CALC

A Big Data Analysis Tool

What is CALC?

The ability of machine learning systems to recognize patterns accurately has advanced by leaps and bounds in recent years. These systems are becoming widely used in industries involving identification, classification, and prediction. So-called black boxes - systems where it is difficult to explain where exactly the results come from, deep learning being a prominent example - are now known for producing better results than conventional models.

However, there is a demand to understand the causal relation or logical connection behind the curtain of such big data; we need technologies that clearly delineate these black boxes. In response to this demand, this project attempts to use causal models to interpret the many phenomena that are difficult to understand using traditional notions of correlation or similarity.

[regular correlation analysis vs. CALC]
Takashi Isozaki of Sony CSL has proposed high-precision inferring algorithms and their extensions those that can deal with latent variables, and has proposed some unique methods for it, such as thermodynamic statistical analysis.

This suite of technologies, given the collective name CALC, was refined with the use of actual Sony Group data and used to solve a variety of analysis-related problems. Along the way, we received lots of feedback that advanced our technology and our knowhow. CALC proved to have applications in a wide range of data analysis involving design, manufacturing, marketing and services in fields from electronics and finance to entertainment.

Based on these results, in 2017 Sony CSL and two other companies, Information Services International-Dentsu (ISID) Ltd. and Koozyt Inc., began a service that sells CALC technology licenses, provides analysis consulting, and so on. Currently the team includes Atsushi Noda of Sony CSL, people from the Sony Group, ISID and Koozyt, and Calc technology and systems R&D is progressing.

[what you understand with correlation analysis (L) vs. what you understand with CALC (R)]

2019/02/22
Go to top