Existing Solutions
![Document image Document image](https://images.archbee.com/a-cXOL8NuAm7OnEOlQ3FQ/c2fwB1ocFt-iOeJe5aZ6-_gensyn-15.png?format=webp)
Present day ML computing solutions are either very expensive to access, in part because of the oligopolistic market structure, or not able to provide the complex compute required to power large-scale AI models. These solutions include services by providers such as AWS, Google Cloud Platform, Microsoft Azure, etc.
AI and machine learning have started drifting away from the cloud, in part because of the exessive costs but also because the increased performance of in-house solutions.
Voluntary grid computing services and decentralized blockchain protocols could be used to overcome aforementioned challenges by providing complex compute in a cost-efficient way.
Let’s take a look at both approaches to evaluate if they are able to provide trustless compute in a cost-efficient way.
The first option for trustless compute are voluntary grid computing services such as SETI@Home and BONIC. These services provide trustless, voluntarily-networked, latent computing power.
However, grid computing solutions are not ideal for ML compute:
- Compute is predominantly used to solve embarrassingly parallel problems.
- Majority of ML problems are inherently state dependent, requiring new methods for parallelisation and verification.
- Volunteer networks model participants as rational actors in a philanthropic system, adding financial transactions for the payment of ML computing services changes the incentive mechanisms and introduces exploitation vectors.
Distributed ledger technologies such as decentralized blockchain protocols merge the concepts of financial incentivization, trustlessness, and grid computing. With the help of a blockchain, a computational environment can be created that rewards compute providers (i.e. nodes) through gas fees. Thanks to these features, blockchain technology is more suitable to provide trustless compute than aforementioned grid copmuting services.
However, existing blockchain protocols such as Bitcoin and Ethereum are unsuitable for deep learning as shown below:
- Trustless consensus can only be achieved by extremely expensive on-chain replication of work.
- On-chain replication increases time and costs of computation significantly.
Even blockchains that perform computational work off-chain are not well suited for deep learning models requiring extreme computational resources.
Gensyn seeks to overcome these challenges by creating a decentralized machine learning compute protocol.
![Doc contributor](https://s3.amazonaws.com/archbee-animals/lion.png)