Digital_Twin_Data.zip

============================================================DIgital Twin JSON Configuration Documentation============================================================<br>This document describes the structure and contents of the provided .json configuration file. The file encodes model configuration, data configuration, trained parameters, and sample input/output.<br>This files results from the in-silicon simulation of the HORN model, applying the experimental constraints of the anabrid Model 1.<br>------------------------------------------------------------Top-Level Structure------------------------------------------------------------<br>The JSON file contains the following main keys:<br>- config_model -&gt; Model architecture and training configuration- config_data -&gt; Dataset information and preprocessing settings- parameters_model -&gt; Model weights, masks, and learned parameters- input -&gt; Input vector provided to the model (e.g., MNIST image)- label -&gt; Ground-truth target label- prediction -&gt; Model’s predicted label- activity -&gt; Model’s internal activity traces<br>------------------------------------------------------------config_model------------------------------------------------------------<br>Defines the neural model’s structure, constraints, and initialization.<br>- constraint_ulmann : true (enables structural constraint)- name : "horn" (model type/name)- params : dictionary of hyperparameters and structural settings<br>Parameters (params):- gamma : damping constant (0.01)- h : scaling factor (1.0)- omega : frequency parameter (0.224)- sigma : "lin" (linear scaling of sigma)- state_init : initial hidden state (fixed at 0)- n_hidden : number of hidden units (4)- readout : linear readout at time = -1- seed : random initialization seed (1)- n_params : number of learnable parameters (0)<br>Weight Matrices:- whh (hidden-to-hidden): - mask : "nodiag" (no self-connections) - constraints : clip values to [0.0, 9.0] - bias : false - pre_factor : -1<br>- wih (input-to-hidden): - constraints : clip values to [0.0, 9.0] - bias : false - pre_factor : -1<br>------------------------------------------------------------config_data------------------------------------------------------------<br>Specifies dataset and preprocessing options.<br>- name : "mnist"- params : - normalize : false (no input normalization) - seed : 1- variant : "base"- version : "untracked"<br>------------------------------------------------------------parameters_model------------------------------------------------------------<br>Contains learned parameters and masks.<br>Scalars:- h : [1.0]- omega : [0.224]- gamma : [0.01]- alpha : [0.04]<br>Weights:- i2h.weight : Input -&gt; Hidden connections (shape: 4x1)- h2h.mask_mat : Hidden -&gt; Hidden connectivity mask (4x4, no diagonals)- h2h.weight : Hidden -&gt; Hidden weights (4x4, masked)- h2o.weight : Hidden -&gt; Output connections (10x4)- h2o.bias : Output biases (10-dim)<br>------------------------------------------------------------input------------------------------------------------------------<br>Represents a flattened MNIST image (28x28 = 784 values).- Pixel intensities range from 0.0 (black) to 1.0 (white).- Example shows a digit image with partial grayscale pixel values.<br>------------------------------------------------------------label &amp; prediction------------------------------------------------------------<br>- label : Ground-truth class - prediction : Model output class<br>------------------------------------------------------------activity------------------------------------------------------------<br>This section (truncated in example) stores hidden unit activity traces during model execution.- Tracks how hidden states evolve over time.- Useful for analyzing model dynamics.<br>------------------------------------------------------------Summary------------------------------------------------------------<br>This JSON file stores both the model definition and a training snapshot including:<br>- Neural architecture (config_model)- Dataset setup (config_data)- Weights and masks (parameters_model)- Example inference on MNIST (input, label, prediction)<br>