The following is a summary of “Ensemble averaging: What can we learn from skewed feature distributions?”, published in the January 2023 issue of Ophthalmology by Lakovlev, et al.
According to several research, the average feature of a set of items may be precisely estimated by observers. However, there was disagreement over how the visual system uses the data from each distinct item. Some models proposed arithmetically averaging and sampling some or all elements. In another approach, median parts are given more weight than outliers, which is known as “robust averaging.” Teng et al. (2021), who investigated motion direction averaging in skewed feature distributions and discovered systematic biases toward their modes, recently proposed one variant of a robust averaging model. They considered these biases to provide support for robust averaging and proposed a probabilistic weighting methodology based on virtual loss function reduction.
For a study, researchers replicated systematic skew-related biases in four trials in an additional feature domain called orientation averaging. Importantly, they demonstrated that the form of the entire feature distribution strongly defined the bias magnitude and not just the mean or mode locations. They examined a model that, in a physiologically reasonable manner, considered such distribution-dependent biases and resilient averaging.
The approach was predicated on well-established methods of spatial pooling and population encoding of regional properties by neurons with expansive receptive fields. Furthermore, ithe pooled population response model could be viewed as a neural implementation of the computational algorithms of information sampling and robust averaging in ensemble perception because both the loss functions model and the population coding model with a winner-take-all decoding rule accurately predicted the observed patterns.