---
license: openrail
---

# Monkey V4 Data Driven + Attention Readout Model Card

Neural Encoding model for Macaque V4. The model is a combination of a data driven core and an attention readout layer.

<p align="center"><img src="./assets/schematic.png" width="100%" alt="Data Driven V4 Schematic" /></p>

## Model Details
### Model Description

This model is a combination of a data driven core and an attention readout layer.
The data driven core is a convolutional neural network and the attention readout layer is a multihead attention layer with each head trained to predict the firing rates of a neuron in Macaque V4.

### Model Sources
For research purposes, we recommend our `nnvision` Github repository (https://github.com/sinzlab/nnvision), which contains the code for the model defintions and training.

- Repository: https://github.com/sinzlab/nnvision
- Paper: https://www.biorxiv.org/content/10.1101/2023.05.18.541176v1

### Intended Use

The model is intended for research purposes only.

### Model Use

The model can be used to predict the firing rates of neurons in Macaque V4 given an image.

#### nnvision
The model can be used in Python with the `nnvision` package.
```python
import torch
from nnvision.models.trained_models.v4_data_driven import v4_multihead_attention_ensemble_model

input_image = torch.rand(1, 100, 100)
firing_rate = v4_multihead_attention_ensemble_model(input_image, data_key="all_sessions")
```

### energy-guided diffusion
The model can be used in Python with the `energy-guided-diffusion` package.
```python
from egg.models import models
model = models['data_driven']['train']
```