Smart contracts on the EVM are very limited. This makes it challenging to create applications that require more compute and data beyond the state of the blockchain. Furthermore, data is mostly public, making it difficult to remain private. ZKP (Zero Knowledge Proofs) offer a way to scale Ethereum and help offer privacy where needed. However, writing custom circuits novel applications is difficult.
Using machine learning, developers don't need to write custom circuits and can specify intended behaviors and have a computer learn the circuit. Leveraging on the EZKL library, Gelkin helps developers extend their smart contracts with ZKML (Zero Knowledge Machine Learning). Developers can then extend their smart contracts by leveraging on existing verifiers or by deploying verifiers themselves through the service.
Through Gelkin, EZKL, and ZKML, smart contracts will be able to offer more complex functionality like:
On-chain credit scoring to offer more kinds of DeFi loan arrangements
Anomaly detection, to detect potentially malicious actors and prevent attacks
Dynamic airdrop mechanisms with real-time incentives to consistently incentivize behaviors, versus the current model of retroactive merkle drops
Gas-cost savings, by rolling up many conditions into a single ZK proof and ML output that can be used on-chain.
ZKML is at a pivotal moment. Advances with projects like RISC0 and Halo2 make more computationally intensive ZK feasible. With WebGPU on the horizon, hardware acceleration on the web for edge compute becomes an interesting possibility. While huge GPT-3 compute will not be available, in the near to mid term simpler ML models can be used onchain. With this opportunity present, Gelkin seeks to help accelerate ZKML by making ZKML easier for developers.