Distributed learning of genAI embedded whole slide images for digital pathology projections (Praful).

Demo code: https://github.com/cloud4bio/cloud4bio.github.com/wiki/Federation-Demo 

Journal Club
A whole-slide foundation model for digital pathology from real-world data
Nature. 2024; 630(8015): 181–188. doi: 10.1038/s41586-024-07441-w

...

Hackathon

Let's go through Praful's embedding, and distributed learning, step by step.

...

WebNN is the basis for native AI deployment in the browser: https://webmachinelearning.github.io/webnn-intro/

Praful's simplified NN module is fully working.

Praful - simpleNN - https://github.com/episphere/tfjs/wiki 

```javascript
const { TFJSModel } = await import("https://episphere.github.io/tfjs/modelBuilder.js");
const model = new TFJSModel({});
await model.loadData({
    "datasetURL": "https://episphere.github.io/ai/data/irisOHE.json",
    "labelFieldName": "species",   
    "trainTestRatio": 0.8
});
await model.build();
await model.train();
await model.test();
```

A simplified decentralized learning demo:

// Data Loading:

const tfjsModelBuilder = await import("https://episphere.github.io/tfjs/modelBuilder.js")

const m = new tfjsModelBuilder.TFJSModel({})

await m.loadData({

    'datasetURL': "https://episphere.github.io/ai/data/irisOHE.json",

    "labelFieldName": "species",

    "trainTestRatio": 0.8

})

document.addEventListener("webFed_newPeer", (event)=>{

    const delta = event.detail

    const newPeerOperations = delta.filter(operation => {

        return operation.hasOwnProperty("insert")

    })

    if (newPeerOperations.length > 0) {

        const newPeerNames = newPeerOperations.map(operation => {

            const insertOperations = operation.insert.filter(insertOp => insertOp.hasOwnProperty("name") && insertOp.hasOwnProperty("_webRTCPeerId"))

            return insertOperations.map(op => op.name)

        }).flat()

        newPeerNames.forEach(newPeer => {

            console.log(`New Peer ${newPeer} joined.`)

        })

    }

})

const webFed = await import("https://episphere.github.io/lab/webFed_yjs.js")

const fed = new webFed.WebFed({

    'signalingServer': "https://signalyjs-df59a68bd6e6.herokuapp.com",

    'selfName': "Chrome",

    'federationName': "testFederation",

    'dataset': m.dataset

})

document.addEventListener("webFed_synced", (event) =>{

    if (event.detail.synced) {

        const allPeers = fed.getAllPeers()

        allPeers.forEach(peer => {

            if (!peer.isSelf)

                console.log(`Connected to peer ${peer.name}`)

        })

    }

})

// Modeling + Federation

await fed.initializeModel({

    'framework': "tfjs",

    'inputShape': [4],

    'architecture': [5, 3, 3],

    'activation': "tanh",

    'bias': true,

    'outputActivation': "softmax",

    'optimizer': "adam",

    'lossFunc': "categoricalCrossentropy",

    'metrics': ["accuracy"],

    'batchSize': 30

})

await fed.setTrainingParams({})

await fed.startTraining()

• MNIST Fed: https://episphere.github.io/lab/mnistFed.html 

Reference manuscripts

• Heterogeneous Federated Learning: State-of-the-art and Research Challenges: https://arxiv.org/pdf/2307.10616 

• Braid HTTP Proposal: https://github.com/braid-org/braid-spec

• Braid Interactive Demo: https://braid.org/demo/interact 

Note also that October 15th-16th, 2024, the NCI Office of Data Sharing (ODS) has been hosting the Annual Data Sharing Symposium titled Driving Cancer Advances through Impactful Research.