Feature-space selection with banded ridge regression

Tour, Tom Dupré la; Eickenberg, Michael; Núñez-Elizalde, Anwar O.; Gallant, Jack L.

doi:10.1016/j.neuroimage.2022.119728

Cited by 48 publications

(75 citation statements)

References 84 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, because different features may have substantially different predictive powers, and may correlate with one another to different extents, single regularization parameter may be insufficient to allow for accurate TRF estimation of all features in a model (Nunez-Elizalde et al, 2019). To account for this issue, banded ridge regression methods have recently been introduced, which apply individualized ridge parameters to different features in a model (Nunez-Elizalde et al, 2019; Dupré la Tour et al, 2022). This method was not used in the present work, as we chose to build our analysis pipelines using publicly available tools (i.e., mTRF toolbox), which did not have this functionality implemented, yet (but see Crosse et al, 2021 for the description of forthcoming banded ridge regression functionality).…”

Section: Discussionmentioning

confidence: 99%

The effects of data quantity on performance of temporal response function analyses of natural speech processing

Mesík

Wojtczak

2022

Preprint

View full text Add to dashboard Cite

In recent years, temporal response function (TRF) analyses of non-invasive recordings of neural activity evoked by continuous naturalistic stimuli have become increasingly popular for characterizing response properties within the auditory hierarchy. However, despite this rise in TRF usage, relatively few educational resources for these tools exist. Here we use a dual-talker continuous speech paradigm to demonstrate how a key parameter of experimental design, the quantity of acquired data, influences TRF analyses fit to either individual data (subject-specific analyses), or group data (generic analyses). We show that although model performance monotonically increases with data quantity, the amount of data required to achieve significant prediction accuracies can vary substantially based on whether the fitted model contains densely (e.g., acoustic envelope) or sparsely (e.g., lexical surprisal) spaced features, especially when the goal of the analyses is to capture the aspect of neural responses that co-vary with the amplitude of the modelled features. Moreover, we demonstrate that generic models can exhibit high performance on small amounts of test data (4-8 min), as long as they are trained on a sufficiently large data set. As such, they may be particularly useful for clinical and multi-task study designs. Finally, we show that the regularization procedure used in fitting TRF models can interact with the quantity of data used to fit the models, with larger training quantities resulting in systematically larger TRF amplitudes. Together, demonstrations in this work should aid the learning process of new users of TRF analyses, and in combination with other tools, such as piloting and power analyses, may serve as a detailed reference for choosing acquisition duration in future studies.

show abstract

Section: Discussionmentioning

confidence: 99%

The effects of data quantity on performance of temporal response function analyses of natural speech processing

Mesík

Wojtczak

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…For this purpose, we constructed whole-brain voxel-wise encoding models for the following four settings (see Figure 2 bottom and Appendix A for implementation details): We first built linear models to predict voxel activity from the following three latent representations of the LDM independently: z, c , and z c . Although z c and z produce different images, they result in similar prediction maps on the cortex (see 4.2.1). Therefore, we incorporated them into a single model, and further examined how they differ by mapping the unique variance explained by each feature onto cortex [23]. To control the balance between the appearance of the original image and the semantic fidelity of the conditional text, we varied the level of noise added to z .…”

Section: Methodsmentioning

confidence: 99%

High-resolution image reconstruction with latent diffusion models from human brain activity

Takagi

Nishimoto

2022

Preprint

View full text Add to dashboard Cite

Reconstructing visual experiences from human brain activity offers a unique way to understand how the brain represents the world, and to interpret the connection between computer vision models and our visual system. While deep generative models have recently been employed for this task, reconstructing realistic images with high semantic fidelity is still a challenging problem. Here, we propose a new method based on a diffusion model (DM) to reconstruct images from human brain activity obtained via functional magnetic resonance imaging (fMRI). More specifically, we rely on a latent diffusion model (LDM) termed Stable Diffusion. This model reduces the computational cost of DMs, while preserving their high generative performance. We also characterize the inner mechanisms of the LDM by studying how its different components (such as the latent vector of image Z, conditioning inputs C, and different elements of the denoising U-Net) relate to distinct brain functions. We show that our proposed method can reconstruct high-resolution images with high fidelity in straightforward fashion, without the need for any additional training and fine-tuning of complex deep-learning models. We also provide a quantitative interpretation of different LDM components from a neuroscientific perspective. Overall, our study proposes a promising method for reconstructing images from human brain activity, and provides a new framework for understanding DMs.

show abstract

“…Fitting a joint model with ridge regression allows considering the complementarity of different feature spaces but subjects all models (feature sets) to a unique regularization. As the optimal regularization required when fitting each individual feature space may differ (since it depends, among others, on factors such as number of features and features covariances) (27), fitting a joint model with one regularization parameter may be suboptimal and can be extended to banded ridge regression. In banded ridge regression, separate regularization per parameters for each feature space are optimized, which ultimately improves model performance by reducing spurious correlations and ignoring non-predictive feature spaces (27, 28).…”

Section: Methodsmentioning

confidence: 99%

“…As the optimal regularization required when fitting each individual feature space may differ (since it depends, among others, on factors such as number of features and features covariances) (27), fitting a joint model with one regularization parameter may be suboptimal and can be extended to banded ridge regression. In banded ridge regression, separate regularization per parameters for each feature space are optimized, which ultimately improves model performance by reducing spurious correlations and ignoring non-predictive feature spaces (27, 28). In the present work we used banded ridge regression to fit the three encoding models and performed a decomposition of the variance explained by each of the models following established procedures (27).…”

Section: Methodsmentioning

confidence: 99%

Voxelwise encoding models of body stimuli reveal a representational gradient from low-level visual features to postural features in extrastriate body area

Marrazzo

Martino

Lage‐Castellanos

et al. 2022

Preprint

View full text Add to dashboard Cite

Previous research has focused on the role of the extrastriate body area (EBA) in categoryspecific body representation, but the specific features that are represented in this area are not well understood. This study used ultra-high field fMRI and banded ridge regression to investigate the coding of body images by comparing the performance of three encoding models in predicting brain activity in ventral visual cortex and specifically the EBA. Our results suggest that EBA represents body stimuli based on a combination of low-level visual features and postural features.

show abstract

Feature-space selection with banded ridge regression

Cited by 48 publications

References 84 publications

The effects of data quantity on performance of temporal response function analyses of natural speech processing

The effects of data quantity on performance of temporal response function analyses of natural speech processing

High-resolution image reconstruction with latent diffusion models from human brain activity

Voxelwise encoding models of body stimuli reveal a representational gradient from low-level visual features to postural features in extrastriate body area

Contact Info

Product

Resources

About