Dimensionality reduction with normalizing flows

Despite the large number of active neurons in the cortex, for various brain regions, the activity of neural populations is expected to live on a low-dimensional manifold [1]. Among the most common tools to estimate the mapping to this manifold, along with its dimension, are many variants of principal component analysis [2]. Despite their apparent success, these procedures have the disadvantage that they assume only linear correlations and that their performance, when used as a generative model, is poor.To be able to fully learn the statistics of neural activity and to generate artificial samples, we make use of normalizing flows (NFs) [3, 4, 5]. These neural networks learn a dimension-preserving estimator of the data probability distribution. They are outstanding in comparison to generative adversarial networks (GANs) and variational autoencoders (VAEs) for their simplicity ‒ only one invertible network is learned ‒ and for their exact estimation of the likelihood due to tractable Jacobians at each building block.We aim to modify NFs such that they can discriminate relevant (in manifold) from noise (out of manifold) dimensions. To this end, we penalize the participation of each single latent variable in the reconstruction of the data through the inverse mapping (following a different reasoning than [6]). We can thus not only give an estimate of the dimensionality of the activity sub-space but also describe the underlying manifold without the need to discard any information.We prove the validity of our modification on controlled data sets of different complexity. We emphasize, in particular, differences between affine and additive coupling layers in normalizing flows [7], and show that the former lead to pathologies when the data topology is non-trivial, or when the data set is composed of classes with different volumes. We further illustrate the power of our modified NFs by reconstructing data using only a few dimensions.We finally apply this technique to identify manifolds in EEG recordings from a dataset showing high gamma activity (described in [8]), obtained from 128 electrodes during four different movement tasks.AcknowledgementsThis project is funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - 368482240/GRK2416; and by the German Federal Ministry for Education and Research (BMBF Grant 01IS19077A to Jülich).References [1] Gao, P., Trautmann, E., Yu, B., Santhanam, G., Ryu, S., Shenoy, K., & Ganguli, S. (2017). A theory of multineuronal dimensionality, dynamics and measurement. BioRxiv, 214262., 10.1101/214262 [2] Gallego, J. A., Perich, M. G., Miller, L. E., & Solla, S. A. (2017). Neural manifolds for the control of movement. Neuron, 94(5), 978-984., 10.1016/j.neuron.2017.05.025 [3] Dinh, L., Krueger, D., & Bengio, Y. (2014). Nice: Non-linear independent components estimation. arXiv preprint arXiv:1410.8516., 10.48550/arXiv.1410.8516 [4] Dinh, L., Sohl-Dickstein, J., & Bengio, S. (2016). Density estimation using real nvp. arXiv preprint arXiv:1605.08803., 10.48550/arXiv.1605.08803 [5] Kingma, D. P., & Dhariwal, P. (2018). Glow: Generative flow with invertible 1x1 convolutions. Advances in neural information processing systems, 31. [6] Cunningham, E., Cobb, A., & Jha, S. (2022). Principal manifold flows. arXiv preprint arXiv:2202.07037., 10.48550/arXiv.2202.07037 [7] Behrmann, J., Vicol, P., Wang, K. C., Grosse, R., & Jacobsen, J. H. (2021). Understanding and mitigating exploding inverses in invertible neural networks. In International Conference on Artificial Intelligence and Statistics (pp. 1792-1800). PMLR. [8] Schirrmeister, R. T., Springenberg, J. T., Fiederer, L. D. J., Glasstetter, M., Eggensperger, K., Tangermann, M., ... & Ball, T. (2017). Deep learning with convolutional neural networks for EEG decoding and visualization. Human brain mapping, 38(11), 5391-5420., 10.1002/hbm.23730

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.