Cooperative multi-agent systems can be naturally used to model many real world problems, such as network packet routing and the coordination of autonomous vehicles. There is a great need for new reinforcement learning methods that can efficiently learn decentralised policies for such systems. To this end, we propose a new multi-agent actor-critic method called counterfactual multi-agent (COMA) policy gradients. COMA uses a centralised critic to estimate the Q-function and decentralised actors to optimise the agents' policies. In addition, to address the challenges of multi-agent credit assignment, it uses a counterfactual baseline that marginalises out a single agent's action, while keeping the other agents' actions fixed. COMA also uses a critic representation that allows the counterfactual baseline to be computed efficiently in a single forward pass. We evaluate COMA in the testbed of StarCraft unit micromanagement, using a decentralised variant with significant partial observability. COMA significantly improves average performance over other multi-agent actor-critic methods in this setting, and the best performing agents are competitive with state-of-the-art centralised controllers that get access to the full state.

We propose a multiagent distributed actor-critic algorithm for multitask reinforcement learning (MRL), named Diff-DAC. The agents are connected, forming a (possibly sparse) network. Each agent is assigned a task and has access to data from this local task only. During the learning process, the agents are able to communicate some parameters to their neighbors. Since the agents incorporate their neighbors' parameters into their own learning rules, the information is diffused across the network, and they can learn a common policy that generalizes well across all tasks. Diff-DAC is scalable since the computational complexity and communication overhead per agent grow with the number of neighbors, rather than with the total number of agents. Moreover, the algorithm is fully distributed in the sense that agents self-organize, with no need for coordinator node. Diff-DAC follows an actor-critic scheme where the value function and the policy are approximated with deep neural networks, being able to learn expressive policies from raw data. As a by-product of Diff-DAC's derivation from duality theory, we provide novel insights into the standard actor-critic framework, showing that it is actually an instance of the dual ascent method to approximate the solution of a linear program. Experiments illustrate the performance of the algorithm in the cart-pole, inverted pendulum, and swing-up cart-pole environments.

Mindfire, a new foundation with the goal of "decoding the mind" to help develop true artificial intelligence (AI) is launching November 17th in Zurich, Switzerland. Futurism spoke with the founder of Starmind and president of the foundation, Pascal Kaufmann to learn more about its goals and the path to reach them. "We cannot achieve True AI until we understand actual intelligence. Intelligence has evolved as a means of nature to successfully guide us through an ever-changing environment. This gave rise to behavior, emotions, and consciousness. These critical factors must be taken into account in how we develop AI. This is the purpose of the Mindfire Foundation," he explains.

In this paper, we consider a class of finite-sum convex optimization problems defined over a distributed multiagent network with $m$ agents connected to a central server. In particular, the objective function consists of the average of $m$ ($\ge 1$) smooth components associated with each network agent together with a strongly convex term. Our major contribution is to develop a new randomized incremental gradient algorithm, namely random gradient extrapolation method (RGEM), which does not require any exact gradient evaluation even for the initial point, but can achieve the optimal ${\cal O}(\log(1/\epsilon))$ complexity bound in terms of the total number of gradient evaluations of component functions to solve the finite-sum problems. Furthermore, we demonstrate that for stochastic finite-sum optimization problems, RGEM maintains the optimal ${\cal O}(1/\epsilon)$ complexity (up to a certain logarithmic factor) in terms of the number of stochastic gradient computations, but attains an ${\cal O}(\log(1/\epsilon))$ complexity in terms of communication rounds (each round involves only one agent). It is worth noting that the former bound is independent of the number of agents $m$, while the latter one only linearly depends on $m$ or even $\sqrt m$ for ill-conditioned problems. To the best of our knowledge, this is the first time that these complexity bounds have been obtained for distributed and stochastic optimization problems. Moreover, our algorithms were developed based on a novel dual perspective of Nesterov's accelerated gradient method.

Much research has been done to apply auctions, markets, and negotiation mechanisms to solve the multiagent task allocation problem. However, there has been very little work on human-agent group task allocation. We believe that the notion of bounty hunting has good properties for human-agent group interaction in dynamic task allocation problems. We use previous experimental results comparing bounty hunting with auction-like methods to argue why it would be particularly adept at handling scenarios with unreliable collaborators and unexpectedly hard tasks: scenarios we believe highlight difficulties involved in working with humans collaborators.

We argue that social simulation can help us understand social justice issues. In particular, modeling certain social dynamics within computational systems can be used to creatively explore and better understand the social and identity dynamics of oppression. Writing theories of oppression in code forces us to explicate everything, and question what we leave out or what we can’t account for. As an early step in this direction, we present an in-progress social simulation of group discussion in activist meetings, developed in the already-existing AI system, Ensemble. Through this minimal, highly constrained social arena, we can explore wide-reaching phenomena like privilege, intersectionality, and power dynamics in nonhierarchical groups, but in a way that’s grounded in concrete, person-to-person interactions. We propose that this kind of social simulation can aid in the process of unlearning hegemonic ways of being, and imagining liberatory alternatives.

In this paper, we investigate how to learn to control a group of cooperative agents with limited sensing capabilities such as robot swarms. The agents have only very basic sensor capabilities, yet in a group they can accomplish sophisticated tasks, such as distributed assembly or search and rescue tasks. Learning a policy for a group of agents is difficult due to distributed partial observability of the state. Here, we follow a guided approach where a critic has central access to the global state during learning, which simplifies the policy evaluation problem from a reinforcement learning point of view. For example, we can get the positions of all robots of the swarm using a camera image of a scene. This camera image is only available to the critic and not to the control policies of the robots. We follow an actor-critic approach, where the actors base their decisions only on locally sensed information. In contrast, the critic is learned based on the true global state. Our algorithm uses deep reinforcement learning to approximate both the Q-function and the policy. The performance of the algorithm is evaluated on two tasks with simple simulated 2D agents: 1) finding and maintaining a certain distance to each others and 2) locating a target.

We consider the well-studied cake cutting problem in which the goal is to find an envy-free allocation based on queries from $n$ agents. The problem has received attention in computer science, mathematics, and economics. It has been a major open problem whether there exists a discrete and bounded envy-free protocol. We resolve the problem by proposing a discrete and bounded envy-free protocol for any number of agents. The maximum number of queries required by the protocol is $n^{n^{n^{n^{n^n}}}}$. We additionally show that even if we do not run our protocol to completion, it can find in at most $n^3{(n^2)}^n$ queries a partial allocation of the cake that achieves proportionality (each agent gets at least $1/n$ of the value of the whole cake) and envy-freeness. Finally we show that an envy-free partial allocation can be computed in at most $n^3{(n^2)}^n$ queries such that each agent gets a connected piece that gives the agent at least $1/(3n)$ of the value of the whole cake.

Using the previously developed concepts of semantic spacetime, I explore the interpretation of knowledge representations, and their structure, as a semantic system, within the framework of promise theory. By assigning interpretations to phenomena, from observers to observed, we may approach a simple description of knowledge-based functional systems, with direct practical utility. The focus is especially on the interpretation of concepts, associative knowledge, and context awareness. The inference seems to be that most if not all of these concepts emerge from purely semantic spacetime properties, which opens the possibility for a more generalized understanding of what constitutes a learning, or even `intelligent' system. Some key principles emerge for effective knowledge representation: 1) separation of spacetime scales, 2) the recurrence of four irreducible types of association, by which intent propagates: aggregation, causation, cooperation, and similarity, 3) the need for discrimination of identities (discrete), which is assisted by distinguishing timeline simultaneity from sequential events, and 4) the ability to learn (memory). It is at least plausible that emergent knowledge abstraction capabilities have their origin in basic spacetime structures. These notes present a unified view of mostly well-known results; they allow us to see information models, knowledge representations, machine learning, and semantic networking (transport and information base) in a common framework. The notion of `smart spaces' thus encompasses artificial systems as well as living systems, across many different scales, e.g. smart cities and organizations.

In this work, we design and evaluate a computational learning model that enables a human-robot team to co-develop joint strategies for performing novel tasks that require coordination. The joint strategies are learned through "perturbation training," a human team-training strategy that requires team members to practice variations of a given task to help their team generalize to new variants of that task. We formally define the problem of human-robot perturbation training and develop and evaluate the first end-to-end framework for such training, which incorporates a multi-agent transfer learning algorithm, human-robot co-learning framework and communication protocol. Our transfer learning algorithm, Adaptive Perturbation Training (AdaPT), is a hybrid of transfer and reinforcement learning techniques that learns quickly and robustly for new task variants. We empirically validate the benefits of AdaPT through comparison to other hybrid reinforcement and transfer learning techniques aimed at transferring knowledge from multiple source tasks to a single target task. We also demonstrate that AdaPT's rapid learning supports live interaction between a person and a robot, during which the human-robot team trains to achieve a high level of performance for new task variants. We augment AdaPT with a co-learning framework and a computational bi-directional communication protocol so that the robot can co-train with a person during live interaction. Results from large-scale human subject experiments (n=48) indicate that AdaPT enables an agent to learn in a manner compatible with a human's own learning process, and that a robot undergoing perturbation training with a human results in a high level of team performance. Finally, we demonstrate that human-robot training using AdaPT in a simulation environment produces effective performance for a team incorporating an embodied robot partner.