Privacy protection and nonconvexity are two challenging problems in decentralized optimization and learning involving sensitive data. Despite some recent advances addressing each of the two problems separately, no results have been reported that have theoretical guarantees on both privacy protection and saddle/maximum avoidance in decentralized nonconvex optimization. We propose a new algorithm for decentralized nonconvex optimization that can enable both rigorous differential privacy and saddle/maximum avoiding performance. The new algorithm allows the incorporation of persistent additive noise to enable rigorous differential privacy for data samples, gradients, and intermediate optimization variables without losing provable convergence, and thus circumventing the dilemma of trading accuracy for privacy in differential privacy design. More interestingly, the algorithm is theoretically proven to be able to efficiently { guarantee accuracy by avoiding} convergence to local maxima and saddle points, which has not been reported before in the literature on decentralized nonconvex optimization. The algorithm is efficient in both communication (it only shares one variable in each iteration) and computation (it is encryption-free), and hence is promising for large-scale nonconvex optimization and learning involving high-dimensional optimization parameters. Numerical experiments for both a decentralized estimation problem and an Independent Component Analysis (ICA) problem confirm the effectiveness of the proposed approach.

Adequate strategizing of agents behaviors is essential to solving cooperative MARL problems. One intuitively beneficial yet uncommon method in this domain is predicting agents future behaviors and planning accordingly. Leveraging this point, we propose a two-level hierarchical architecture that combines a novel information-theoretic objective with a trajectory prediction model to learn a strategy. To this end, we introduce a latent policy that learns two types of latent strategies: individual $z_A$, and relational $z_R$ using a modified Graph Attention Network module to extract interaction features. We encourage each agent to behave according to the strategy by conditioning its local $Q$ functions on $z_A$, and we further equip agents with a shared $Q$ function that conditions on $z_R$. Additionally, we introduce two regularizers to allow predicted trajectories to be accurate and rewarding. Empirical results on Google Research Football (GRF) and StarCraft (SC) II micromanagement tasks show that our method establishes a new state of the art being, to the best of our knowledge, the first MARL algorithm to solve all super hard SC II scenarios as well as the GRF full game with a win rate higher than $95\%$, thus outperforming all existing methods. Videos and brief overview of the methods and results are available at: https://sites.google.com/view/hier-strats-marl/home.

Particle dynamics and multi-agent systems provide accurate dynamical models for studying and forecasting the behavior of complex interacting systems. They often take the form of a high-dimensional system of differential equations parameterized by an interaction kernel that models the underlying attractive or repulsive forces between agents. We consider the problem of constructing a data-based approximation of the interacting forces directly from noisy observations of the paths of the agents in time. The learned interaction kernels are then used to predict the agents behavior over a longer time interval. The approximation developed in this work uses a randomized feature algorithm and a sparse randomized feature approach. Sparsity-promoting regression provides a mechanism for pruning the randomly generated features which was observed to be beneficial when one has limited data, in particular, leading to less overfitting than other approaches. In addition, imposing sparsity reduces the kernel evaluation cost which significantly lowers the simulation cost for forecasting the multi-agent systems. Our method is applied to various examples, including first-order systems with homogeneous and heterogeneous interactions, second order homogeneous systems, and a new sheep swarming system.

DCOP algorithms usually rely on interaction graphs to operate. In open and dynamic environments, such methods need to address how this interaction graph is generated and maintained among agents. Existing methods require reconstructing the entire graph upon detecting changes in the environment or assuming that new agents know potential neighbors to facilitate connection. We propose a novel distributed interaction graph construction algorithm to address this problem. The proposed method does not assume a predefined constraint graph and stabilizes after disruptive changes in the environment. We evaluate our approach by pairing it with existing DCOP algorithms to solve several generated dynamic problems. The experiment results show that the proposed algorithm effectively constructs and maintains a stable multi-agent interaction graph for open and dynamic environments.

While Shapley's analysis was originally thought to quantify the worth of human agents in a team, Research in the field of Multi-Agent Systems (MAS) suggests its application is straightforward to every other possible transferable viable pathways to solve complex tasks [1]. In a MAS utility coalitional game that respects the needed mathematical environment, every agent is, in principle, an individual independent properties. of one another with its own characteristics and skills. The field of possible applications of Shapley and Myerson The main idea is that by assigning to each agent a specific subtask analyses or their generalizations is broad. Shapley analysis or according to its perks and hence exploiting a delocalized its suitable generalizations can be applied for instance to estimate control, it is possible to solve a problem more efficiently. The the contributions of basketball players in a match using the human society itself is an example of a MAS since groups of recorded match data and statistics [3]. If the practitioner possesses individuals usually train according to their nature to exercise some information about the connectivity of interactions, specific professions that require different expertise: medical or, e.g., spatial rules of the game that restrict the interaction personnel, firefighters, engineers, etc. When analyzing the behavior among agents, Shapley and Myerson analyses can be used to of agents in a MAS a question arises immediately: according assess the importance of vertices, i.e., agents, in graphs. Recent to a common goal to be reached, which agent is contributing works investigated the Shapley and Myerson analyses of the most, and which are its most important individual transportation networks [4] and bus-holding strategies [5].

Almost all multi-agent reinforcement learning algorithms without communication follow the principle of centralized training with decentralized execution. During centralized training, agents can be guided by the same signals, such as the global state. During decentralized execution, however, agents lack the shared signal. Inspired by viewpoint invariance and contrastive learning, we propose consensus learning for cooperative multi-agent reinforcement learning in this paper. Although based on local observations, different agents can infer the same consensus in discrete space. During decentralized execution, we feed the inferred consensus as an explicit input to the network of agents, thereby developing their spirit of cooperation. Our proposed method can be extended to various multi-agent reinforcement learning algorithms with small model changes. Moreover, we carry out them on some fully cooperative tasks and get convincing results.

In the Approval Participatory Budgeting problem an agent prefers a set of projects $W'$ over $W$ if she approves strictly more projects in $W'$. A set of projects $W$ is in the core, if there is no other set of projects $W'$ and set of agents $K$ that both prefer $W'$ over $W$ and can fund $W'$. It is an open problem whether the core can be empty, even when project costs are uniform. the latter case is known as the multiwinner voting core. We show that in any instance with uniform costs or with at most four projects (and any number of agents), the core is nonempty.

In this paper, we consider the partial gathering problem of mobile agents in synchronous dynamic bidirectional ring networks. When k agents are distributed in the network, the partial gathering problem requires, for a given positive integer g (< k), that agents terminate in a configuration such that either at least g agents or no agent exists at each node. So far, the partial gathering problem has been considered in static graphs. In this paper, we start considering partial gathering in dynamic graphs. As a first step, we consider this problem in 1-interval connected rings, that is, one of the links in a ring may be missing at each time step. In such networks, focusing on the relationship between the values of k and g, we fully characterize the solvability of the partial gathering problem and analyze the move complexity of the proposed algorithms when the problem can be solved. First, we show that the g-partial gathering problem is unsolvable when k <= 2g. Second, we show that the problem can be solved with O(n log g) time and the total number of O(gn log g) moves when 2g + 1 <= k <= 3g - 2. Third, we show that the problem can be solved with O(n) time and the total number of O(kn) moves when 3g - 1 <= k <= 8g - 4. Notice that since k = O(g) holds when 3g - 1 <= k <= 8g - 4, the move complexity O(kn) in this case can be represented also as O(gn). Finally, we show that the problem can be solved with O(n) time and the total number of O(gn) moves when k >= 8g - 3. These results mean that the partial gathering problem can be solved also in dynamic rings when k >= 2g + 1. In addition, agents require a total number of \Omega(gn) moves to solve the partial (resp., total) gathering problem. Thus, when k >= 3g - 1, agents can solve the partial gathering problem with the asymptotically optimal total number of O(gn) moves.

A critical challenge in multi-agent reinforcement learning (MARL) is for multiple agents to efficiently accomplish complex, long-horizon tasks. The agents often have difficulties in cooperating on common goals, dividing complex tasks, and planning through several stages to make progress. We propose to address these challenges by guiding agents with programs designed for parallelization, since programs as a representation contain rich structural and semantic information, and are widely used as abstractions for long-horizon tasks. Specifically, we introduce Efficient Multi-Agent Reinforcement Learning with Parallel Program Guidance (E-MAPP), a novel framework that leverages parallel programs to guide multiple agents to efficiently accomplish goals that require planning over 10+ stages. E-MAPP integrates the structural information from a parallel program, promotes the cooperative behaviors grounded in program semantics, and improves the time efficiency via a task allocator. We conduct extensive experiments on a series of challenging, long-horizon cooperative tasks in the Overcooked environment. Results show that E-MAPP outperforms strong baselines in terms of the completion rate, time efficiency, and zero-shot generalization ability by a large margin.

We address the problem of learning the legitimacy of other agents in a multiagent network when an unknown subset is comprised of malicious actors. We specifically derive results for the case of directed graphs and where stochastic side information, or observations of trust, is available. We refer to this as ``learning trust'' since agents must identify which neighbors in the network are reliable, and we derive a protocol to achieve this. We also provide analytical results showing that under this protocol i) agents can learn the legitimacy of all other agents almost surely, and that ii) the opinions of the agents converge in mean to the true legitimacy of all other agents in the network. Lastly, we provide numerical studies showing that our convergence results hold in practice for various network topologies and variations in the number of malicious agents in the network.