Domain-Specific BERT Innovates Business Use of NLP
The performance of large-scale AI models has seen massive improvement, driving increased competition. As applications are being explored that leverage these models, the natural language processing (NLP) industry has begun to examine business applications for Financial BERT (Bidirectional Encoder Representations from Transformers), an AI language model developed by NTT DATA for the financial industry. Based on NTT BERT, which was built by NTT Laboratories using one of the largest Japanese-language text datasets in Japan, Financial BERT has learned financial text data uniquely compiled by NTT DATA. It eliminates the need to build domain-specific dictionaries or rules when analyzing documents that contain financial jargon or finance-specific context. This reduces the AI building process, while enabling highly accurate results. We are now building an infrastructure upon which optimized versions of BERT (initially as Japanese-language models) can be easily created for domains beyond finance. It will handle processes ranging from domain-specific data collection to model building, enabling these models to be applied to business practices.
Data Collaboration Technology Securely Integrates Learning Data
As the value of data-driven business rises, the use of AI across industries is furthering optimization and creating new opportunities. Meanwhile, concerns over leaks of confidential information are also increasing, necessitating a security-backed data collection technology. NTT DATA is working with academic institutions to develop a data collaboration technology that assures that each organization can keep its own information private while integrating with each other's data to build a highly accurate model. This technology works by converting confidential data to unreadable and neutral formats. This enables the creation of models that use learning data containing more variations and attributes, thereby producing more accurate AI models. The technology also uses fewer computing resources than other distributed learning methods, lowering the implementation threshold and making business application easier. Cross-industry data use will only grow. NTT DATA is actively engaged in the development and deployment of collaboration technology to accelerate this trend and create new value.
LOOKING AHEAD: Technology Trends Driving Digital Innovation
01.The Transformative Power of AI
The continuing growth of AI has led to advancements in increased model size and performance. These innovations are being applied to commodity AI research. Technologies supporting new AI uses will emerge, including efforts to improve learning data preparation. AI will evolve from a purpose-built tool to a broader exploratory technology.
02.The Complication of IT Infrastructures
As AI dramatically increases hardware and network performance requirements, it is driving innovations in miniaturization, new materials and processing methods – and a move to purpose-built, software-specific hardware designs. In supporting these challenging innovations, cloud service providers will develop expertise and lead the way on best practice.
03.Software-Driven Evolution
Competition among service providers will focus on delivering differentiated customer experiences through software. Manufacturers will rely on software to increase product value and accelerate deployment. Accordingly, companies will seek AI tools to increase productivity and new organizational structures to drive continuous improvement.
04.The Growth of Consolidated Data
Data collection and analysis are essential for effective planning and decision-making in an increasingly data-driven society. As this trend accelerates, technologies that integrate data and perform cross-sectional analysis will improve, as will technological counter-measures to protect privacy.
05.Simulation Takes on New Challenges
IT-based simulation has become increasingly accurate, expanding its application in automobile design, drug development and other fields. AI will make it easier for simulations to mimic reality while helping reduce calculations. Finally, by supporting the discovery of new materials and proteins, AI will enable new paths in research and development.
06.Distance Accelerates Automation
Technologies that capture human work and automate those tasks are proven to increase productivity for remote workers. By enabling AI to learn as an apprentice does by observing a master – and eliminating the need for step-by-step programming – these technologies are bringing AI to new applications
Contact GroupBDO to help drive new business development opportunities in the mega-enterprise BPS arena.
clientservices@groupbdo.com