Page 46 - AIH-2-3

Artificial Intelligence in Health Omnichain FL in healthcare AI

do not exchange local data. Rather, they transmit the foundational layer, known as Layer-0, that interconnects all
parameters of their local training results to a central blockchains, regardless of their smart contract technologies,
server, thereby achieving a collaborative AI model training thereby allowing all federated learning processes to operate
objective. 3 atop this infrastructure. This represents a highly compatible
6
However, the concern within federated learning is super multi-chain ecosystem that mitigates the limitations
incentivizing entities to contribute valid health data while of individual blockchains, ultimately serving the needs of
penalizing malicious actors. The latency and subjectivity AI training. During the construction of the Omnichain,
inherent in manual evaluations do not provide an optimal Cosmos SDK is a pivotal technology whose standardized
solution. The immutable and self-executing nature of development toolkit enables seamless communication and
smart contracts within blockchain addresses this issue transactions among unique parallel blockchains. With this
by enabling all participants to ascertain results instantly interoperability, the Cosmos SDK opens up a world of
without the need for intermediaries, thus facilitating a possibilities, allowing data, tokens, assets, and logic to be
reliable reward distribution mechanism for federated transmitted across multiple blockchains in a highly secure
learning to create an automated and standardized process. 4 and trustworthy manner. 7
Consequently, these two advantages position federated As illustrated in Figure 1, Omnichain supported by the
learning and blockchain as inherently complementary. Each Cosmos SDK facilitates communication among disparate
participant is assigned a blockchain identity, known as the blockchains. Omnichain connects various blockchains and
public address, wherein health data is stored, manifesting allows users to deploy smart contracts directly onto it by
in a distributed manner. Participants declare their public unified smart contracts tailored for diverse blockchains
address to engage in the federated learning process and, on the Omnichain. Furthermore, the federated learning
thereby, receive rewards according to smart contract process is capable of directly interfacing with the
criteria. Thus, in the healthcare sector, where pertinent Omnichain and executes these contracts to establish
ethical and sensitive considerations regarding health data incentive mechanisms. Simultaneously, smart contracts
are paramount, implementing real-time federated learning can be distributed across multiple blockchains without
for health information privacy protection and utilizing needing to consider the barriers arising from differing
blockchain to develop more intelligent incentive strategies consensus mechanisms or programming languages,
holds significant promise. eventually mitigating performance degradation arising
from blockchain heterogeneity while ensuring the integrity
2. Omnichain paradigm in federated of privacy and incentive mechanisms, which ultimately
learning aids in the training of AI models.
This innovative amalgamation of federated learning and Notably, this process does not impede the independent
blockchain effectively addresses the privacy concerns generation of blocks by individual blockchains, indicating
surrounding health data during the training process while that the construction of Layer-0 does not necessitate the
significantly enhancing the motivation of participants implementation of Layer-1 through a forking mechanism.
to contribute health data. However, AI practitioners This advancement effectively dismantles barriers between
have encountered challenges with this integration due different blockchains, resulting in a qualitative leap in both
to the heterogeneity among different blockchains. For the fluidity and functionality of information exchange.
instance, when a federated learning framework that was For example, once a federated learning model completes
originally developed in Solidity language and deployed a round of training updates on Ethereum, the trained
on Ethereum mainnet seeks to migrate to Solana, it model can be transmitted to Solana through Omnichain
may be reimplemented in Rust language. Meanwhile, to continue training. This approach eliminates the need for
implementing federated learning on diverse blockchains reliance on a single-point oracle or centralized bridge and
necessitates reconfiguring environments to accommodate likewise reduces the opportunities for attackers to exploit
various consensus mechanisms. The aforementioned cross-chain bridge vulnerabilities.
factors result in performance degradation for federated Moving a further step, Omnichain can also be
learning. This situation underscores the need for a integrated into existing hybrid solutions. Differential
5
unified blockchain environment that ensures a consistent privacy (DP) is a strategy aimed at mitigating the risks of
execution standard for conducting federated learning on side-channel attacks or differential analyses of parameter
health data across different blockchains. updates in federated learning. Injecting mathematically
The emergence of the concept of the Omnichain quantifiable noise into the parameter reporting and
addresses this critical issue. Omnichain constructs a novel aggregation processes eventually increases the difficulty

Volume 2 Issue 3 (2025) 40 doi: 10.36922/aih.5753