Vision scientists have long pursued an understanding of how the visual system manages a deluge of information within its biological limitations. One computational strategy that the visual system exploits to mitigate its information processing bottleneck is ensemble coding–an information integration process that extracts summary statistics from complex visual scenes. Despite ample evidence showing that humans can quickly represent summary statistics, the computational mechanisms of ensemble coding remain largely unknown. In this thesis, I bridge ensemble coding with the theory of efficient coding, proposing that our visual system can further optimize its information processing through efficient ensemble coding. In Chapter 2, I formulate the efficient ensemble coding as a normative computational framework. With the core assumption that sensory encoding rapidly adapts to the short-term, dynamic ensemble statistics in the experiment, I demonstrate that the model accurately captures various aspects of ensemble coding behavior across multiple feature domains, including the overweighting of ensemble stimuli with feature values close to the decision boundary (inlying elements). This model provides a normative explanation for this intriguing phenomenon: it enhances ensemble coding accuracy within the system’s resource constraints. In Chapters 3 and 4, I present compelling evidence for this efficient ensemble coding model by validating its two key predictions. I show that the strength of efficient ensemble coding increases as the visual system progressively adapts to a dynamic, approximately Gaussian distribution of ensemble stimuli, resulting in an increasingly overweighting of inlying elements (Chapter 3). Conversely, observers shift to a more equal weighting scheme when their visual system adapts to a dynamic uniform stimulus distribution (Chapter 4). In Chapter 5, I further demonstrate that a misalignment between the ensemble and reference onset disrupts efficient ensemble coding, thereby impairing ensemble coding accuracy. I conclude with a discussion of the empirical and theoretical implications of this work in Chapter 6. Taken together, this thesis extends the boundaries of efficient coding theory and provides a general computational model for ensemble coding, offering valuable insights into its underlying computational processes.