Наредни састанак Семинара биће одржан онлајн у четвртак, 13. јуна 2024. са почетком у 13 часова.
Предавач: Татјана Давидовић, Mathematical Institute of the Serbian Academy of Sciences and Arts
Наслов предавања: PROOF-OF-USEFUL-WORK CONSENSUS PROTOCOL BASED ON SOLVING REAL-LIFE OPTIMIZATION PROBLEMS
Апстракт: BlockChain (BC) is a distributed database system, popular for its innovative, unsupervised maintaining process: it uses the so-called consensus protocol (CP) to avoid inference of any third party of absolute trust. As the main issues in maintaining BC, security, privacy, consistency, and energy consumption are identified. According to the recent literature, some of these issues can be formulated as Combinatorial Optimization (CO) problems, and this fact motivated us to consider incorporating the CO approaches into the BC. The main goal of this talk is to summarize the results, related to the above mentioned topic, achieved during the realization of AI4TrustBC project (2020-2023).
We proposed a new Combinatorial Optimization Consensus Protocol (COCP), based on Proof-of-Useful-Work (PoU) concept that assumes solving instances of the real-life CO problems. The main advantages of COCP are efficient utilization of computing resources, solving the real-life instances of CO problems, and providing a broad range of incentives for the various BC participants. We enumerate potential benefits of the COCP with respect to practical impacts and savings in power consumption, describe in detail some illustrative examples, and identify several challenges that should be resolved in order to implement a useful, secure, and efficient PoUW consensus protocol.
We have been developing the BC framework that combines the two above-mentioned research fields: BC and CO. It involves some basic steps toward the implementation of COCP. The first challenge to be resolved involves the existence of efficient methods for the underlying CO problems. Due to their complexity, we developed various types of heuristic methods to be utilized in the COCP. Most of these methods are problem-dependent stochastic heuristics or metaheuristics. Their usability within COCP requires careful analysis and estimation of the time necessary to provide solutions of desired quality. We proposed to apply Machine Learning techniques in this phase and to ensure fairness in the mining process. In addition, we considered the incorporation of the existing methods into our framework with an aim to spread the applicability of the proposed COCP to various CO domains. The search for the existing algorithms should be organized within CP as a part of useful work.
The presented research results were published in 2 journal articles and 5 conference papers, and have been obtained in the collaboration with:Project members: Milan Todorović, Dragan Urošević, Tatjana Jakšić-Kruger, Luka Matijević, Đorđe Jovanović (MISANU) Foreign researchers: Dušan Ramljak, Abhay Haridas, Barat Sharma (PSU) Students: Uroš Maleš (ETF), Dragutin Ostojić (PMF KG), Ognjen Nešković, Pavle Sekešan (MF).
Напомена: Предавања се могу пратити на даљину преко странице:
https://miteam.mi.sanu.ac.rs/asset/tkKTsEvjqDmEEDx9a