Professor Chao\u2019s research to date has engaged deeply with this challenge, tackling it from different angles. To address the unique challenges of data in privacy- and security-critical scenarios, he has developed algorithmic and strategic improvements to federated learning, a training approach which utilizes decentralized data sources while protecting raw data. He co-led the first systematic study of 3D perception generalizability in AV systems across diverse environments, and co-developed the first visual foundation model dedicated to biodiversity, in pursuit of enhancing biological research by making the overwhelmingly varied traits of organisms computable from images.\u00a0<\/span><\/p>\n Chao also worked to shore up under-resourced and imbalanced training data for deep learning models via algorithms which exploit multiple views of data, among other adaptive approaches. Each project has been grounded in the understanding that technology will ultimately be more accurate and effective if it can embrace the imperfect, dynamic\u00a0 nature of the data that constitutes \u201creal life.\u201d<\/span><\/p>\n![\"\"](\"\/eng\/files\/2026\/01\/ece.faculty.26.chao_.png\")
Outside the tidy binaries of digital computation, the world is a messy place. Few things in life are clearly observable and predictable, whether you\u2019re looking at individual human behavior, traffic, organisms, or the weather. This is a truth that brand-new BU ECE Assistant Professor Wei-Lun (Harry) Chao not only recognizes, but has centered in his research: real-world applications of AI\/ML, especially those with significant safety implications, have to take that inherent messiness and imperfection into account. Large \u201chigh-quality\u201d datasets have unavoidable limitations; they are laborious and costly to collect and annotate due to the nature of the data they collect. In order to develop generalizable AI models, robust machine learning technologies must be developed to compensate for these limitations.<\/span><\/p>\n