Feature Selection Methods for Zero-Shot Learning of Neural Activity

Abstract

Dimensionality poses a serious challenge when making predictions from human neuroimaging data. Across imaging modalities, large pools of potential neural features (e.g., responses from particular voxels, electrodes, and temporal windows) have to be related to typically limited sets of stimuli and samples. In recent years, zero-shot prediction models have been introduced for mapping between neural signals and semantic attributes, which allows for classification of stimulus classes not explicitly included in the training set. While choices about feature selection can have a substantial impact when closed-set accuracy, open-set robustness, and runtime are competing design objectives, no systematic study of feature selection for these models has been reported. Instead, a relatively straightforward feature stability approach has been adopted and successfully applied across models and imaging modalities. To characterize the tradeoffs in feature selection for zero-shot learning, we compared correlation-based stability to several other feature selection techniques on comparable data sets from two distinct imaging modalities: functional Magnetic Resonance Imaging and Electrocorticography. While most of the feature selection methods resulted in similar zero-shot prediction accuracies and spatial/spectral patterns of selected features, there was one exception; A novel feature/attribute correlation approach was able to achieve those accuracies with far fewer features, suggesting the potential for simpler prediction models that yield high zero-shot classification accuracy.

Graph-Based Deep Learning for Medical Diagnosis and Analysis: Past, Present and Future

“…Another interesting variant of transfer learning is zero-shot learning (ZSL), which aims to predict the correct class without being exposed to any instances belonging to that class in the training dataset. Although zero-shot learning is flourishing in the field of computer vision [ 218 ], it is seldom used for biomedical signal analysis, though zero-shot learning has recently been used to recognize unknown EEG signals [ 219 ]. Recently, GCNs have shown a lot of promise for zero-shot learning.…”

Modeling Brain Representations of Words' Concreteness in Context Using GPT‐2 and Human Ratings

“…We adopted the methodology of Mitchell et al. (2008), which is a straightforward choice in encoding studies involving word vectors (Anderson, Zinszer, & Raizada, 2016; Caceres et al., 2017). This procedure was carried out separately for each subject separately and for each train‐test split (see Section 5.2).…”

“…Dedicated methods have been devised in the literature for encoding studies like ours, where a mapping function is learnt between a vectorial model and brain data (Mitchell et al, 2008;Pereira et al, 2018). We adopted the methodology of Mitchell et al (2008), which is a straightforward choice in encoding studies involving word vectors (Anderson, Zinszer, & Raizada, 2016;Caceres et al, 2017). This procedure was carried out separately for each subject separately and for each train-test split (see Section 5.2).…”

Graph-Based Deep Learning for Medical Diagnosis and Analysis: Past, Present and Future

“…Another interesting variant of transfer learning is zero-shot learning (ZSL) which aims to predict the correct class without being exposed to any instances belonging to that class in the training dataset. Although zero-shot learning is flourishing in the field of computer vision [264], it is seldomly used for biomedical signal analysis, though zero-shot learning has recently been used to recognize unknown EEG signals [265]. Recently, graph convolutional networks have shown a lot of promise for zero-shot learning.…”