SOTA's Seminarhttps://verse-astro.vercel.app/en-UShttps://verse-astro.vercel.app/posts/Weiyang%20Jin/https://verse-astro.vercel.app/posts/Weiyang%20Jin/Think about the pros and cons of native mllm and SD backone from the perspective of the current strongest closed-source model.Sat, 05 Apr 2025 00:00:00 GMT<!--  --> **Talker**: [Weiyang Jin](https://github.com/WayneJin0918) **Bio**: Weiyang Jin is now the intern at New York University advised by Prof. Saining Xie. His research interests are about MLLMs/VLMs and visual representation learning **Key note**: - Introdues the defination of native MLLMs and current works. - Analyze the possible architecture of the current strongest closed-source model. - Introducing insights from recent papers about revealing the model principle. - Discuss the pros and cons of both systems and possible future development directions. [Bilibili Link](https://b23.tv/sR9g9co)https://verse-astro.vercel.app/posts/Ning%20Gao/https://verse-astro.vercel.app/posts/Ning%20Gao/Analyze model applications of single and dual systems. Discuss the differences between work in contemporary periods such as Gemini-robotics and GR00T.Mon, 31 Mar 2025 00:00:00 GMT<!--  --> **Talker**: [Ning Gao](https://axi404.top/) **Bio**: Ning Gao is a 3rd year student from XJTU. He is conducting a scientific internship in the Embodied Intelligence Center of Shanghai Artificial Intelligence Laboratory (SHAILAB), and is engaged in research in the field of Embodied Intelligence Manipulation **Key note**: - Paper sharing and discussion on the dual-system model in Gemini-Robotics, HiRobot, and other relevant frameworks. - Provide a personal definition of the dual-system framework in embodied AI. - Comparative analysis of end-to-end, dual-system, and prompt-based models. Highlight the advantages and limitations of each approach in embodied tasks. - Define the function in VLA and VLM for embodied system - Discussion future trend in embodied AIhttps://verse-astro.vercel.app/posts/Huanran%20Chen/https://verse-astro.vercel.app/posts/Huanran%20Chen/Explore the robustness of autoregressive models and diffusion models, and think about the analysis of the upper and lower boundaries of robustness and its applications.Sat, 29 Mar 2025 00:00:00 GMT<!--  --> **Talker**: [Huanran Chen](https://huanranchen.github.io/) **Bio**: Huanran Chen is a PhD student from TSAIL (Fall 2025), advised by Prof. Jun Zhu, and closely collaborate with Prof. [Yinpeng Dong]. He has a keen interest in the physics of machine learning. His unattainable yet motivating dream is to elevate AI to the realm of science, making every phenomenon explainable and predictable. **Key note**: - What is worst-case robustness? - Why worst-case robustness? - Upper bounding worst-case robustness - Lower bounding worst-case robustness (knapsack problem) - Specific application to text domain - Theoretical insights (e.g., diffusion>MaskGen>ARM) - Applications [slides](https://drive.google.com/file/d/19QrJOxkKdknQkQiKYeoZmYVP5C4OGCKO/view)