Stochastic models and evaluation in blockchain and recommendation

Abstract

In this thesis, we study two modeling and evaluation problems and propose novel methods to improve system performance. In the first part, we consider improving throughput and security in a blockchain system. Blockchains based on the celebrated Nakamoto consensus protocol have shown promise in several applications, including cryptocurrencies. However, these blockchains face inherent scalability limits due to the consensus properties of the protocol. In particular, the 'consistency' property demonstrates a tight trade-off between block production speed and the system's security in terms of resisting adversarial attacks. To address this, we propose a novel method, Ironclad, which enhances the blockchain's consistency bound by assigning different weights to randomly selected blocks. We apply our method to the original Nakamoto protocol and rigorously prove that this combination can significantly improve the consistency bound by analyzing the fundamental consensus properties. This improvement enables a much faster block production rate than the original Nakamoto protocol while maintaining the same security guarantee. In the second part, we study how to evaluate and debias a recommendation system. Recommendation systems aim to predict users' feedback on items they have not yet encountered. Confounding bias arises due to the presence of unmeasured variables (e.g., a user's socio-economic status) that can affect both a user's exposure to items and their feedback. Existing methods either (1) make untenable assumptions about these unmeasured variables or (2) directly infer latent confounders from users' exposure histories. However, they cannot guarantee the accurate identification of counterfactual feedback, leading to biased predictions. In this work, we propose a novel method, identifiable deconfounder (iDCF), which utilizes a set of proxy variables (e.g., observed user features) to resolve this non-identification issue. The proposed iDCF is a general deconfounded recommendation framework that applies proximal causal inference to infer unmeasured confounders and accurately identify counterfactual feed-back with theoretical guarantees. Extensive experiments on various real-world and synthetic datasets verify the effectiveness and robustness of our proposed method.

Date of Award	2024
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Jiheng ZHANG (Supervisor)