DeepchainAda, CDDLink and ValCert


Recently there has been a lot of talk about AI models showing bias. For example some of the healthcare related models were biased while suggesting treatments. One of the root causes of this is lack of data. Usually many projects cannot share data because of Privacy issues. So models get trained with limited data.

One advancement that happened was the concept of federated learning or distributed learning. In this there is a central server and model gets trained where data is and only model parameters are send to a central sever which aggregates this.

Blockchain like Cardano can help decentralize this federated learning. Along with this there are also concepts of Split Neural Nets that help in training. Multiple papers have talked about this and people have implemented this in mainly an experimental chain or some in Eth. So DeepchainAda focuses on building this distributed learning framework on Cardano. The idea is to use the Smart Contracts to orchestrate these tasks. Couple this with SingularityNet, the model can be enabled to work in the decentralized marketplace along with other AI models.

So in short the DeepchainAda project focuses on getting the Plutus Smart Contracts to help Distributed Machine Learning and interface the model with SingularityNet Framework using the AI DSL (Idris). The aim is to also enable NuNET (distributed computing) incase data provider does not have compute. All of this should be achieved in such a way that Privacy and Auditability is preserved 


What is PGWAD pool project CDDLink? Collaborated Distributed Deeplearning is focused on getting models trained on #NuNET. This is exploratory project as NuNet is evolving & it will have various compute scale available. This should eventually Link to DeepchainAda 

Why CDDLink? 

Training DL model needs lot of compute & only big techs can afford. Small guy like me have to rely on AWS to train my models, very centralized. NuNET provides distributed HW & hence very apt for such usecase. But NuNET provides various class of HW so provides ability to choose.

With various HW types it provides a choice of heterogeneous compute. CDDLink will explore this space. NuNet also provides volunteer computing & CDDLink will explore these. Speed & convergence of training should not be affected by this & that is the goal.


Smart contracts on eUTXO blockchains like Cardano and Ergo are based on functional languages. The contracts are mainly validators and hence suitable for formal verification. In this project we will provide framework to write validator in Coq and extraction to functional languages using MetaCoq and ConCert framework. We also reuse the property based testing of the ConCert framework.

This is part of Blockchain layer of DeepchainAda focusing on formal verification of contracts