Daishi Kato

Creating GUIs for Web services

Next-generation Web browsers Will have form navigation, variable-size lists, and dynamic enumeration built in to increase user friendliness.

Exponential Family Tensor Factorization for Missing-Values Prediction and Anomaly Detection

In this paper, we study probabilistic modeling of heterogeneously attributed multi-dimensional arrays. The model can manage the heterogeneity by employing an individual exponential-family distribution for each attribute of the tensor array. These entries are connected by latent variables and are shared information across the different attributes. Because a Bayesian inference for our model is intractable, we cast the EM algorithm approximated by using the Lap lace method and Gaussian process. This approximation enables us to derive a predictive distribution for missing values in a consistent manner. Simulation experiments show that our method outperforms other methods such as PARAFAC and Tucker decomposition in missing-values prediction for cross-national statistics and is also applicable to discover anomalies in heterogeneous office-logging data.

Adding AI to Web Services

The FX-Project consisted of members of the Stanford Logic Group industrial visitors from NEC and Intec Web & Genome working together to develop a new technologies based upon the combination of web services and techniques from artificial intelligence using our experience in AI-based software agents. This two-year project ran from April of 2001 until the end of March 2002 and explored the then emerging functionality of web services. This paper is a result of our findings.

Exponential family tensor factorization: an online extension and applications

In this paper, we propose a new probabilistic model of heterogeneously attributed multi-dimensional arrays. The model can manage heterogeneity by employing individual exponential family distributions for each attribute of the tensor array. Entries of the tensor are connected by latent variables and share information across the different attributes through the latent variables. The assumption of heterogeneity makes a Bayesian inference intractable, and we cast the EM algorithm approximated by the Laplace method and Gaussian process. We also extended the proposal algorithm for online learning. We apply our method to missing-values prediction and anomaly detection problems and show that our method outperforms conventional approaches that do not consider heterogeneity.

A scalable interest-oriented peer-to-peer pub/sub network

There has been a big challenge in structured peer-to-peer overlay network research area. Generally, a structured overlay network involves nodes evenly or based on their resource availabilities, and gathers nodes’ resources to achieve some bigger tasks. The challenge here is to gather resources based on nodes’ interests, and only interested nodes are involved in a certain task. Toward this challenge, we propose a new scheme to a peer-to-peer publish/subscribe network. Publish/subscribe represents a new paradigm for distributed content delivery. It provides an alternative to address-based communication due to its ability to decouple communication between the source and the destination. We propose a Bloom filter based mapping scheme to map IDs to nodes’ interests in addition to new interest proximity metric to forward events and to build nodes’ routing tables. We also propose a new approach called “shared interest approach” for network discovery. To evaluate the algorithms proposed in this work, we conducted simulations in both static and dynamic settings, and found a low false positive rate. We also discuss about a well-known application called Twitter, and show how our scheme would work in a real environment.

Flavor visualization: Taste guidance in co-cooking system for coexistence

Currently we are developing a co-cooking system that helps users to make similar tasting dishes, even though users may all be in remote locations and potentially cooking at different times.

An emulator for peer-to-peer distributed hash tables

Distributed hash tables (DHTs) are one of the hottest topics in large-scale peer-to-peer network research. We propose a method for evaluating DHTs by emulator, which allows us to evaluate not only DHT algorithms but also DHT implementations. Evaluating DHT implementations is important for DHT application developers because their performance influences application design. We developed a DHT emulator that runs in a local environment, and controls several DHT implementations based on a scenario. Because a scenario allows us to repeat evaluations, we can compare DHTs by one scenario and find behavior patterns by slightly changed scenarios. Five use cases are demonstrated to show the capabilities of Peeremu, and some results show DHT characteristics that cannot be obtained by simulating DHT algorithms. We hope this method helps application developers to understand DHTs and utilize them to create a better user experience.

Latency model of a distributed hash table with big routing table

In peer-to-peer research, one of the most popular areas is distributed hash table (DHT). Among many topics in the DHT area, this paper focuses on DHT latencies, which are mainly caused by its basic multi-hop lookup function. Some DHTs already have the capability of building big routing tables to reduce hop counts. However, none of them are explicitly trying to enlarge the routing table and lower the hop count. This paper provides a simple latency model of DHT and discusses how big routing tables help reduce latency.

The programming of robots by haptic means

According to the theory of constructivism, children can learn a lot not just from being taught but also through their play experiences, especially when they are designing and creating things. By playing with blocks, for example, children learn by exploration and experimentation how the physical world works [Piaget 1955]. It is also important for learners with different knowledge to collaborate in order to arrive at a shared understanding [Duffy and Jonassen 1992]. Based on these ideas, our objective has been to develop a system to program robots using the sense of touch. We believe that enabling robot programming through physical touch encourages play, learning, sharing and collaboration.

Group feature extraction based on matrix factorization from long-range office-logging data

To increase the productivity of knowledge workers, it is necessary to manage their organization so that they are motivated to collaborate with each other for their synergy. However, it is difficult for managers to grasp the explicit interactions of workers in the organization all the time. Owing to advanced communications technology, and the reduced size and improved capabilities of computers, we are able to record group behaviors as logging data in the office. The aim of this study is to extract features of group behavior from long-range office-logging data. We apply principal component analysis to the data matrix whose element is the mean travel velocity calculated from an individual’s trajectory per day. The results demonstrate the feasibility of our approach, since nontrivial informative group features can be extracted.