Usage¶

Despite advancements in the field of eXplainable Artificial Intelligence (XAI), many existing XAI tools remain complex and challenging to use.

The XPLAIN platform aims to address this challenge by making XAI more accessible, by providing an intuitive, ready-to-use environment that consolidates various XAI techniques helping users to better understand and apply AI insights effectively.

Notebooks & Datasets¶

We offer a range of tutorials and how-to guides presented as use cases in interactive notebooks. Users can explore and experiment with these resources to explore various XAI methods available. Additionally, datasets and trained models are accessible in our HuggingFace repository, and the notebooks can be opened and tested directly on Google Colab.

On our website, you will thus find:

• Tutorial Notebooks: Step-by-step guides to help users understand and apply XAI methods.
• Use Case Notebooks: Demonstrations of how XAI and our platform can address real-world problems.
• Datasets and Models: Embedded datasets and pre-trained models to facilitate experimentation.
• Other XAI Resourcse: Literature and resources that explain how Explainable Artificial Intelligence works.

These practical resources aim to help users understand and utilize the various features available on our platform.

graph LR
    subgraph Huggingface["Huggingface Repository"]
        direction LR
        models["Pre-trained models"]
        datasets["Datasets"]
    end

    notebooks["Use cases Notebooks"]
    colab["Google Colab"]

    models --> notebooks
    datasets --> notebooks
    notebooks --> |interactive session|colab

Online deployment through Google Colab allows for easy access and experimentation with our algorithms, tutorials, and use cases, without the need to install Python and XAI libraries on a local machine.

The interactive notebooks can be found in the Notebooks section.

Algorithms code¶

In algos we introduce our XAI algorithms. They are developed in an open source manner in our GitHub repositories. They consist of Python and C++ code using the pybind11 binding library.

graph TB
    subgraph Repo["GitHub repository"]
        subgraph hesxplainAlgo["Our XAI algorithms"]
            direction LR
            algoRuleextract["rule-extraction"]
            algoFuge["fuge"]
            algoDimlp["dimlpfidex"]
        end

        mlxplain["mlxplain"]
    end

    subgraph ExternalAlgo["Existing XAI algorithms"]
        existingAlgo1["SHAP"]
        existingAlgo2["LIME"]
        existingAlgo3["..."]
        omnixai["OmniXAI"]
    end

    hesxplainAlgo --> mlxplain
    existingAlgo1 --> omnixai
    existingAlgo2 --> omnixai
    existingAlgo3 --> omnixai
    omnixai --> mlxplain

The platform assembles our own developed algorithms as well as incorporating all algorithms proposed by OmniXAI through the overaching MLXplain package. This comprehensive suite of tools for XAI enables users to compare and choose the best techniques for their needs.

Explore our algorithms in the Algorithms section.

Python packages¶

Our algorithms are packaged in the latest standards-compliant formats and stored on our platform under an open, permissive, and business friendly licence.

The packaging process is entirely automated from the source repositories. This simplifies code maintenance and release processes on the public Python Index, allowing data scientists to obtain these algorithms quickly from their command line to integrate them into their workflow without friction.

graph LR
    subgraph repository["GitHub Repository"]
        sourcecode["Source Code"]
    end

    subgraph pypi["Python Index Repository"]
        direction LR
        mlxplain["MLxplain package"]
        dimlpfidex["dimlpfidex package"]
        rulesextract["rules-extraction package"]
    end

    subgraph user["User"]
        cli["Terminal CLI"]
    end

    repository --> |automated compilation|pypi
    pypi --> |pip install|cli

To obtain the Python package, follow the Installation guide.

Containerized images¶

Additionally to the use cases available as online interactive session, the platform offers ready-to-use deployment solutions in the form of downloadable Docker images available in the public GitHub artifact registry. This streamlined deployment of notebooks and datasets includes all the necessary libraries, allowing users to concentrate on explainability issues.

The following solutions are provided to facilitate deployment:

• Notebook Server Package: A pre-configured environment for running Jupyter notebooks.
• Python XAI Libraries: All essential XAI libraries, including our custom algorithms and OmniXAI.
• CUDA Support: Hardware-accelerated support for notebooks for users with NVIDIA graphics cards.

graph LR
    subgraph Repo["GitHub repository"]
            subgraph Docker["Docker Resources"]
                direction LR
                notebookServer["Jupyter server"]
                notebookLibs["PyTorch & CUDA"]
                dockerfile["Dockerfile"]
            end
        end

        subgraph Registry["Artifact Registry"]
            dockerImages["Docker Images"]
        end

        subgraph user["User"]
            privateServer["Private deployment"]
        end

    Docker --> Registry
    Registry --> privateServer

These containerized versions of our platform components allow for easy deployment on local machines or within private cloud environments. This is especially useful for advanced users and enterprises, where a private deployment of the platform allows some key advantages:

• Private Datasets: Securely manage and use private datasets within the HES-XPLAIN platform.
• Deployment Flexibility: Deploy our solutions within private infrastructures using Kubernetes or other orchestration tools.

Platform expansion¶

The platform is designed to be a flexible and collaborative tool, allowing anyone to add new algorithms to improve its explanatory capabilities. It also supports the creation of new notebooks for improved usability.

We have already used and extended the existing OmniXAI framework, leveraging its extensive documentation to incorporate new algorithms and ensure their seamless integration.

We strongly encourage contributions not only to the platform itself but also to the development of our algorithms. We aim to continuously improve the platform and expand its functionality in a collaborative manner, making it a valuable resource for the community.

Architecture overview¶

graph LR

    subgraph External["External projects"]
        direction LR
        omnixai["OmniXAI"]
    end

    subgraph Repo["GitHub repository"]
        direction LR
        souceCode["Source code"]
        dockerres["Docker resources"]
        notebooksSrc["Notebook sources"]
    end

    subgraph Pypi["Python Index repository"]
        direction TB
        mlxplain["MLxplain package"]
        dimlpfidex["dimlpfidex package"]
        rulesextract["rules-extraction package"]
    end

    subgraph Website["Website"]
        direction LR
        overview["Project overview"]
        notebookLinks["Notebooks index"]
        docs["Docs. & API ref."]
    end

    subgraph Datasets["Dataset repositories"]
        direction LR
        kaggle["Kaggle"]
        huggingface["Hugging face"]
    end

    subgraph Notebooks["Notebooks"]
        direction LR
        colab["Google colab"]
        docker["Docker images"]
    end

    subgraph Endusers["End users"]
        direction LR
        decisionMakers["Decision makers"]
        engineers["Engineers"]
        dataScientists["Data scientists"]
        enthousiasts["Students & ML enthousiasts"]
    end

    omnixai-- "empowers" -->mlxplain
    souceCode-- "is compiled into" -->Pypi
    dockerres-- "supplies" -->docker
    souceCode-- "is documented in" -->docs 
    notebooksSrc-- "are hosted by" -->Notebooks
    Notebooks-- "provide help to" -->Endusers
    Datasets-- "can be used with" -->Notebooks
    notebookLinks-- "links to" -->Notebooks
    Website-- "enlight" -->Endusers
    Pypi-- "can be used by" -->Endusers
    Pypi-- "are used in" -->Notebooks