Download to read offline
Uploaded byOcean Protocol
Beyond Online PDFs
The document discusses the Ocean Protocol, emphasizing the importance of the data economy and the challenges of unlocking its potential, including data fragmentation and organizational silos. It outlines the roles of AI orchestration and compute-to-data protocols in facilitating data access while maintaining privacy. The roadmap for Ocean Protocol aims to create a permissionless ecosystem that encourages data discovery, ownership, and usage within a generalized AI asset framework.
More Related Content
![Blockchains for AI [With New Applications]](https://cdn.slidesharecdn.com/ss_thumbnails/20170522blockchainsforai-consensus2017-170523111907-thumbnail.jpg?width=640&height=640&fit=bounds)
Beyond Online PDFs
- 1. Ocean Protocol Beyond onlinePDFs Marcus Jones Monday November 3rd, 2019 Blockchain for Science @ Beta Haus, Berlin Powered by
- 2. Overview - Motivation: TheData Economy => The AI Economy - Ocean Protocol Overview - Marketplace - Data Science - Static assets in Ocean - Generalized assets = Services - No Data Escape / Federated Learning / Compute to Data - Beyond PDFs 2
- 3. Motivation - Whatis the Data Economy? 33 The Data Economy measures the overall impacts of the data market on the economy as a whole. It involves the following list of data enabled by digital technologies. The data economy also includes the direct, indirect, and induced effects of the data market on the economy. * Source: A study commissioned by the EU https://ec.europa.eu/newsroom/dae/document.cfm?doc_id=444 00 The value of the data economy in 2016 was worth nearly 2% (330 Billion Euros) of the European GDP. BY 2020, this will rise to 729 Bn -> Globally many Trillion EUR. ● Data Generation ● Data Collection ● Data Storage ● Data Processing ● Data Distribution ● Data Analysis ● Data Exploitation
- 4. Incentive to hoardand silo data “It is not who has the best algorithm that wins. It’s who has the most data.” - Banko, M., & Brill, E. (2001). Scaling to very very large corpora for natural language disambiguation. https://doi.org/10.3115/1073012.1073017 44
- 5. Banko + Brillproven by Deep Learning 55 Grace, K., Salvatier, J., Dafoe, A., Zhang, B., & Evans, O. (2018). Viewpoint: When will ai exceed human performance? Evidence from ai experts. Journal of Artificial Intelligence Research. https://doi.org/10.1613/jair.1.11222
- 6. Unlocking the dataeconomy: Data Orchestration The end-to-end automation of data-driven processes, raw data -> value 6 - data fragmentation - confusing provenance - organizational silos - complicated technology integrations ______________________ loss of potential value
- 7. Unlocking the dataeconomy: AI Orchestration Challenges in AI-driven processes 7 - data fragmentation - even more confusing provenance - organizational silos - complicated technology integrations - code authorship - training history - productionisation - logging ______________________ loss of potential value!
- 8.
- 9. Composition & registration ofstatic AI assets into Ocean
- 10.
- 11. Datasets registered intoOcean Protocol
- 12. AI assets registeredinto Ocean Protocol
- 13. Entire workflow asan asset
- 14. (A real data-scienceworkflow…)
- 15. Composition of astatic ‘asset’
- 16.
- 17. Satellite imagery: did:op:3fdcc402b9994d88828e82f9be16e40eaf8eed10036c48ae9a826415e3ca46ce Jupyter Notebook: did:op:5268ca64d7d843acae995f1712f0941c0dae57fb3ec0491bb6dda83d93f534c0 Pre-trainedmodel: did:op:4b8a4bd15e8e429ba7918e8d9005dc58923d0ce408834e0ea9089cd41fc780b3 Example: Deforestation detection in the Amazon
- 19. Example: (Publishing anasset) https://commons.nile.dev-ocean.com/
- 20. Key concepts Open source Decentralized Permissionless Communitygoverned Incentivized _______________ Unlocking data 2020
- 21.
- 22. Ocean Protocol EnhancementProposals 2222 Join the discussion and ideation at: https://github.com/oceanprotocol/OEPs
- 23. Networks • Local testnetwork • Public test network • Ethereum Mainnet • Commons network • Priced network 2323
- 24.
- 25. The generalized AIasset - a service
- 26. The generalized AIasset (it’s a service) SSH access to my GPU @ 2k OCEAN / hr
- 27. The generalized AIasset (it’s a service) SSH access to my GPU @ 2k OCEAN / hr Calculate the deforestedareas in uploaded image @10 OCEAN / image
- 28. The generalized AIasset (it’s a service) SSH access to my GPU @ 2k OCEAN / hr Calculate the deforestedareas in uploaded image @10 OCEAN / image Download the stream of IOT data @ 50 OCEAN / hr
- 29. The generalized AIasset (it’s a service) SSH access to my GPU @ 2k OCEAN / hr Calculate the deforestedareas in uploaded image @10 OCEAN / image Download the stream of IOT data @ 50 OCEAN / hr Train a federated model on our genetic data @ 500 OCEAN / 1%-increase
- 30. Research topics: Ownershipof AI Assets - Tied to identity - Ownership loss through escape of – Data – AI Model – ? - Shared ownership? - Delegated rights?
- 31. Research topics: Pricingof Digital Assets - Fixed price – Conditional – Prorated over time, accuracy, etc. - Negotiated - Bonding-curve based - Bounties/contests (See “Data Pricing” blog series series) https://blog.oceanprotocol.com/tagged/bonding-curves https://blog.oceanprotocol.com/lets-talk-about- data-pricing-part-i-bbc9cf781d9f
- 32.
- 33. protocol with incentives Bringmodel to the data f(x) private data modeling pipeline privately train model private model model predictions Data stays behind firewall 33
- 34. Compute to datamodes - On premise compute 3434
- 35. Compute to datamodes - On premise compute … third party premise compute 3535
- 36. Compute to datamodes - On premise compute … third party premise compute … federated learning 3636
- 37. Compute to datamodes - On premise compute ... third party premise compute … federated learning … homomorphic encryption 3737
- 38. Compute to datamodes - On premise compute … third party premise compute … federated learning … homomorphic encryption … production model as an AI Asset 3838
- 39. Federated / On-line/ Incremental learning Horizontal : Same feature space Vertical : Same ID space Transfer : Different feature and ID space 3939
- 40.
- 42. Orchestration of aResearch Ecosystem 4242
- 43. Orchestration of ResearchEcosystem 4343 100 OCEAN / Hour 594 Upvotes Complete version history Delegated partial payments Dataset never leaves premise, compute-to-data
- 44.
- 45.
- 46.
- 47. The road ahead(Ocean Protocol) Unlock data 1) Allow and incentivize discovery and access to data 2) Broaden the scope to allow composite general assets 3) Build compute-to-data Build a permissionless ecosystem 4) Expand incentivization mechanisms 5) Fully permissionless and ownerless 4747 Research and Experiment Provenance Curation Bonding curves Federated learning Identity Shared ownership Delegated ownership Non-fungible tokens
- 48. The road ahead(BigChainDB) Find product <> market fit 4848