Communication(NTT Communication Science Laboratories)
|To the site||http://www.kecl.ntt.co.jp/rps/english/Research_e/Research_Index_e.html|
We study basic technologies based on statistical machine learning and data mining.
For training deep neural network models, we need a huge amount of data. However, in real-world settings, enough data can be unavailable. Meta-learning is attracting attention for a technique to improve performance with a limited amount of data by learning how to learn from various tasks. We are working on research to meta-learn from various tasks, such as spatial analysis, anomaly detection, and time-series forecasting.
Spatio-temporal data analysis
With the development of information and communication technology, a variety of spatio-temporal data related to weather, traffic, etc., are becoming readily available. Time-series forecasting and knowledge extraction for spatio-temporal data are important in science and industry. We are developing methods based on deep learning and Gaussian processes for refining coarse-grained aggregated data gathered from cities and for learning and simulating physical phenomena from observation data.
Information diffusion analysis on social networks
News, rumors, and reputation are diffused through social networks. We model such information diffusion processes based on probabilistic models, and use them for predicting future trends, and for discovering hidden influence relationships among users.
Our research activities include various phases, including proposing new theories and modeling, developing effective algorithms and data structures, and applying techniques to new interesting applications. We are interested in processing various data, such as Web and language data, speech sounds, images, and sensor data. Our everyday efforts are aimed at the world's first proposal and verification of new techniques, or the world's best performance of certain tasks. Students can use rich computer and human resources of NTT Communication Science Laboratories such as large clusters of high-performance servers.
- Yusuke Tanaka, Tomoharu Iwata, Naonori Ueda, Symplectic Spectrum Gaussian Processes, Learning Hamiltonians from Noisy and Sparse Data, NeurIPS, 2022.
- Tomoharu Iwata, Yusuke Tanaka, Few-shot learning for spatial regression via neural embedding-based Gaussian processes, Machine Learning, 2021.
- Tomoharu Iwata, Atsutoshi Kumagai, Meta-learning from Tasks with Heterogeneous Attribute Spaces, NeurIPS, 2020.
- Yusuke Tanaka, Toshiyuki Tanaka, Tomoharu Iwata, Takeshi Kurashima, Maya Okawa, Yasunori Akagi, Hiroyuki Toda, Spatially Aggregated Gaussian Processes with Multivariate Areal Outputs, NeurIPS, 2019.