Newsroom
国产短视频CSE graduate student wins best doctoral consortium award at an International Conference
国产短视频CSE PhD candidate Saandeep Lakshminarayan won best doctoral consortium award at the IEEE International Conference 2023 on Automatic Face and Gesture Recognition Doctoral Consortium for his application: "Towards Explainable Affective Computing using Efficient and Tractable Representation Learning."
The IEEE International Conference on Automatic Face and Gesture Recognition is a premier forum for researchers, educators, industry professionals, and PhD students to network, present, and discuss the latest advancements in the field of affective computing.
Under the guidance of Shaun Canavan, Assistant Professor in CSE, Saandeep has been researching how to use machine learning to create explainable facial expression recognition technology. The goal being to understand why the machine does what it does. This is necessary because AI, as a machine of pure correlation, often takes unexpected, and sometimes detrimental paths to its end goals. When it comes to facial expression recognition technology, with all its ethical implications, a machine that accidentally takes shortcuts could be potentially dangerous or unreliable.
Saandeep Lakshminarayan鈥檚 winning paper showcased his current progress to mitigating these risks using Normalizing Flows and Supervised Contrastive Learning as a development path. 鈥淭he former helps capture the underlying nature of the dataset (probability distribution) and the latter builds meaningful low-dimensional representation of the original data, exhibiting semantic information,鈥� explained Saandeep. 鈥淏y efficient fusion of the underlying mathematical equations of the two, we can observe why our deep learning model predicts an expression based on the low-dimensional plot.鈥�