Project overview
As well as data analytics for sensor data in manufacturing industries,
we are also focusing on data analytics for supporting business decisions
based on various data generated by business activities. Our objective is
"data analytics for converting data to business value", and we pursue this goal in collaboration with our partners such
as leading-edge clients, other IBM research labs, other IBM departments,
and IBM's business partners.
In many industries, globalization is becoming one of the most important
issues. Business have discovered that making use of vast amount of heterogeneous
data distributed over the world for decision making is enormously challenging,
and so data analytics technologies can play an essential role. We hope
our research efforts in data analytics lead IBM and our clients to transforming
themselves into globally integrated enterprises.
It is challenging, but at the same time, thrilling and exciting to bring
our cutting edge skills in machine learning and data mining to the solution
of complex real-world problems troubling our leading-edge partners in various
industries such as financial services, distribution, and retail sales.
(See also our notable results in manufacturing industries). Also, when we recognize that IBM is a big, multinational company with
many departments, its internal data is one of the most fascinating candidates
for study.
In parallel with making an impact on data analytics technologies in the
real world, we are also focusing on research and development into new data
analytics techniques. Risk-sensitive data analytics that explicitly handles (long-term) risks and opportunities in machine learning is one of the themes that we have been working on for years.
Recently, we developed new methods that are highly suitable for complex, dynamic and heterogeneous real-world data. For example, we developed sampling methods for biased data （Hido&Kashima, SDM2008), a method that explains changes in time-varying data （Hido et al., PAKDD2008), and a learning method which enables accurate predictions of imminent changes （Tsuboi et al, SDM2008、Sugiyama et al, NIPS2007).