Deep-level collaborations between theoretical computer science, mathematics, and statistics are indispensable to build theoretical foundations of data sciences, urgently driven by the ever-increasing scale of data and complexity of modern machine learning algorithms. However, culture and language barriers between different disciplines have been obstacles to make mutual-benefit collaborations, inspire new ideas, and obtain synergistic achievements. UA-TRIPODS aims to build a highly collegial research institute to bring three communities together collaborating on fundamental areas of theoretical data science, using “common communication languages” and working with “the same mindset”. The institute also strives to provide cross-disciplinary education and training experiences for next-generation data scientists. This talk will share our three-year experiences, highlighting our strategies, pitfalls, and achievements as well.