The Taiwan Banker

During his May visit to Taiwan, Nvidia CEO Jensen Huang revealed that it is teaming up with Cathay United Bank (CUB), E.SUN, and CTBC on AI. Coincidentally, these three domestic financial holding giants are also key players in the Financial Supervisory Commission's (FSC’s) FinTech Alliance. Cathay Financial Holdings (CFH) Executive Vice President Jeremy Liu said that its cooperation with Nvidia includes technical exchanges such as AI framework training, going beyond simple hardware computing power. Its AI application scope is wide-ranging, including risk management, process improvement and customer identification, integrating both cloud and on-prem deployment. CUB launched an AI management associate (MA) recruitment plan this year to scale up its AI applications and cultivate talent.Liu said that CFH has deeply applied Nvidia’s NeMo framework for AI training, and has used its GPUs to strengthen its computing power. Executive Vice President Ming-Chiao Liang said that the focus of its cooperation with Nvidia is to build future financial AI agents for Taiwan through Nvidia's technical strength and computing power. Initial results have been achieved over the past three months in building a large language model (LLM) with knowledge of finance and credit in Taiwan. Nvidia has many frameworks for agentic AI models, which will help to build a foundation for more diverse applications in the future.Liu said that the focus of CFH’s continued cooperation with Nvidia will be to increase the technological capabilities of its AI agents in learning, memory processing, environmental interaction, and learning and growth. Its longer-term goal is to create an ecosystem of mutual cooperation between agents and cross-industry partners.CFH has established a dataset for training AI agents, including laws, regulations, and financial knowledge, and has tested its training results by having agents simulate the Taiwan financial credit license examination. Liu also explained the various AI agents currently available at CFH, saying that it aims for omni-scenario smart finance, including price comparison, discounts, payment, wealth management, credit cards, investment, CUBE, credit, lending, insurance, asset management, and payments. It can also personalize agents for individual customers.Jyh-Shing Roger Jang, Chief Technology Officer of E.SUN Financial Holdings, said that E.SUN will launch two core AI projects, aiming to improve customer experience with smarter and faster services: an AI chatbot and fraud prevention. Nvidia has contributed greatly to this, especially through its technical frameworks and computing power. Jang further explained that E.SUN will introduce Nvidia H100 GPUs and NeMo to improve computing performance while reducing costs. Jang said that after confirming that the technology is feasible, safe, and reliable, E.SUN will discuss relevant standards with the FSC under the framework of the Fintech Industry Alliance to steadily add new external services.Regarding fraud prevention, Jang said Nvidia’s main contribution is computing power through Nvidia Rapids and with H100 GPUs. Jang mentioned that E.SUN will embed graph analysis into its existing transaction fraud detection model, visualizing complex transaction relationships, increasing detection accuracy two times, and accelerating end-to-end fraud model training by about 880 times. In his view, these two applications will not only make internal operations more efficient, but also bring security and convenience to consumers.Jang believes that the biggest contribution of Nvidia technology in fraud prevention is its program libraries, including an industry framework with strong performance. A convenient software framework and fast GPUs are two sides of the same coin, both of which are strengths of Nvidia. Jang said that most of Nvidia's program library is processing graphics drawn by different nodes, used not only for fraud prevention, but also for a diverse array of other applications as well.E.SUN will also focus on chatbots. Jang said that E.SUN will build a new agent platform integrating multiple chatbots into a single dialogue interface, while accelerating development and security testing through a standardized architecture. The bank needs to do KYC procedures on account opening and other tasks, for instance. The integrated use of AI will enable every server to call this KYC agent.As with Cathay, NeMo is also a key point in E.SUN’s cooperation with Nvidia. Jang said that Nvidia provides financial institutions with many frameworks, including a variety of APIs to call program algorithms. It can write programs to allow GPUs to run in parallel, and then use the characteristics of its GPUs to achieve acceleration.Jang also said that these frameworks can retrain different language models. Different business segments will have different LLMs. For example, some models only specialize in credit underwriting, and can only be used within the bank and cannot be connected to the external cloud. Liu said that the "financial brain" for AI banking is built on top of LLMs, and different services can be connected to create more agents.CTBC, meanwhile, said that GPUs are a major focus of its cooperation with Nvidia. Currently, CTBC is using AI extensively to improve its internal efficiency and risk control, such as developing technology to improve the effectiveness of fraud prevention. It is also continuing its technical exchanges and cooperation with Nvidia on areas such as model performance and accelerating the development of generative AI application scenarios.Looking ahead, Liang said that the financial industry will enter a new stage of greater personalization through AI, from online to mobile banking. Liang also analyzed the market changes which come with the ability to make autonomous decisions, respond to changes in the environment, and perform integrated tasks across domains, including large changes in customer needs, higher difficulty of risk control, and explosive growth in information volume and process complexity. The author is a senior reporter on the financial industry