What is Stained Glass Transform?

Introduction

Stained Glass Transform holistically transforms data to a randomized re‑representation through a stochastic data transformation which is learned based on Protopia AI’s patented technology. These representations are irreversible to the original data but are understandable by the target AI model. Randomized re‑representations can be used by AI tasks with high accuracy without exposing the original raw data.

Stained Glass Transform is data agnostic and can be applied to any data type, including text, images, audio, video, and more.

A Stained Glass Transform learns a re-representation so that a particular model will still understand the transformed data. This means that a Stained Glass Transform is model-specific, i.e. changing a model will require generating a new Stained Glass Transform. However, a Stained Glass Transform can be built for any Deep Learning model, including those that are already trained and in production.

Example of Stained Glass Transform with various data types

Large language model generation

flowchart LR
    subgraph DATA [" "]
        direction TB
        DATA_IMG[Rewrite the following email to be more professional.
Hey Maria,
Your broken laptop is fixed now. Come pick it up at the repair store
where you dropped it last week. We will be there from noon to five.
Anytime then is good for us.
Peace,
Your Repair Team]
        DATA_LBL[Original Data]
        DATA_IMG ~~~ DATA_LBL
    end

    subgraph SGT [" "]
        direction TB
        SGT_IMG(<img src="images/stained_glass_transform_logo.png"/>)
        SGT_LBL[Stained Glass Transform]
        SGT_IMG ~~~ SGT_LBL
    end

    subgraph RRR [" "]
        direction TB
        RRR_IMG["}); ', rite}, // >?[<') /** >?[<sy */ ]; };
 …᥀ */?[<'); ){ // }, '); // }; ()); },  // ]>?[<'> }, ', ',
 on}); '); ?[<᥀ }); }, ', }; …᥀>?[<iNdEx()); }); */ ]; }; '); \'>"]
        RRR_LBL[Randomized Re-representation]
        RRR_IMG ~~~ RRR_LBL
    end

    style DATA_IMG fill:#FFFFFF00, stroke:#FFFFFF00
    style SGT_IMG fill:#FFFFFF00, stroke:#FFFFFF00
    style RRR_IMG fill:#FFFFFF00, stroke:#FFFFFF00

    DATA --> SGT --> RRR

Network log data classification

flowchart LR
    subgraph DATA [" "]
        direction TB
        DATA_IMG[15, 1, 72, Europe \ Berlin \ Sales,
Gerrard S, 25. 33. 8. 119, kenta.
s, FileCopy, File copy Source,\\
192. 168. 102. 241 \ PUBLIC \
Confidential \ customer
information 2. XLS,,,
confidential folderー,,,]
        DATA_LBL[Original Data]
        DATA_IMG ~~~ DATA_LBL
    end

    subgraph SGT [" "]
        direction TB
        SGT_IMG(<img src="images/stained_glass_transform_logo.png"/>)
        SGT_LBL[Stained Glass Transform]
        SGT_IMG ~~~ SGT_LBL
    end

    subgraph RRR [" "]
        direction TB
        RRR_IMG[Front Solitary Inheritance, 73
Catch Remember Shanghai
Ndlauuri Muroran Axillary
Nakamura Trump Card Shimbashi
Rado for short Military history
Sonouchi Masuda Body
Opportunity Chasing ft No. Ritsu
Animal Royon 2 Maneki ioni]
        RRR_LBL[Randomized Re-representation]
        RRR_IMG ~~~ RRR_LBL
    end

    style DATA_IMG fill:#FFFFFF00, stroke:#FFFFFF00
    style SGT_IMG fill:#FFFFFF00, stroke:#FFFFFF00
    style RRR_IMG fill:#FFFFFF00, stroke:#FFFFFF00

    DATA --> SGT --> RRR

Facial recognition

flowchart LR
    subgraph DATA [" "]
        direction TB
        DATA_IMG[<img src="images/navy/navy_raw.png"/>]
        DATA_LBL[Original Data]
        DATA_IMG ~~~ DATA_LBL
    end

    subgraph SGT [" "]
        direction TB
        SGT_IMG(<img src="images/stained_glass_transform_logo.png"/>)
        SGT_LBL[Stained Glass Transform]
        SGT_IMG ~~~ SGT_LBL
    end

    subgraph RRR [" "]
        direction TB
        RRR_IMG[<img src="images/navy/navy_transformed.png"/>]
        RRR_LBL[Randomized Re-representation]
        RRR_IMG ~~~ RRR_LBL
    end

    style DATA_IMG fill:#FFFFFF00, stroke:#FFFFFF00
    style SGT_IMG fill:#FFFFFF00, stroke:#FFFFFF00
    style RRR_IMG fill:#FFFFFF00, stroke:#FFFFFF00

    DATA --> SGT --> RRR