Identify Hierarchical Structures from Task-Based fMRI Data via Hybrid Spatiotemporal Neural Architecture Search Net
- 6.2k Downloads
The hierarchical organization of brain function has been an established concept in the neuroscience field. Recently, these hierarchical organizations have been extensively investigated in terms of how such hierarchical functional networks are constructed in the human brain via a variety of deep learning models. However, a key problem of how to determine the optimal neural architecture (NA), e.g., hyper parameters, of deep model has not been solved yet. To address this question, in this work, a novel Hybrid Spatiotemporal Neural Architecture Search Net (HS-NASNet) is proposed by jointly using Evolutionary Optimizer, Deep Belief Networks (DBN) and Deep LASSO to reasonably determine the NA, thus revealing the latent hierarchical spatiotemporal features based on the Human Connectome Project (HCP) 900 fMRI datasets. Briefly, this HS-NASNet can automatically search the global optimal NA of DBN given the search space, and then the optimized DBN can extract the weights between two adjacent layers of the optimal NA, which are then treated as the hierarchical temporal dictionaries for Deep LASSO to identify the corresponding hierarchical spatial maps. Our results demonstrate that the optimized deep model can achieve accurate fMRI signal reconstruction and identify spatiotemporal functional networks exhibiting multiscale properties that can be well characterized and interpreted based on current neuroscience knowledge.
KeywordsHierarchical organization Neural architecture search Evolutionary optimization Deep Belief Network Task-based fMRI LASSO
- 4.Real, E., Aggarwal, A., Huang, Y., Le, Q.V.: Regularized evolution for image classifier architecture search. arXiv preprint arXiv:1802.01548 (2018)
- 6.Cortes, C., Gonzalvo, X., Kuznetsov, V., Mohri, M., Yang, S.: AdaNet: adaptive structural learning of artificial neural networks. In: Proceedings of the 34th International Conference on Machine Learning-Volume 70, pp. 874–883. JMLR.org, August 2017Google Scholar