文件列表:

app/
escrow/
minter/

# Hackathon Solution For this hackathon, we managed to add standard Escrow procedures combined with the Solana token 2022 extension standard (described in [details here](https://github.com/UtilinkAI/Utilink/blob/main/escrow/README.md)). In addition to that, we minted [our token](https://explorer.solana.com/address/Gj8suaHabEVRJGC75i6RFeezjz43s6scBE6tCyTjJAUE?cluster=devnet) to the devnet. We automated the process, so we can perform a few more tests before minting our token to the mainnet. We also present a mocked version of our dApp - how it will function and look once the proper implementation phase begins. For further details, you can [visit our site](https://utilink.xyz/) and [follow us on X](https://twitter.com/utilinkAI). # Motivation Our motivation stems from the recognition of the increasing demand for accessible, transparent, and cost-effective AI solutions. By leveraging Solana blockchain technology and the Utilink token (ULNK), we aim to create a sustainable ecosystem where participants are incentivized to contribute, collaborate, and innovate. Ultimately, our goal is to make advanced AI capabilities accessible to all while fostering a community-driven approach to development and growth. # Project description and architecture Utilink is entering the web3 game to disrupt the centralized services market for training AI models by providing a blockchain-enabled platform for distributed training and decentralized delivery of machine learning models. At current our system design is underway incorporating advanced cryptographic techniques, secure p2p data sharing, and federated machine learning technology. Leveraging the speed of the Solana network, known for its high throughput and low transaction costs, as well as the integrity of zero-knowledge proofs, Utilink will enable verification of training results without revealing the underlying data in order to coordinate between different participants in the ecosystem.  ## Zero Knowledge Training Logic Unit (ZK TLU) - Dividing the training dataset into smaller chunks that can be processed by different nodes, ensuring data privacy through encryption and decentralised storage. - Training models on multiple nodes simultaneously to speed up the training process. - Combining the training results from different nodes into a final model, using techniques like federated learning to maintain privacy and security. Although many proposals of Incentivized and Decentralized Federated Learning Frameworks (FLFs) exist, we have not yet seen any full-fledged production-ready FLF. Through ZK TLU we Facilitate distributed training across multiple nodes to leverage computational resources efficiently.  # Examples **Job Provider** ``` Alice has some satellite data and she wants to train a Machine Learning model to perform image segmentation. However, Alice does not have enough resources to train as large a model as she would wish, so she comes to Utilink. Upon successful authentication, she can encrypt her data and upload it to a p2p storage to induce security. Data is broken into smaller pieces and prepared for federated learning (FL). Once the data and her model architecture are ready we share them with the pool of workers that would train Alice's model. ``` **Worker** ``` Bob on the other side has a couple of quite powerful GPUs at hand and he does not use them at their full capacity. He decides to lend them through Utilink and get incentivised for this. Bob joins the pool of federated workers and is able to help training large models. ``` **Inferrer** ``` John already owns a trained model, which he is quite proud of and he wants to share it with the community. He can either make a trained model public or upload a completely different one (trained offline) to a p2p storage through Utilink. Either way, he gets incentivised per usage. ``` # Team Building such a platform requires a multidisciplinary approach, combining expertise in blockchain development, cryptography, distributed computing, machine learning, design, and business analysis. 

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：