Announcement

Collapse
No announcement yet.

[FREE PLUGIN] Integrate ChaosAI (My deep learning framework) in Unreal Engine.

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

    [FREE PLUGIN] Integrate ChaosAI (My deep learning framework) in Unreal Engine.

    Update 21-08-2020:
    This week I have not been able to dedicate part of my free time, soon I will have more time.

    Update 10-08-2020:
    I have already been able to train a network based on D3QN thanks to which an actor learns the ability to walk, a locomotion system.

    Update 07-08-2020:
    I have started to develop the demo of the deep reinforcement learning system of a locomotion system.

    Update 06-08-2020:
    I have already refactored the ChaosAI source code.
    I have integrated it into a Unreal Engine c ++ plugin and it compiles perfectly.

    [W.I.P]
    I am going to create a demo to demonstrate the power of ChaosAI, the first demo is going to be about a locomotion system based on deep reinforcement learning. I think it will be interesting.

    [Original Post]
    Hello friend of the Unreal Engine community, my name is Domy, I am going to tell you about the new project I started a few days ago that I think can be very interesting for all game developers, students, scientists and professionals who use Unreal Engine.

    I am the creator of ChaosAI, a Deep Learning framework developed in c ++ with the possibility of using CUDA if you want to use gpu, currently working to add OpenCL 2.0 & 3.0 for AMD gpu.

    The objective is to create a FREE PLUGIN with all the power of ChaosAI within Unreal Engine and convert ChaosAI into free software for the whole community to improve it.

    I know there are binding to use frameworks like TensorFlow with python but what I am going to develop is different, it is to integrate in the form of a ChaosAI plugin fully optimized for Unreal Engine and that anyone can use it, from AI scientists creating complex parameters and models to developers of games that have no knowledge of AI.

    Everyone can use Blueprint to create their networks, parameters and everything necessary.

    An interesting thing is to use pre-trained models, for example for different types of animals, enemies, vehicles.

    Being fully optimized in c ++ you should not lose performance in games, always the part that spends the most resources is the training part but it will not be like this in deep reinforcement training since one of the modes is continuous training, for example In a racing game, opponents can learn based on their driving style and the learned model is saved.

    With ChaosAI you can get the full potential of Deep Learning, for example deep reinforcement learning, below I am going to summarize the some beautiful things that with ChaosAI you can do.

    Give life to the NPCs, scenarios, thanks to the DQN and many of its variants that ChaosAI has, for example D3QN.

    Using NLP with NPCs thanks to the use of recurring networks (LSTM, GRU, BiLSTM) than ChaosAI and thanks to this the dialogues of your games can be more dynamic, also if we use D3QN we can punish or re-host the NPCs based on their way of talk .

    Thanks to the GAN, AE, VAE networks we will be able to create textures, scenarios, and music when users train their own networks of this type or use pre-trained models.

    And many more features, everything you can imagine thanks to the power of AI.

    Next I am going to expose the characteristics of ChaosAI.
    c ++
    CUDA for Nvidia GPUs
    [W.I.P. -> TensorCores & RT Cores] for Nvidia GPUs
    [W.I.P. -> OpenCL 2.0, OpenCL 3.0] for AMD GPUs
    Tree parity machine (TPM) -> [Neural cryptography]
    [W.I.P. -> Adversarial neural cryptography (NGANs) -> [Neural cryptography]
    Deep Reinforcement Learning (Agents, DQN, DDQN, D3QN)
    Deep Convolution Networks.
    Deep Deconvolution Networks.
    Deep Recurrent Networks (BiLSTM, LSTM, GRU)
    Generative adversarial network (GANs)
    Autoencoders, VAE.
    Stochastic gradient descent (SGD) & (Mini-Batch & Momentum)
    [W.I.P -> Stochastic gradient descent implicit updates (ISGD)]
    Direct Feedback Alignment (DFA)
    Rectified Linear Unit (ReLU)
    Exponential Linear Unit (ELU)
    Leaky Rectified Linear Unit (Leaky ReLU)
    Scaled Exponential Linear Unit (SELU)
    Gaussian Error Linear Unit (GELU)
    Regularizations L1 & L2
    Residual Blocks.
    Max Pooling,
    Average Pooling,
    Global Max Pooling.
    Global Average Pooling
    Add,concat,max,clip,norm,dropout,softmax,linear,sigmoid,tanh.

    Friends forgive me for the technical part but it is necessary for people who want to know what characteristics ChaosAI has.

    I started the integration of ChaosAI a few days ago and I am doing it in my few free hours, I will update this post with all the advances that I am going to do, I tell you that I am CTO in an A.I.development company.

    I have created this post to listen to your opinions and resolve your doubts, I would love to hear what you would like to do with AI in Unreal Engine and what ChaosAI may lack or what it can be useful for you.

    I need your opinions, I would like to listen to all of you.

    Linkedin https://www.linkedin.com/in/domingo-chamorro-40461187/
    Twitter @domyoriginal

    Greetings friends.
    Last edited by domyoriginal; 08-21-2020, 05:15 AM.

    #2
    This looks very amazing !
    I am looking forward to your work : )

    Comment

    Working...
    X