Target Inference

Target inference focuses on accurately predicting desired outcomes or states from available data, encompassing diverse applications from robotic control to personalized medicine. Current research emphasizes improving the efficiency and robustness of inference, particularly addressing challenges like handling unexpected environmental changes (e.g., using environment-centric active inference) and mitigating privacy risks in collaborative inference (e.g., through privacy-oriented pruning). These advancements are crucial for enhancing the reliability and security of machine learning systems across various domains, ranging from dialogue systems leveraging transformer encoders to resource-constrained edge devices.