August 9th, Guoqian Ecological Circle
The fourth special training event of 2024 was successfully held
The theme of this event is
AI Big Model Security
Host: Guoqian Medical Technology (Suzhou) Co., Ltd
Guoqian Venture Capital Management (Suzhou) Co., Ltd
Special guest speakers for this issue: Dr. Qiu Han and Dr. Dong Yinpeng from Tsinghua University
The Background
In the past two years, with the emergence of artificial intelligence models such as ChatGPT and Sora, the attention of various sectors of society to AI has been increasing. Both the scientific research industry and the investment and financing industry are full of expectations for the application of artificial intelligence in future human society. The AI models in the field of medical and health care are highly regarded for their ability to solve clinical needs more quickly and accurately. However, due to the unique nature of the medical field, the security of AI models is crucial and directly affects their implementation in the healthcare industry.
Events Review
In response to the significant importance of model security in artificial intelligence products in the healthcare field, the National Thousand Ecosystem Theme Training Event invited Associate Professor Qiu Han from the Network Research Institute of Tsinghua University and Dr. Dong Yinpeng from the Department of Computer Science and Technology to share on the theme of "AI Big Model Security".
The theme of Professor Qiu's speech is "Research on Content Security of Large Language Models", which introduces the development history and basic principles of large language models, the output control of harmful content by large models, and the impact of large model illusions on vertical domain applications through practical GPT model cases in a simple and understandable manner. In his speech, Professor Qiu explained the limitations, biases, and reasons for unsafe content output of large models by analyzing the GPT training process. He believed that the data-driven nature would directly affect the security of large models. Meanwhile, Professor Qiu believes that the autoregressive reasoning mode of the big language model is related to model hallucinations, and the responses of different models under hallucinations or human interference will have different manifestations. Reducing the illusions of generative artificial intelligence and improving model accuracy are the main challenges for the widespread use of large models in vertical fields, and there is still a long way to go.
The topic shared by Dr. Dong is "Research on the Security and Trustworthiness of Diffusion Probability Models". With solid theoretical foundation, Dr. Dong focuses on introducing the basic concepts and classification of diffusion models in the field of image generation, the robustness and security judgment of diffusion models, the security processing of model generated images, and the application of diffusion models to improve the robustness of image generation and recognition. Dr. Dong combined his research experience with his speech to analyze practical application cases and mathematical methods of various diffusion models, deepening the audience's understanding of the security risks of generative artificial intelligence.
During the sharing process, two experts and the audience had a lively discussion on different methods of big language models, the integration of physical concepts and big models, the impact of medical data on big model hallucinations, future trends of big models, and how to set prompt words in model training.
The current artificial intelligence is impacting traditional industries, and the new paradigm has a huge impact on product positioning, product content, and product development. The application of large-scale models presents new opportunities and challenges for medical enterprises. I look forward to the sparks generated from this exchange to inspire entrepreneurs and help companies innovate and develop in the era of artificial intelligence!
About Guoqian
Guoqian Medical Technology (Suzhou) Co., Ltd. is jointly invested and established by Dr. Zhang Xu and seven other national major talent engineering experts who have successfully innovated and started businesses in the fields of biomedical and medical devices, as well as shareholders in other fields.
Guoqian Medical is committed to investing in medical device innovation technology industry, entrepreneurial incubation, and precise empowerment. Guoqian Venture Capital Management (Suzhou) Co., Ltd., a wholly-owned subsidiary, is a venture capital firm specializing in early-stage equity investment in medical devices. Adhering to the business philosophy of "achieving others' dreams", Guoqian Venture Capital invests and aggregates industries through its managed equity investment funds, while Guoqian Medical creates an industrial ecosystem through incubation and empowerment. The two business platforms are highly coordinated, creating a high-end medical equipment industry cluster and ecosystem in Suzhou High tech Zone.