EthOptimise: AI-Driven Restaking Optimisation Agent

# EthOptimise: AI-Driven Restaking Optimisation Agent for Ethereum ## :bulb: Project Abstract :::success EthOptimise aims to develop an AI-driven economic agent for Ethereum restaking. ::: This project focuses on optimising restaking strategies for validators and stakers by analysing Ethereum's unique market conditions, staking rewards, validator performance, and network security metrics. The goal is to maximise returns for stakers and enhance the overall security and efficiency of the Ethereum network. ## :pushpin: Objectives Maturing AI technology brings with it opportunities and challenges in the emerging combined field of AI & Crypto. Each technology has unique qualities that can be leveraged to the benefit or detriment of the other. Key objectives for this project are to progress the following concepts: :::success #### :small_blue_diamond: *Advanced Decision-Making Algorithms*: ::: Incorporate sophisticated AI and machine learning algorithms capable of making complex decisions based on real-time data analysis, historical trends, and predictive modeling. These algorithms would need to evaluate market conditions, network status, and validator performance to make informed restaking decisions without human intervention. :::success #### :small_blue_diamond: *Dynamic Risk Management*: ::: Implement dynamic risk assessment models that continuously evaluate the risk-reward ratio of staking strategies. The agent would use these models to adjust its strategies in response to changing network conditions, aiming to optimise returns while minimising exposure to slashing risks and validator underperformance. :::success #### :small_blue_diamond: *Self-Adjusting Algorithms*: ::: Design the AI algorithms to be self-learning and adaptive, enabling the agent to refine its decision-making processes based on outcomes and evolving market dynamics. This feature would be crucial for maintaining optimal performance over time, as the Ethereum ecosystem and its economic models evolve. ## :dart: Outcomes :::success This project incorporates two major developments in the Ethereum ecosystem: - the proliferation of **restaking**, and - the maturation of **AI technologies**. Therefore the key benefits to the greater Ethereum ecosystem lies in the syngergy of these two major developments. ::: Moreover, additional research into restaking will contribute to a greater understanding of benefits and risks thereof. Leveraging AI to assist in these tasks and learning from emergencing trends and data will further benefit the Ethereum ecosystem and deepen our understanding of the possibilities and implications of employing AI agents with varying degrees of autonomy. ## :books: Grant Scope The algorithms and AI agent will be designed specifically for matters relating to Ethereum staking and restaking. Within the constraint of a proposed 9 month project duration, the goal is to design an autonomous AI agent to incorporate some, but not all, aspects of the three research objectives mentioned above. :::success **Expected output**: - Document detailing the literature research for staking and restaking. This includes researching restaking providers such as Eigenlayer. - Data collection for Ethereum staking and one restaking provider. - Data collection for validator rewards and penalties. - A working prototype of the AI agent that monitors, alerts and digests new information from one restaking provider. - Blog post to share the research. - Open source code for AI agent. - Conference or journal paper. ::: ## :busts_in_silhouette: Project team The project team of three comprises of two researchers and one expert academic collaborator. :::success ### Sandra Johnson (researcher): ##### 20 hours per week ::: Sandra is a staff researcher at Consensys Software and has a PhD in Environmental Statistics. She is a senior adjunct lecturer at the School of Mathematics, Queensland University of Technology (QUT), Brisbane. Prior to joining ConsenSys she worked as a Data Scientist at Flight Centre and as a Research Fellow in Applied Statistics at QUT. Her research interests include data science technologies, decision making under uncertainty, Bayesian network modelling and statistical modelling applied to the Ethereum ecosystem. [[Google scholar]](https://scholar.google.com.au/citations?hl=en&user=1gsap5oAAAAJ) [[LinkedIn]](https://www.linkedin.com/in/sandjohnson/) :::success ### Roberto Saltini (researcher): ##### 4 hours per week ::: Roberto is a senior staff researcher at Consensys Software where he leads the Dependable Distributed System team. Roberto has been involved in a number of projects involving both the design and analysis, often performed by leveraging formal methods, of distributed protocols applied to blockchain systems, such as the Ethereum fork choice protocol, the Distributed Validator Technology (DVT) protocol, and the BFT protocols IBFT, QBFT and BigFooT. [[Google scholar]](https://scholar.google.com/citations?user=J29ceBQAAAAJ&hl=en) [[LinkedIn]](https://www.linkedin.com/in/roberto-saltini/) :::success ### Kerrie Mengersen (academic collaborator): ##### 2 hours per week ::: Kerrie Mengersen is a Distinguished Professor of Statistics at QUT and the Director of the QUT Centre for Data Science (CDS). CDS encompasses around 180 researchers from across the University with expertise in all facets of data collection, curation, privacy, modelling, visualisation and analysis, underpinning applications in twelve key domains including health, business, digital systems and social systems. She is also a co-founder of the Australian Data Science Network (ADSN) which brings together 32 centres in data science across the country. This concentration of expertise will be available to the proposed project. Dr Mengersen's research sits at the intersection of computational and applied statistics and machine learning, and focuses on developing ways to efficiently collect, analyse, share and trust diverse data sources. Her applied work focuses on health, environment and industry. [[Google scholar]](https://scholar.google.com.au/citations?hl=en&user=eiD83s4AAAAJ) [[LinkedIn]](https://www.linkedin.com/in/kerrie-mengersen-197347208/) # :memo: Background The primary researcher on this project has worked with the expert researcher and academic collaborator for 5 years and 15 years, respectively. ::: success ### :small_blue_diamond: Publications & Blog Posts ::: The publications and blog posts listed here may not directly relate to the proposed project, but demonstrates the expertise of the research team to employ the approaches mentioned in the papers, as well as domain knowledge. :::success #### Sandra Johnson: ::: - [1] S. Johnson,K. Mengersen and P.O'Callaghan "Proposer selection with increased MaxEB (EIP-7251)", 2024. [Link to blog](https://ethresear.ch/t/proposer-selection-with-increased-maxeb-eip-7251/18144) - [2] S. Johnson, D. Hyland-Wood, A. L. Madsen, and K. Mengersen, “Stateful to Stateless: Modelling Stateless Ethereum,” Electron. Proc. Theor. Comput. Sci., vol. 355, pp. 27–39, Mar. 2022. [Link to paper](https://arxiv.org/abs/2203.12435v1) - [3] S. Johnson, B. Cristescu, J. T. Davis, D. W. Johnson, and K. Mengersen, Now You See Them, Soon You Won’t: Statistical and Mathematical Models for Cheetah Conservation Management. 2017. :::success #### Roberto Saltini: ::: - [1] R. Saltini. "BigFooT: A robust optimal-latency BFT blockchain consensus protocol with dynamic validator membership." Comput. Networks 204: 108632 (2022) [Link to paper](https://www.researchgate.net/profile/Roberto-Saltini/publication/356793770_BigFooT_A_robust_optimal-latency_BFT_blockchain_consensus_protocol_with_dynamic_validator_membership/links/62439f078068956f3c57c1f4/BigFooT-A-robust-optimal-latency-BFT-blockchain-consensus-protocol-with-dynamic-validator-membership.pdf?origin=publication_detail) - [2] R. Saltini, D. Hyland-Wood. "IBFT 2.0: A Safe and Live Variation of the IBFT Blockchain Consensus Protocol for Eventually Synchronous Networks". CoRR abs/1909.10194 (2019) [Link to paper](https://arxiv.org/abs/1909.10194) - [3] P. Robinson, D. Hyland-Wood, R. Saltini, S. Johnson, J. Brainard. "Atomic Crosschain Transactions for Ethereum Private Sidechains". CoRR abs/1904.12079 (2019) [Link to paper](https://arxiv.org/abs/1904.12079) :::success #### Kerrie Mengersen: ::: - [1] A. Snoswell, B. Hyland-Wood, K. Mengersen, T. Orth, B. Cook, “QUT Centre For Data Science submission on the draft G7 Guiding principles for organizations developing advanced AI systems” Analysis and Policy Observatory (APO), 2023. [Link to paper](https://scholar.google.com.au/citations?view_op=view_citation&hl=en&user=eiD83s4AAAAJ&sortby=pubdate&citation_for_view=eiD83s4AAAAJ:FKYJxdYMdFIC) - [2] O.Forbes, E. Santos-Fernandez, P.P.Y. Wu, H.B. Xie, P.E Schwenn, J. Lagopoulos, L. Mills, D.D. Sacks, D.F. Hermens, K. Mengersen, “clusterBMA: Bayesian model averaging for clustering” Plos one Vol.18 (8), 2023. [Link to paper](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0288000) - [3] I. Ullah, K. Mengersen, R.J. Hyndman, and J. McGree, "Detection of cybersecurity attacks through analysis of web browsing activities using principal component analysis", 2021. [Link to paper](https://scholar.google.com.au/citations?view_op=view_citation&hl=en&user=eiD83s4AAAAJ&cstart=80&sortby=pubdate&citation_for_view=eiD83s4AAAAJ:dFKc6_kCK1wC) # :bar_chart: Methodology The above-mentioned research objectives encompass a longer roadmap than this proposed project. Moreover, we envision that after development of the various agents and algorithms, the efficacy of each deliverable will be monitored and evaluated with respect to autonomy, and its responsiveness to changing conditions on which it was trained. The project therefore aims to contribute to the research objectives in the following way: #### :small_blue_diamond: *Data collection and pre-processing*: The precursor to all three objectives mentioned above is the collection, pre-processing and interpretation of the data available to inform models and learning. #### :small_blue_diamond: *Self-adjusting algorithms for decision-making and risk management*: The strength of AI is its ability to dynamically learn from the present to better predict the future. Thus, our knowledge today becomes our prior tomorrow, which is combined with tomorrow's new information to become our updated knowledge base. Therefore the application of Bayesian learning and associated methods are very relevant. In designing advanced decision-making algorithms, we aim to personalise the process, i.e. learning from the whole cohort, but providing targeted decision support for individuals or subgroups. This is achievable via a [Bayesian] hierarchical model that allows for individual behaviour as realisations (or deviations) from a collective population. Moreover, during development of the algorithms, various other statistical approaches may also be used and/or combined, such as multi-criteria learning/optimisation through accurate forecasting/prediction/future decision-making combined with security. # :timer_clock: Timeline We summarise the milestones and deliverables in the table below, and visualise a proposed timeline spanning 9 months. | **Milestone** | **Expected deliverable** | |:----------------|:------------| | *Milestone 1*: Literature research on staking & restaking | Document detailing the literature research for staking and restaking. This includes researching restaking providers such as Eigenlayer| | *Milestone 2*: Data gathering & wrangling for staking & restaking | Datasets of collected and processed data for staking and a restaking provider.| | *Milestone 3*: Data gathering & wrangling for validator rewards and penalties| Datasets of collected and processed data | | *Milestone 4*: Develop prototype of an AI agent to monitor, alert and digest new information from one restaking provider.| Working prototype of the AI agent| | *Milestone 5*: Blog post to share research| Blog post| | *Milestone 6*: Open source code for AI agent| Add AI agent code to public GitHub repo| | *Milestone 7*: Document research as a conference paper or journal paper | Conference or journal paper| ![Screenshot 2024-03-10 at 6.37.31 pm](https://hackmd.io/_uploads/r1DGTDjp6.png) ## :money_with_wings: Budget The estimated costs to fund this grant proposal for a 9 month project: ### :small_blue_diamond:Principal Researchers Costs: - Primary researcher: (US$ 22,000.00 * 0.5) * 9 = US$99,000.00 - Expert Researcher: (US$ 22,000.00 * 0.1) * 9 = US$19,800.00 - Collaborators: (US$ 22,000.00 * 0.05)* 9 = US$ 9,900.00 ### :small_blue_diamond:Indirect costs: - Publication fees in journals - US$1,000.00 (Pay to have open access.) :::success **Total funding request: US$129,700.00** :::