Inverted encoding models estimate sensible channel responses for sensible models

Abstract

for helpful discussions and comments on the manuscript. Abstract (233 words)In a commentary published in eNeuro, Gardner & Liu (2019) discuss the role of model specification in interpreting the output of complex models of neural data. As a case study, they suggest that one variant of such analyses, the inverted encoding model (IEM) analysis framework, should not be used to assay properties of "stimulus representations" because the ability to apply linear transformations at various stages of the analysis procedure renders results 'arbitrary'. As we discuss, the specification of all models is arbitrary to the extent that an experimenter makes choices based on current knowledge of the model system. However, the results derived from any given model, such as the reconstructed channel response profiles obtained from an IEM analysis, are uniquely defined and are arbitrary only in the sense that changes in the model can predictably change results. Moreover, with knowledge of the model used for IEM analyses, the results remain informative as comparisons between reconstructed channel response profiles across task conditions using a fixed encoding model -the most common use of the IEM technique -can generally capture changes in population-level representation magnitude across linear transformations. Thus, changes in the magnitude of the response profiles across conditions are preserved, even across unprincipled linear transforms. IEM-based channel response profiles should therefore not be considered arbitrary when the model is clearly specified and guided by our best understanding of neural population representations in the brain regions being analyzed.Intuitions derived from this case study are important to consider when interpreting results from all model-based analyses, which are similarly contingent upon the specification of the models used.

Neural dynamics of the attentional blink revealed by encoding orientation selectivity during rapid visual presentation

“…This analysis verified that each RSVP item could be encoded independently. We aligned all RSVP items across trials (N = 12,000; 600 trials by 20 items) and used a fixed encoding model for training and testing 67,68 (Figs. 6-8).…”

Neural dynamics of the attentional blink revealed by encoding orientation selectivity during rapid visual presentation

“…This analysis verified that each RSVP item could be encoded independently. We aligned all RSVP items across trials (N = 12,000; 600 trials by 20 items) and used a fixed encoding model for training and testing (Figures 6-8) 64, 65 . This meant we trained and tested all encoding models across all items (both targets and distractors) regardless of trial type 12, 13 .…”

Behavior-dependent directional tuning in the human visual-navigation network

“…Future studies employing more dynamic model tests such as inverted encoding modeling (proof of principle in Supplementary Fig. 7) could address these questions directly by reconstructing the modeled behavioral features more dynamically (Gardner and Liu, 2019;Kriegeskorte and Douglas, 2019;Liu et al, 2018;Naselaris et al, 2011;Sprague et al, 2018Sprague et al, , 2019. In addition, high temporal resolution measures such as magnetoencephalography (MEG) or intracranial recordings in combination with VR-tasks (Kunz et al, 2019) could monitor trial by trial changes in tuning as a function of behavioral performance.…”