Introduction

Traditional cloud computing solutions, like those provided by AWS and Alibaba Cloud, offer high-quality computing resources but come with high costs. Decentralized cloud computing is a new approach that uses blockchain technology to enable computing resources worldwide to join the network as nodes. These nodes can provide computing power and earn tokens as incentives. Decentralized computing has many applications, including graphic rendering, video transcoding, artificial intelligence, and machine learning. In the current bull market, AI has become a popular focus. The AI industry is growing rapidly, with computational complexity potentially doubling every three months, leading to a significant increase in demand for computing power. This rising cost of decentralized computing is challenging for individuals and small businesses involved in machine learning who need cloud computing services. Gensyn aims to democratize AI through decentralization, reducing the cost of computing power needed for learning. Based on the Substripe protocol, Gensyn uses smart contracts to facilitate allocating and rewarding machine learning tasks. It also aims to create a large-scale distributed deep learning computing protocol, combining probabilistic learning proofs with cryptocurrency and incentive mechanisms to offer a more efficient and scalable computing model for the AI field. This article will delve into the operation logic of Gensyn’s protocol and the current state of its development.

What is Gensyn?

Gensyn is a GPU computing network specifically designed for machine learning. It increases computing power for machine learning by leveraging various long-tail computing devices worldwide, such as small data centers, personal gaming PCs, and Macs. Although still in development, Gensyn has made significant progress in its phased product development. The economic model is yet to be released, and the protocol is planned to be integrated into the Polkadot ecosystem.

The Gensyn team is based in London, UK. The co-founders have PhDs in computer science and were early entrants into the blockchain industry. Other team members also have experience in artificial intelligence, and the team is expanding. Financially, the team is well-supported. They received $1.1 million in funding in July 2021, $6.5 million in seed funding led by Eden Block in March 2022, and $43 million in Series A funding from a16z in June 2023. Several other investors have also backed this funding. The team has indicated that this funding round will be used to grow the team and speed up the protocol’s launch.

Who is Involved?

The Gensyn ecosystem includes four key roles: Submitters, Solvers, Verifiers, and Reporters.

• Submitters are Gensyn users who submit tasks requiring computation and pay the associated fees.
• Solvers are the primary workers who train the models and create proofs for verification.
• Verifiers play a crucial role in the ecosystem. Model training in machine learning is often a non-deterministic process due to random initialization, algorithm optimization, and data perturbation. Verifiers bridge the gap between non-deterministic training and deterministic computation. They validate the model by solving mathematical proofs and comparing the model’s output with the expected results to ensure its reliability.
• Reporters are the system’s final safeguard. They review the Verifiers’ work and can raise challenges to earn rewards.

How Does It Work?

The operational process of Gensyn’s product includes six stages: task submission, model training, proof generation, proof verification, challenge, and settlement. The model training stage occurs off-chain, while proof verification and economic incentives happen on-chain.

1. First, the submitter must upload three types of files: metadata of the task and hyperparameters, model binary files, and publicly available pre-processed training data. These files are critical components in machine learning model training.
2. After analysis, the task enters a public task pool, and a single solver is selected to execute the task. The solver will execute the task based on the cloud data uploaded by the submitter, the provided model, and the training model.
3. While executing the training task, the solver must also set checkpoints at planned intervals and store metadata during the training process to generate learning proofs. This ensures that the verifier can accurately replicate the optimization steps later. This process constructs a set of proven, pre-trained base models that provide the foundation for subsequent optimization steps.
4. After completing the task, the solver must mark the task’s completion status on-chain and place the learning proofs in a publicly verifiable location for the verifier. Verifiers obtain verification tasks from the public task pool, rerun part of the proofs, and perform distance calculations. The blockchain uses these distances to determine whether the verification matches the learning proof.
5. After verifying the learning proof, reporters can replicate the verifier’s work to check if it has been correctly executed. If a reporter believes the verification was incorrectly executed, they can initiate an arbitration challenge against the verifier to earn rewards. These rewards come from the verifier’s deposit or the reward pool.
6. In this process, participants receive corresponding rewards based on probabilistic and deterministic checks’ conclusions.

Cost and Benefits

Large enterprises typically have the budget to opt for centralized computing services. In contrast, Gensyn’s primary users are small businesses, individual developers, and research teams involved in machine learning. These users are often price-sensitive and cannot afford the high costs of centralized computing power. The major advantage of decentralized machine learning is the significant cost reduction. Gensyn’s official pricing shows that their service costs only $0.40 per hour, compared to $2 per hour for equivalent computing power from AWS, resulting in an 80% cost reduction.

Source：docs.gensyn

Opportunities and Risks

Gensyn targets highly sensitive users to computing costs, which means it addresses a relatively smaller market. While the protocol’s vision aligns with current market trends, it faces several risk factors. For example, at the start of the process, users must upload the model framework, training data, and hyperparameters to the Gensyn network. Using open-source data does not involve privacy issues, but uploading proprietary models can lead to information leaks.

The devices using the Gensyn network can vary greatly in computing power, storage capacity, and network connectivity. Gensyn transfers model parameters, tasks, and verification data between different devices. Devices with lower network bandwidth may experience transmission delays, impacting task distribution and result verification. Consequently, the differences in device capabilities can affect the overall efficiency of the system.

Conclusion

Gensyn is a GPU computing network dedicated to machine learning, aiming to connect developers and solvers while leveraging global resources to lower the costs associated with machine learning. Its vision aligns well with current market trends and AI hot topics. However, Gensyn is still in the development phase, primarily attracting small businesses, individual developers, and research teams who are price-sensitive. The market for this service is still relatively small, and the project will face substantial challenges in achieving widespread practical implementation.