Interpretable Nonlinear Dynamic Modeling of Neural Trajectories

Authors: Yuan Zhao, Il Memming Park

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show that our model can recover qualitative features of the phase portrait such as attractors, slow points, and bifurcations, while also producing reliable long-term future predictions in a variety of dynamical models and in real neural data. We apply the proposed method to a variety of low-dimensional neural models in theoretical neuroscience. Table 1: Model errors
Researcher Affiliation Academia Yuan Zhao and Il Memming Park Department of Neurobiology and Behavior Department of Applied Mathematics and Statistics Institute for Advanced Computational Science Stony Brook University, NY 11794 {yuan.zhao, memming.park}@stonybrook.edu
Pseudocode No No pseudocode or algorithm blocks are present in the paper.
Open Source Code No The paper mentions using TensorFlow [14] for implementation but does not provide access to its own source code for the methodology described.
Open Datasets No To test the model on data obtained from cortex, we use a set of trajectories obtained from the variational Gaussian latent process (vLGP) model [26].
Dataset Splits No We use 19 trajectories for training and the last one for testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory) used for running the experiments.
Software Dependencies No The paper mentions 'TensorFlow [14]' but does not provide specific version numbers for TensorFlow or any other software dependencies.
Experiment Setup Yes We estimate the parameters {Wg, WB, τ, c, σ} by minimizing the loss function through gradient descent (Adam [13]) implemented within Tensor Flow [14]. We initialize the matrices Wg and WB by truncated standard normal distribution, the centers {ci} by the centroids of the K-means clustering on the training set, and the kernel width σ by the average euclidean distance between the centers. The model with 10 basis functions learned the dynamics from 90 training trajectories (30 per coherence c = 0, 0.5, 0.5). We train the model with 50 basis functions on 100 simulated trajectories... The duration is 200 and the time step is 0.1.