative periodic component analysis for music source separation

Non-negative matrix factorization (NMF) is attracting a lot of attention as a powerful technique for music transcription and audio source separation. With this approach, the magnitude (or power) spectrogram of a mixed signal, interpreted as non-negative matrix Y, is factorized into the product of two non-negative matrices, dictionary matrix H and activation matrix U. Each template vector in the dictionary matrix corresponds to the prototype spectrum of a certain sound component. So that NMF can output a musically meaningful as well as accurate decomposition, we must extend the NMF model HU with reasonable assumptions. One such assumption involves the temporal regularity underlying the onset occurrences of musical notes and drum sounds. In particular, the periodicity is most apparent in the onset timings of bass instruments and drum sounds. Motivated by this fact, this paper proposes a new constrained NMF that appends the objective function of NMF a criterion that promotes the periodicity of the time-varying amplitude associated with each basis spectrum and derives an iterative algorithm for solving the regularized optimization problem of interest. The proposed method is particularly noteworthy in that it makes it possible to extract audio events that occur periodically in an unsupervised manner. Unsupervised and supervised audio source separation experiments show that the proposed method significantly outperforms conventional approaches including original NMF and periodicity-aware music/voice separation.

Blocked Time-Step Algorithm for Accelerating k-Means and Fuzzy c-Means

The k-means and fuzzy c-means algorithm are widely used for discovering clusters in data. In this paper, we propose a new acceleration method using blocked time step algorithm. It reduces the number of distance calculations efficiently by changing the frequency of updating membership grades (searching the nearest cluster centers). By restricted time step with the discrete number as a power of two, distance for the objective function will be calculated only synchronized objects at the blocked time. As a results, the experiments show the proposed algorithm reduces distance calculations by 20 -- 60% while keeping the clustering accuracy.