arXiv subjects

Zhixuan Fang

Publications and source records attributed to Zhixuan Fang.

At least 19 recordsLinked to original sources

arXiv2026

CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents

Current Mixture-of-Agents (MoA) paradigms generally treat query routing and agent fine-tuning as separate processes, limiting their ability to respond to evolving agent capabilities. This disconnect prevents routing strategies from adapting to evolving agent capabilities during post-training and prevents agents from achieving synergistic data-driven specialization. To resolve this, we introduce CERA-MoA (Co-Evolving Router with continually learning Agents for Mixture-of-Agents), an iterative reinforcement learning framework where the dynamic router and independent agent policies co-evolve. We design a predictive familiarity estimator that leverages mid-layer hidden states to evaluate semantic competence among agents, avoiding the overhead of full rollouts. Based on these familiarity scores, a cumulative-threshold adaptive routing mechanism dynamically activates a tailored minimal agent subset, achieving a trade-off between task performance and efficiency. By proactively allocating targeted training samples to agents based on their evolving competence, CERA-MoA promotes capability differentiation. Extensive experiments across various domains demonstrate that CERA-MoA outperforms state-of-the-art static-agent routing and fix-workflow fine-tuning baselines.

cs.AI↗

arXiv2026

Clownfish: Scaling DAG-based BFT Consensus via Sparse Edges

Directed Acyclic Graph (DAG) based BFT protocols have demonstrated the capability to achieve significantly high throughput in practice. Recent advancements focused on minimizing the good-case latency of these protocols, approaching the theoretical lower bound. However, the high communication complexity inherent in existing DAG-based protocols limits their scalability. This primarily arises because each vertex in the DAG must include a linear number of edges (references) to vertices from previous rounds. We present Clownfish, a partially synchronous DAG-based BFT protocol designed to address the scalability bottleneck. Clownfish achieves lower communication complexity by selectively reducing the number of edges in DAG vertices. When using a communication-optimal consistent broadcast, Clownfish attains quadratic total communication complexity per round, outperforming prior DAG-based protocols. Clownfish also reduces the additional latency in failure cases by optimizing the round advancement rule. Additionally, Clownfish supports multiple leaders per round to reduce average latency while maintaining its lower communication complexity. Our experimental evaluation demonstrates that Clownfish provides significantly better scalability than existing DAG-based protocols.

cs.DC↗

arXiv2026

Proof-of-Learning with Incentive Security

Most concurrent blockchain systems rely heavily on the Proof-of-Work (PoW) or Proof-of-Stake (PoS) mechanisms for decentralized consensus and security assurance. However, the substantial energy expenditure stemming from computationally intensive yet meaningless tasks has raised considerable concerns surrounding traditional PoW approaches, The PoS mechanism, while free of energy consumption, is subject to security and economic issues. Addressing these issues, the paradigm of Proof-of-Useful-Work (PoUW) seeks to employ challenges of practical significance as PoW, thereby imbuing energy consumption with tangible value. While previous efforts in Proof of Learning (PoL) explored the utilization of deep learning model training SGD tasks as PoUW challenges, recent research has revealed its vulnerabilities to adversarial attacks and the theoretical hardness in crafting a byzantine-secure PoL mechanism. In this paper, we introduce the concept of incentive-security that incentivizes rational provers to behave honestly for their best interest, bypassing the existing hardness to design a PoL mechanism with computational efficiency, a provable incentive-security guarantee and controllable difficulty. Particularly, our work is secure against two attacks, and also improves the computational overhead from $Θ(1)$ to $O(\frac{\log E}{E})$. Furthermore, while most recent research assumes trusted problem providers and verifiers, our design also guarantees frontend incentive-security even when problem providers are untrusted, and verifier incentive-security that bypasses the Verifier's Dilemma. By incorporating ML training into blockchain consensus mechanisms with provable guarantees, our research not only proposes an eco-friendly solution to blockchain systems, but also provides a proposal for a completely decentralized computing power market in the new AI age.

cs.CR↗

arXiv2026

The Power of Information for Intermediate States in Contract Design

In the conventional principal-agent problem, a principal delegates a task to an agent and formulates a contract to incentivize the agent's actions on behalf of the principal. However, this framework overlooks the information that is possibly available during the delegation process in some scenarios. To address this limitation, we propose a novel model that incorporates multiple intermediate states to capture such information revealed during the delegation. Furthermore, to evaluate the impact of the information embedded in these intermediate states, we introduce two distinct contracts: the pay-halfway contract, which provides payments based not only on final outcomes but also on intermediate states, and the terminate-halfway contract, which allows the principal to terminate the delegation process upon encountering undesirable intermediate states. This leads to the question of whether and how these contract types can leverage intermediate-state information? In particular, we ask: Can these contract types outperform standard contracts, and if so, when and to what extent? We answer the first question affirmatively and provide several important insights regarding the second, shedding light on the circumstances in which intermediate-state-aware contracts yield substantial advantages.

cs.GT↗

arXiv2025

Learning with Limited Shared Information in Multi-agent Multi-armed Bandit

Multi-agent multi-armed bandit (MAMAB) is a classic collaborative learning model and has gained much attention in recent years. However, existing studies do not consider the case where an agent may refuse to share all her information with others, e.g., when some of the data contains personal privacy. In this paper, we propose a novel limited shared information multi-agent multi-armed bandit (LSI-MAMAB) model in which each agent only shares the information that she is willing to share, and propose the Balanced-ETC algorithm to help multiple agents collaborate efficiently with limited shared information. Our analysis shows that Balanced-ETC is asymptotically optimal and its average regret (on each agent) approaches a constant when there are sufficient agents involved. Moreover, to encourage agents to participate in this collaborative learning, an incentive mechanism is proposed to make sure each agent can benefit from the collaboration system. Finally, we present experimental results to validate our theoretical results.

cs.LG↗

arXiv2025

Efficient and Optimal Policy Gradient Algorithm for Corrupted Multi-armed Bandits

In this paper, we consider the stochastic multi-armed bandits problem with adversarial corruptions, where the random rewards of the arms are partially modified by an adversary to fool the algorithm. We apply the policy gradient algorithm SAMBA to this setting, and show that it is computationally efficient, and achieves a state-of-the-art $O(K\log T/Δ) + O(C/Δ)$ regret upper bound, where $K$ is the number of arms, $C$ is the unknown corruption level, $Δ$ is the minimum expected reward gap between the best arm and other ones, and $T$ is the time horizon. Compared with the best existing efficient algorithm (e.g., CBARBAR), whose regret upper bound is $O(K\log^2 T/Δ) + O(C)$, we show that SAMBA reduces one $\log T$ factor in the regret bound, while maintaining the corruption-dependent term to be linear with $C$. This is indeed asymptotically optimal. We also conduct simulations to demonstrate the effectiveness of SAMBA, and the results show that SAMBA outperforms existing baselines.

cs.LG↗

arXiv2024

Cost-efficient Crowdsourcing for Span-based Sequence Labeling: Worker Selection and Data Augmentation

This paper introduces a novel crowdsourcing worker selection algorithm, enhancing annotation quality and reducing costs. Unlike previous studies targeting simpler tasks, this study contends with the complexities of label interdependencies in sequence labeling. The proposed algorithm utilizes a Combinatorial Multi-Armed Bandit (CMAB) approach for worker selection, and a cost-effective human feedback mechanism. The challenge of dealing with imbalanced and small-scale datasets, which hinders offline simulation of worker selection, is tackled using an innovative data augmentation method termed shifting, expanding, and shrinking (SES). Rigorous testing on CoNLL 2003 NER and Chinese OEI datasets showcased the algorithm's efficiency, with an increase in F1 score up to 100.04% of the expert-only baseline, alongside cost savings up to 65.97%. The paper also encompasses a dataset-independent test emulating annotation evaluation through a Bernoulli distribution, which still led to an impressive 97.56% F1 score of the expert baseline and 59.88% cost savings. Furthermore, our approach can be seamlessly integrated into Reinforcement Learning from Human Feedback (RLHF) systems, offering a cost-effective solution for obtaining human feedback.

cs.CL↗

arXiv2024

The Earth is Flat because...: Investigating LLMs' Belief towards Misinformation via Persuasive Conversation

Large language models (LLMs) encapsulate vast amounts of knowledge but still remain vulnerable to external misinformation. Existing research mainly studied this susceptibility behavior in a single-turn setting. However, belief can change during a multi-turn conversation, especially a persuasive one. Therefore, in this study, we delve into LLMs' susceptibility to persuasive conversations, particularly on factual questions that they can answer correctly. We first curate the Farm (i.e., Fact to Misinform) dataset, which contains factual questions paired with systematically generated persuasive misinformation. Then, we develop a testing framework to track LLMs' belief changes in a persuasive dialogue. Through extensive experiments, we find that LLMs' correct beliefs on factual knowledge can be easily manipulated by various persuasive strategies.

cs.CL↗

arXiv2024

RL-CFR: Improving Action Abstraction for Imperfect Information Extensive-Form Games with Reinforcement Learning

Effective action abstraction is crucial in tackling challenges associated with large action spaces in Imperfect Information Extensive-Form Games (IIEFGs). However, due to the vast state space and computational complexity in IIEFGs, existing methods often rely on fixed abstractions, resulting in sub-optimal performance. In response, we introduce RL-CFR, a novel reinforcement learning (RL) approach for dynamic action abstraction. RL-CFR builds upon our innovative Markov Decision Process (MDP) formulation, with states corresponding to public information and actions represented as feature vectors indicating specific action abstractions. The reward is defined as the expected payoff difference between the selected and default action abstractions. RL-CFR constructs a game tree with RL-guided action abstractions and utilizes counterfactual regret minimization (CFR) for strategy derivation. Impressively, it can be trained from scratch, achieving higher expected payoff without increased CFR solving time. In experiments on Heads-up No-limit Texas Hold'em, RL-CFR outperforms ReBeL's replication and Slumbot, demonstrating significant win-rate margins of $64\pm 11$ and $84\pm 17$ mbb/hand, respectively.

cs.GT↗

arXiv2024

Tempo: Confidentiality Preservation in Cloud-Based Neural Network Training

Cloud deep learning platforms provide cost-effective deep neural network (DNN) training for customers who lack computation resources. However, cloud systems are often untrustworthy and vulnerable to attackers, leading to growing concerns about model privacy. Recently, researchers have sought to protect data privacy in deep learning by leveraging CPU trusted execution environments (TEEs), which minimize the use of cryptography, but existing works failed to simultaneously utilize the computational resources of GPUs to assist in training and prevent model leakage. This paper presents Tempo, the first cloud-based deep learning system that cooperates with TEE and distributed GPUs for efficient DNN training with model confidentiality preserved. To tackle the challenge of preserving privacy while offloading linear algebraic operations from TEE to GPUs for efficient batch computation, we introduce a customized permutation-based obfuscation algorithm to blind both inputs and model parameters. An optimization mechanism that reduces encryption operations is proposed for faster weight updates during backpropagation to speed up training. We implement Tempo and evaluate it with both training and inference for two prevalent DNNs. Empirical results indicate that Tempo outperforms baselines and offers sufficient privacy protection.

cs.CR↗

arXiv2023

Information Disclosure under Competition in Sharing Systems

Sharing systems have facilitated the redistribution of underused resources by providing convenient online marketplaces for individual sellers and buyers. However, sellers in these systems may not fully disclose the information of their shared commodities, due to strategic behaviors or privacy concerns. Sellers' strategic information disclosure significantly affects buyers' user experiences and systems' reputation. This paper presents the first analytical study on information disclosure and pricing of competing sellers in sharing systems. In particular, we propose a two-stage game framework to capture sellers' strategic behaviors and buyers' decisions. Although the optimization problem is challenging due to sellers' non-convex and non-monotonic objectives, we completely characterize the complex market equilibria by decomposing it into several tractable subproblems. We demonstrate that full disclosure by all sellers or non-disclosure by all sellers will both lead to intense price competition. The former all-disclosure case is never an equilibrium even when all sellers have good commodity qualities and low privacy costs, while the latter non-disclosure case can be an equilibrium under which all sellers get zero profit. We also reveal several critical factors that affect sellers' information disclosure. Interestingly, sellers' sharing capacity limitation and buyers' estimation biases encourage information disclosure as they mitigate sellers' competition.

cs.GT↗

arXiv2023

SAKSHI: Decentralized AI Platforms

Large AI models (e.g., Dall-E, GPT4) have electrified the scientific, technological and societal landscape through their superhuman capabilities. These services are offered largely in a traditional web2.0 format (e.g., OpenAI's GPT4 service). As more large AI models proliferate (personalizing and specializing to a variety of domains), there is a tremendous need to have a neutral trust-free platform that allows the hosting of AI models, clients receiving AI services efficiently, yet in a trust-free, incentive compatible, Byzantine behavior resistant manner. In this paper we propose SAKSHI, a trust-free decentralized platform specifically suited for AI services. The key design principles of SAKSHI are the separation of the data path (where AI query and service is managed) and the control path (where routers and compute and storage hosts are managed) from the transaction path (where the metering and billing of services are managed over a blockchain). This separation is enabled by a "proof of inference" layer which provides cryptographic resistance against a variety of misbehaviors, including poor AI service, nonpayment for service, copying of AI models. This is joint work between multiple universities (Princeton University, University of Illinois at Urbana-Champaign, Tsinghua University, HKUST) and two startup companies (Witness Chain and Eigen Layer).

cs.CR↗

arXiv2023

MISO: Legacy-compatible Privacy-preserving Single Sign-on using Trusted Execution Environments

Single sign-on (SSO) allows users to authenticate to third-party applications through a central identity provider. Despite their wide adoption, deployed SSO systems suffer from privacy problems such as user tracking by the identity provider. While numerous solutions have been proposed by academic papers, none were adopted because they require modifying identity providers, a significant adoption barrier in practice. Solutions do get deployed, however, fail to eliminate major privacy issues. Leveraging Trusted Execution Environments (TEEs), we propose MISO, the first privacy-preserving SSO system that is completely compatible with existing identity providers (such as Google and Facebook). This means MISO can be easily integrated into existing SSO ecosystem today and benefit end users. MISO also enables new functionality that standard SSO cannot offer: MISO allows users to leverage multiple identity providers in a single SSO workflow, potentially in a threshold fashion, to better protect user accounts. We fully implemented MISO based on Intel SGX. Our evaluation shows that MISO can handle high user concurrency with practical performance.

cs.CR↗

arXiv2023

Continuous Mean-Covariance Bandits

Existing risk-aware multi-armed bandit models typically focus on risk measures of individual options such as variance. As a result, they cannot be directly applied to important real-world online decision making problems with correlated options. In this paper, we propose a novel Continuous Mean-Covariance Bandit (CMCB) model to explicitly take into account option correlation. Specifically, in CMCB, there is a learner who sequentially chooses weight vectors on given options and observes random feedback according to the decisions. The agent's objective is to achieve the best trade-off between reward and risk, measured with option covariance. To capture different reward observation scenarios in practice, we consider three feedback settings, i.e., full-information, semi-bandit and full-bandit feedback. We propose novel algorithms with optimal regrets (within logarithmic factors), and provide matching lower bounds to validate their optimalities. The experimental results also demonstrate the superiority of our algorithms. To the best of our knowledge, this is the first work that considers option correlation in risk-aware bandits and explicitly quantifies how arbitrary covariance structures impact the learning performance. The novel analytical techniques we developed for exploiting the estimated covariance to build concentration and bounding the risk of selected actions based on sampling strategy properties can likely find applications in other bandit analysis and be of independent interests.

cs.LG↗

arXiv2022

Crowdsourcing Work as Mining: A Decentralized Computation and Storage Paradigm

Proof-of-Work (PoW) consensus mechanism is popular among current blockchain systems, which leads to an increasing concern about the tremendous waste of energy due to massive meaningless computation. To address this issue, we propose a novel and energy-efficient blockchain system, CrowdMine, which exploits useful crowdsourcing computation to achieve decentralized consensus. CrowdMine solves user-proposed computing tasks and utilizes the computation committed to the task solving process to secure decentralized on-chain storage. With our designed ``Proof of Crowdsourcing Work'' (PoCW) protocol, our system provides an efficient paradigm for computation and storage in a trustless and decentralized environment. We further show that the system can defend against potential attacks on blockchain, including the short-term 51\% attack, the problem-constructing attack, and the solution-stealing attack. We also implement the system with 40 distributed nodes to demonstrate its performance and robustness. To the best of our knowledge, this is the first system that enables decentralized Proof of Useful Work (PoUW) with general user-proposed tasks posted in a permissionless and trustless network.

cs.DC↗

arXiv2022

SoK: Play-to-Earn Projects

Play-to-earn is one of the prospective categories of decentralized applications. The play-to-earn projects combine blockchain technology with entertaining games and finance, attracting various participants. While huge amounts of capital have been poured into these projects, the new crypto niche is considered controversial, and the traditional gaming industry is hesitant to embrace blockchain technology. In addition, there is little systematic research on these projects. In this paper, we delineate play-to-earn projects in terms of economic & governance models and implementation and analyze how blockchain technology can benefit these projects by providing system robustness, transparency, composability, and decentralized governance. We begin by identifying the participants and characterizing the tokens, which are products of composability. We then summarize the roadmap and governance model to exposit there is a transition from centralized governance to decentralized governance. We also classify the implementation of the play-to-earn projects with different extents of robustness and transparency. Finally, we discuss the security & societal challenges for future research in terms of possible attacks, the economics of tokens, and governance.

cs.CR↗

arXiv2022

Effective Multi-User Delay-Constrained Scheduling with Deep Recurrent Reinforcement Learning

Multi-user delay constrained scheduling is important in many real-world applications including wireless communication, live streaming, and cloud computing. Yet, it poses a critical challenge since the scheduler needs to make real-time decisions to guarantee the delay and resource constraints simultaneously without prior information of system dynamics, which can be time-varying and hard to estimate. Moreover, many practical scenarios suffer from partial observability issues, e.g., due to sensing noise or hidden correlation. To tackle these challenges, we propose a deep reinforcement learning (DRL) algorithm, named Recurrent Softmax Delayed Deep Double Deterministic Policy Gradient ($\mathtt{RSD4}$), which is a data-driven method based on a Partially Observed Markov Decision Process (POMDP) formulation. $\mathtt{RSD4}$ guarantees resource and delay constraints by Lagrangian dual and delay-sensitive queues, respectively. It also efficiently tackles partial observability with a memory mechanism enabled by the recurrent neural network (RNN) and introduces user-level decomposition and node-level merging to ensure scalability. Extensive experiments on simulated/real-world datasets demonstrate that $\mathtt{RSD4}$ is robust to system dynamics and partially observable environments, and achieves superior performances over existing DRL and non-DRL-based methods.

cs.LG↗

arXiv2022

TIPS: Transaction Inclusion Protocol with Signaling in DAG-based Blockchain

Directed Acyclic Graph (DAG) is a popular approach to achieve scalability of blockchain networks. Due to its high efficiency in data communication and great scalability, DAG has been widely adopted in many applications such as Internet of Things (IoT) and Decentralized Finance (DeFi). DAG-based blockchain, nevertheless, faces the key challenge of transaction inclusion collision due to the high concurrency and the network delay. Particularly, the transaction inclusion collision in DAG-based blockchain leads to the revenue and throughput dilemmas, which would greatly degrade the system performance. In this paper, we propose "TIPS", the Transaction Inclusion Protocol with Signaling, which broadcasts a signal indicating the transactions in the block. We show that with the prompt broadcast of a signal, TIPS substantially reduces the transaction collision and thus resolves these dilemmas. Moreover, we show that TIPS can defend against both the denial-of-service and the delay-of-service attacks. We also conduct intensive experiments to demonstrate the superior performance of the proposed protocol.

cs.NI↗