- Fujitsu selected for “Research and Development Project of the Enhanced Infrastructures for Post-5G Information and Communication Systems”
- Project aims to enhance Japan’s generative AI capabilities under GENIAC initiative
- Fujitsu to develop LLMs combining knowledge graphs for logical reasoning
- Recent release of Fugaku-LLM showcases enhanced Japanese language proficiency
- Focus on addressing hallucinations in LLMs for increased reliability
- Plans to introduce specialized LLMs for various industries by fiscal 2024
- Development of LLMs specialized in knowledge graph generation and inference
- Approach aims to ensure precise outputs compliant with complex regulatory frameworks
- Future offerings to include evaluation scripts and insights via platforms like Hugging Face and GitHub
Main AI News:
In a strategic move, Fujitsu has been selected to spearhead the “Research and Development Project of the Enhanced Infrastructures for Post-5G Information and Communication Systems/Development of post-5G information and communication systems.” This prestigious subsidy project, initiated by Japan’s New Energy and Industrial Technology Development Organization (NEDO), under Japan’s Ministry of Economy, Trade and Industry’s (METI) Generative AI Accelerator Challenge (GENIAC) initiative, aims to bolster Japan’s prowess in generative AI. Fujitsu will embark on pioneering research and development endeavors, marrying knowledge graphs with large language models (LLMs) to actualize LLMs capable of logical reasoning.
Building upon its extensive portfolio of generative AI technologies tailored for business applications, Fujitsu recently unveiled the Fugaku-LLM. This large language model boasts augmented Japanese language proficiency, marking a significant milestone in Fujitsu’s journey of innovation. The company is committed to fortifying the development of specialized LLMs tailored for diverse industries and enterprises.
Central to Fujitsu’s research and development initiatives is the mitigation of hallucinations in large language models (LLMs). These hallucinations, a common pitfall, occur when generative AI produces seemingly plausible yet inaccurate or irrelevant outputs. Fujitsu is on a mission to pioneer a groundbreaking solution that bolsters the reliability of LLMs by integrating them with knowledge graphs. This integration aims to empower LLMs for applications necessitating stringent compliance and elucidation, such as tort determination in legal proceedings, internal controls in finance, and diagnostic support in healthcare.
Fujitsu envisions that this innovative fusion of LLMs with knowledge graphs will pave the way for the creation of an LLM for logical reasoning, delivering lucid and comprehensible outputs.
As part of its roadmap, Fujitsu aims to introduce this cutting-edge technology to the Japanese market by the culmination of fiscal year 2024. This move aligns with the company’s overarching mission to bolster conversational AI resilience against hallucinations and adversarial attacks, as announced in September 2023.
LLMs Empowering Logical Reasoning
To tackle the reliability conundrum inherent in LLM outputs, Fujitsu has embarked on the development of LLMs specialized in knowledge graph generation and inference. This approach leverages knowledge processing technologies to enable LLMs to logically deduce answers based on generated knowledge graphs derived from natural language regulations and rules. Positioned as the cornerstone of a generative AI trust technology, Fujitsu anticipates the elimination of output instability, ensuring precise outputs compliant with intricate regulatory frameworks.
Under the GENIAC program, Fujitsu will develop two distinct specialized LLMs: one dedicated to knowledge graph generation and the other focused on knowledge graph inference. These LLMs will convert natural language documents into knowledge graphs and logically aggregate information to deliver accurate responses to queries.
To expedite the development process within the project’s time constraints, Fujitsu will initially create a pre-trained LLM shared by both specialized models. This approach streamlines development by enabling LLMs to process both natural language documents and knowledge graphs, augmented by a bilingual corpus. Subsequently, specialized LLMs will undergo concurrent instruction learning processes tailored to their respective functions.
Future Endeavors
Looking ahead, Fujitsu is poised to offer the specialized LLMs for knowledge graph generation and inference, alongside evaluation scripts and invaluable insights garnered throughout the project. These resources will be disseminated through platforms like Hugging Face, GitHub, the Fujitsu blog, and the GENIAC community, in full compliance with pertinent terms of use and licensing agreements. Additionally, Fujitsu plans to unveil these groundbreaking technologies via Fujitsu Kozuchi, its dedicated AI service, facilitating accelerated testing and deployment for users seeking to harness advanced AI capabilities.
Conclusion:
Fujitsu’s pivotal role in the GENIAC project and its innovative strides in large language models signify a significant advancement in AI development. By addressing the reliability challenges associated with LLMs and introducing specialized models tailored for logical reasoning, Fujitsu is poised to revolutionize the AI landscape. This development underscores the company’s commitment to delivering cutting-edge solutions that cater to the evolving needs of industries, promising enhanced compliance, and explainability in AI applications. As Fujitsu prepares to unveil its groundbreaking technologies to the market, businesses can anticipate a transformative shift in how AI is leveraged to drive efficiency and innovation.
