Decision models based on sequence modeling have become prevalent in the field of offline reinforcement learning. However, existing approaches such as Decision Transformer and Trajectory Transformer suffer from poor data efficiency. One important reason is that they fail to extract useful information from potentially high-dimensional and noisy states. To resolve this issue, we propose a data-efficient decision sequence modeling method called Data Efficient Decision Sequence Model (DEDS), which dynamically identifies and filters out task-irrelevant state features for more compact state representations. Specifically, DEDS employs a state mask model guided by the bisimulation metric to ensure that only the most task-relevant state features are preserved for decision-making. Extensive experiments on various environments demonstrate that DEDS outperforms existing offline RL methods and achieves a significantly high data efficiency, especially in tasks with high-dimensional and complex state spaces.
Graph classification is crucial for diverse real-world applications, such as drug discovery and social network analysis. Reinforcement learning (RL) has been introduced to graph tasks for its advantage in sequential decision-making and exploration. However, existing RL-based methods face challenges: improper state/action definitions can trap policies in local optima, while contrastive learning approaches often rely on random sampling, struggling to guarantee effective negative samples. To address these issues, this paper proposes a graph classification framework with stepwise optimization embedded with multi-agent collaboration. Specifically, we first optimize the pooling ratio adjustment agent, adopt more advanced reinforcement learning algorithms, and design richer and more informative action spaces and reward functions to support the agent in conducting more sufficient environmental exploration, thereby approximating the global optimal policy. Additionally, we introduce a sampling agent that adaptively optimizes its sampling strategy during the contrastive learning process, aiming to generate more distinguishable hard negative samples. Extensive experiments on six classic benchmark datasets demonstrate that the proposed method achieves superior performance in graph classification tasks, significantly outperforming existing baseline methods and exhibiting strong competitiveness.
Artificial intelligence relies on access to diverse, high-quality data, yet in sensitive domains such as healthcare, public services, and finance, access remains severely constrained. A key reason is the dominance of rigid compliance mechanisms-fixed rules, static contracts, and manual checks-that cannot adapt to changing contexts, evolving regulations, or diverse stakeholder needs. These inflexible systems slow collaboration, discourage data sharing, and lock away valuable datasets, exacerbating the problem of data scarcity and limiting the development of ethical and effective AI. In this perspective paper, we call for a reimagined architecture of compliant and trustworthy data markets powered by intelligent agents. These agents act on behalf of data providers, data consumers, regulators, and individuals as data subjects; they can negotiate access, assess trust, and enforce legal and ethical requirements in real time. By embedding such capabilities directly into the market infrastructure, we can move beyond one-size-fits-all compliance toward context-aware, fine-grained, and adaptive governance that makes data sharing both flexible and lawful. We propose a research roadmap built on four pillars: privacy-aware and adaptive infrastructures; dynamic and relational trust; automated multi-issue negotiation; and adaptive compliance with legal foresight. We also explore our evaluation plan that will allow us to test the pillars. A healthcare use case illustrates how this approach could address rigidity and data scarcity, enabling innovation while protecting rights. Finally, we position our perspective within relevant fields such as artificial intelligence and multi-agent systems and law, policy, and governance research and show how each contributes to the design of compliant and trustworthy data markets.
Heartbeat synchronization, where agents emit short, periodic, near-simultaneous flashes, is achieved as an emergent behavior of the multi-agent system through a novel decentralized protocol. The proposed protocol operates in large multi-agent systems without requiring agents to share a global frequency or phase. Agents interact iteratively with randomly selected peers, pausing for a random interval before triggering each interaction. Starting from random states, agents update their states through pairwise interactions and continuously emit light when their state meets a condition. Two emergent properties of these flashes are studied: the flash period, which is the interval between consecutive flashes, and the flash duration, which is the duration of each flash. Experiments conducted using a custom-built simulator provide strong empirical evidence that the protocol induces a sufficiently regular flash period and a sufficiently short flash duration to enable effective heartbeat synchronization in real-world applications.
We consider a stable marriage problem on a network, where two groups of agents (men and women) form a network and each only knows their neighbors. We aim to design a novel mechanism that motivates agents to invite their neighbors to join in the matching game if they are not already in it. The difficulty is that invitees may hurt the inviters, which occurs if we apply the standard Deferred Acceptance (DA) mechanism. To induce mutual invitations, one idea is to add restrictions on DA and let each agent only propose to their neighbors, which ensures everyone is eager to invite all their neighbors. The drawback is that agents can never match with non-neighbors. In this paper, we propose Dynamic Deferred Acceptance (DDA) to enable matching with non-neighbors by dynamic alliance construction and sharing process. We demonstrate impossibility results between stability and other desirable properties, and our mechanism achieves the strongest stability in the networks.
