Parameter estimation using residual neural network

---

1. In this work, we represent a practical approach capable of estimating any stochastic process, whether its likelihood function comes from an unknown distribution or the model relys on simulation-based approaches for data generation.

The architecture used in this work is as follows:

---

2. DataSets

We utilized our model on four specific models. Data sets for the Toy Examples are available here, or you can use MVN and MR. (Stefan et al., 2020)

Data sets provided for Drift-Diffusion and Collapsing Drift-Diffusion were based on a data-simulation mechanism, and the codes are available here and here, respectively.

---

3. Required packages

For training:

1. tensorflow 2.x
2. pandas & numpy & matplotlib
3. sklearn

For data generation:

1. numpy
2. scipy
3. multiprocessing
4. functools

number 3 and 4 is needed when using multi threaded

---

4. Pre-trained models

To run inference from the saved models, click on [ MVN , MR , DDM , C-DDM ]