MAGIST-Algorithm

Multi-Agent Generally Intelligent Simultaneous Training Algorithm for Project Zeta!

Official Documentation

We now have official documentation available here!

Working Principle

Data

The data is the most crucial element as the entire intelligence works around it. That is why the AI needs to process it and provide reasonable assumptions. There is another condition, however: the algorithm, in its finished state, MUST be strictly Python code. This means no pre-trained models or presets. It must find its data and process it unsupervised. Although this architecture seems hard-coded in the present state, as more functionality is added, it will be more intelligent and decisive. Here, the data will assume the following structure:

Object -> Common Associated Verbs, Synonyms, Events, Timestamps of usage, Nearby Objects, etc.

This is the "Who, What, When, Where, Why, How" of the data. This data can later be filtered and called upon when inferences are needed. To get here, however, the data must first be extracted from a single image. Since no pre-trained models are allowed, here is the process to follow:

1. K-Means Clustering(find key objects in the image) -> Discriminator for integrity check(see if clustering was performed well)
2. Reverse Image Search and Google Scraping(find the label of image) -> Data Downloader(find dataset from large datasets)
3. Transfer Learn Model -> Object Detector
4. Get a summary from internet sources like Wikipedia; location from LiDAR(future); user from facial detection(future); etc.
5. Store all the data in the NeuralDB

Natural Language Processing

Another key stage of this AGI(Artificial General Intelligence) is human interaction and understanding. MAGIST will constantly listen to conversations and make intelligent decisions. Here is the target process:

1. Record audio data and transcribe it(this is ***the only*** place where a pre-trained model(the transcriber) is used since learning a human language fully unsupervised is incredibly arduous.
2. Use a custom positional embedding with a Self-Attention head to find keywords.
3. Search these terms in the NeuralDB for possible entries.
4. Search unknown terms online and store definitions for future reference. (future)
5. Extract key terms from matching entries. (future)
6. Insert those key terms into a text transformer trained on the collected NeuralDB data to generate a prediction. (future)
7. Utter the prediction. (future)

Usage

This project is still under development. Please contact me at [krishna.shah@deepshift.dev]() if you want immediate access and/or support to MAGIST. Once the algorithm is in a stable state, I will release a Python Package on PyPI and Github for access. There will also be documentation with more instructions.

Installation

This project has many dependencies. Most can be installed using pip. Some require OS-level package managers. This is going to work best in Linux-based systems.

These instructions are for Linux-based systems. In particular for Ubuntu 20.04 LTS based operating systems. Other systems may have errors that will require debugging.

Linux (Ubuntu-based Systems)

First, install Python 3 and pip:

sudo apt install python3 python3-dev python3-pip

Next, we need to install Firefox and its corresponding geckodriver for headless Selenium searches:

sudo apt install firefox

Note: If you get an error regarding the geckodriver, you can install it manually by following the instructions here.

Install System Packages

Next, we need to install the system packages that MAGIST uses.

sudo apt install python3-pyaudio
sudo apt install libasound-dev

MongoDB

Next, we need to install MongoDB. This is a database that MAGIST uses to store its data. Please go to the MongoDB Website and follow the instructions to install it. We have more instructions in the documentation.

Next, create a Python environment. There are 2 ways to do this: Anaconda or VEnv.

Anaconda

First, install Anaconda from https://www.anaconda.com/.

Make the Anaconda environment:

conda create --name myenv

Activate the environment in your current console. Note: You will have to do this every time you want to run MAGIST.

conda activate myenv

Get the latest MAGIST Wheel from our releases page!

Install MAGIST:

pip3 install MAGIST-Algorithm-x.x.x-py3-none-any.whl

VEnv

Make the environment in a designated location.

python3 -m venv /path/to/new/virtual/environment

To activate it, you must travel to that path/bin/ and then run:

source activate

Get the latest MAGIST Wheel from our releases page!

Install MAGIST:

pip3 install MAGIST-Algorithm-x.x.x-py3-none-any.whl

Congratulations! You are all set up to script and use MAGIST

Contributing

Your contribution, monetary or programmatically, is crucial for the rapid development of the algorithm and its training. Please consider contributing. Even minute changes to our README will be greatly appreciated.

Project Zeta

We are building a fully biomimetic robot dog to implement MAGIST into. This will serve as the gateway between MAGIST and the physical world.

Disclaimer

Artificial Intelligence is a powerful field meant only for research and study, and it should be kept that way. The unethical use of MAGIST can have severe repercussions for society and the perpetrator. DeepShift Labs and any programs it develops are strictly for research purposes. Hence, all of our products, MAGIST included, are to be used strictly for research purposes. Misuse of this program can lead to heavy fines and prosecution. Furthermore, to retitle, rebrand, or redistribute without explicitly crediting DeepShift Labs is illegal.