/ValidationTools

Development of statistical tests and tools for comparison of neural activity data with the aim of validating simulations.

Primary LanguageJupyter Notebook

Validation Tools

Development of statistical tests and tools for comparison of neural activity data with the aim of validating neural simulations.

Based on Elephant and neo.
Simulation time 60 x 60 x 5 + 120 s
STDP on true
GENERATE_FROM_EXTERNAL_DATA false
USE_EXTERNAL_STIMULUS false
STATIC_WEIGHTS false

ToDo

  • snakemake
  • Make ValidationBook presentation-ready and integrate into collab
  • Separate Data (Results & Input) into non git controlled dir (to Sciebo)
  • Draw raw distributions for visual comparison
  • analytic T-dependent spectral radius
  • Apply Methods to PSTH Distribution Comparison
    • s. Paper von Grün, Abeles
  • Explore Parameters
    • Blaustein Cluster: Slurm queuing, rsync, requirements (-> Michael)
      • Performance estimate: 2.6MB/run, 11.2s/run (based on 32 runs)
      • Git checkout on cluster?
      • elephant, neo installation?
    • estimate minimum data size to detect features with a certain resolution
      • assemblysize; assemblycorr; assemblycorr with backgroundcorr
  • How to evaluate the significance of high-dim angles?
    • Forward problem to mathematician!
  • Use Depth Functions for multivariate significance of results
  • GED (generalized eigenvalue decompositon): de Cheveigné and Parra "Joint decorrelation ..."
  • Theoretical distribution of correlation matrix eigenvalues
  • Visualisation in -N/ K-space? -> DataHigh (for Win?)
  • PCA directly on binned spiketrains/ inst. firing rate (t)
    • Reduction of spiking variability by trial averaging/ temp. smoothing
    • Activity trace in <=3D k-space
    • luczak harris paper, rate vector, GPFA (-> Junji, Carlos)
  • Implement oscillations analysis (test with brunel)
    • spiketrainwise autocorrelation function
    • population autocorrelation
    • relation to fourier analysis
  • Write function annotations
  • Maybe apply method to Networks with ASSET method (->carlos)
  • Integrate CuBIC (in elephant)
  • Adapt code for collab packages

ORGANIZATIONAL CONSTRUCT

  • METHODS

    • Distribution Comparisons
    • Correlation Matrix
    • Oscillations
    • (Graphs)

  • IMPLEMENTATION

    • Validation Toolbox
      • dist
      • matrix
      • test_data
    • Jupyter Interface
    • => Collab
    • => Elephant

  • WORKFLOW

    • Determine the focus of usage

      • Neuroscience
      • Deep Learning
      • Simulation
      • Experiment Data
      • ..?

    • Test against

      • Neuron model
      • Spike Loss
      • Time steps
      • Numerical Integration
      • Network Model
      • Reference (given by different architecture/hardware)
      • ...

    • Creating references

      • Brainscales models Comparison
      • Universal reference exact NEST?
      • Create reference by regarding distribution of test results

  • USE CASES

    • NEST vs SpiNNaker
    • Brunel network with different solver/ time step/ precision
      • measures to detect state
    • SpiNNaker Izhekevich exact vs inexact