- Semiconductor BusinessHOME
- Products and Services of Macnica,Inc.
-
technical information
-
Events and Seminars
- Handling Manufacturer
- Support
- Inquiry
- Click here to purchase products
- Semiconductor business e-mail magazine registration
The seminar "NVIDIA Financial AI Meet-up with Macnica - AI in your own hands. Practical companies discuss the use of financial AI" was heldon June 24, 2025, and many industry professionals took to the stage to share the latest developments in AI use in the financial industry.
This article introduces the content of the lecture given by Xianchao Wu of NVIDIA Japan.
At the end of the article, we have information about on-demand videos, so please read to the end.
Organizer: Macnica
Sponsor: NVIDIA G.K.
| Lecture | Lecture title | profile | |
| 1 | GTC2025 Digest (Focusing on presentations from financial companies) |
Nvidia G.K. Senior Business Development Manager |
Hiroshi Hirahata Mr. |
| 2 | Strategic AI adoption and business-specific use cases |
SMBC Global Investment & Consulting Co., Ltd. General Manager |
Mr. Hirofumi Yamada |
| 3 | Utilizing the financial specialized model FineNemotron | Senior Solutions Architect, NVIDIA G.K. | Mr. Xianchao Wu |
| 4 | The importance and capabilities of on-premise and local LLM for secure AI utilization in financial institutions | Representative Director and President of Ippu Senkin Co., Ltd. | Hideya Suzuki |
| 5 | Easily start building a local LLM with NVIDIA NIM | Macnica | Mr. Hitoshi Onodera |
| 6 | Expanding Use of Generative AI and Corporate Risks: The Need for AI Security |
Cisco Systems, Inc. Robust Intelligence Country Manager / Cisco Business Development Manager, AI |
Mr. Taiichi Hirata |
| 7 | Kaggle Grandmaster Accelerates Data Science with RAPIDS |
Nvidia G.K. KGMoN（NVIDIA Kaggle Grandmaster） |
Mr. Kazuki Onodera |
| 8 | Accelerating Trading Strategy Development Using RAPIDS and CUDA at a Quantitative Hedge Fund |
Okun Co., Ltd. Project Manager / Data Scientist |
Mr. Nobutaka Takeuchi |
| 9 | Automatic generation of analyst reports using generative AI: Improving accuracy through fine-tuning learning |
aiQ Inc. Representative Director and President |
Mr. Hiroki Yamamoto |
| Special lecture | Working with AI: Using AI to Scale Your Team Discussion Paper |
Financial Services Agency Counselor, Risk Analysis Division, Policy Management Bureau |
Ms. Hozue Igarashi |
Utilizing the financial-specialized model FinNemotron
Xianchao Mr. Wutwist, developing and releasing language models and inference datasets specialized for the financial sector.FollowI was introduced.
The inference model featured this time is "FinNemotron-R1" and an inference dataset based on Japanese financial statements "FinQA-Ja "is.
Financial inference model "FinNemotron-R1"
This model aims to support decision-making in the financial sector through inference based on financial data. Comparing it with existing financial models (Bloomberg, PFN, Fin-R1, etc.), the lack of Japanese language support and issues with open source were pointed out. It was emphasized that the financial sector in particular requires white Box models that can clearly show the inference process, rather than black Box AI.
"FinQA-Ja" inference dataset using Japanese financial statements
NVIDIA has developed the "FinQA-Ja" inference dataset based on Japanese financial statements, utilizing XBRL-formatted financial data published by the Financial Services Agency's EDINET. This dataset generates questions from tabular financial information, and the model learns inference steps to derive answers while providing evidence.
For example, to answer a question such as "What is the ratio of unpaid compensation to total liabilities in the 15th period?", the process includes searching for the relevant columns and rows, calculating, and deriving the answer. Data categorized by difficulty level is also compatible with SFT (supervised fine tuning) and reinforcement learning, enabling responses that are practical in nature.
Model performance and implementation environment
The presentation introduced a highly accurate inference environment using the Qwen-based 7B model. This model achieved an accuracy of approximately 78-80 % on the English dataset and approximately 70% on the Japanese dataset, while also reducing the GPU load. Future deployment of even larger models, such as 32B, 443B, and 240B, is planned. It was also emphasized that the model can be deployed securely and flexibly in cloud and offline environments through NVIDIA platforms such as NeMo and NIM. The ability to choose the model based on corporate privacy protection and operational environment is cited as a major practical advantage.
Simply register to watch the video
If you register using the form below, you will receive a URL to watch the on-demand video of the seminar by NVIDIA's Xianchao Wu introduced here. If you missed it or attended the event and would like to watch it again, please take this opportunity to register!
The latest trends in AI for the financial industry as seen at GTC
Also in the presentation by NVIDIA, they introduced the latest examples and technologies of AI use in the financial industry, including several case studies presented at GTC (one of the world's largest technology conferences hosted by NVIDIA). For more details, please see the conference presentation video below.
Click here for inquiries
