Graph Neural Networks (GNNs) have achieved state-of-the-art performance in solving graph classification tasks. However, most GNN architectures aggregate information from all nodes and edges in a graph, regardless of their relevance to the task at hand, thus hindering the interpretability of their predictions. In contrast to prior work, in this paper we propose a GNN \emph{training} approach that jointly i) finds the most predictive subgraph by removing edges and/or nodes -- -\emph{without making assumptions about the subgraph structure} -- while ii) optimizing the performance of the graph classification task. To that end, we rely on reinforcement learning to solve the resulting bi-level optimization with a reward function based on conformal predictions to account for the current in-training uncertainty of the classifier. Our empirical results on nine different graph classification datasets show that our method competes in performance with baselines while relying on significantly sparser subgraphs, leading to more interpretable GNN-based predictions.

Alan Fern, a professor of computer science at Oregon State University, and a team of researchers have successfully trained a humanoid robot called Digit V3 to stand, walk, pick up a box, and move it from one location to another. Meanwhile, a separate group of researchers from the University of California, Berkeley, have focused on teaching Digit to walk in unfamiliar environments while carrying different loads, without toppling over. Their research is published in a paper in Science Robotics today. Both groups are using an AI technique called sim-to-real reinforcement learning, a burgeoning method of training two-legged robots like Digit. Researchers believe it will lead to more robust, reliable two-legged machines capable of interacting with their surroundings more safely--as well as learning much more quickly.

Choosing an appropriate optimization algorithm is essential to achieving success in optimization challenges. Here we present a new evolutionary algorithm structure that utilizes a reinforcement learning-based agent aimed at addressing these issues. The agent employs a double deep q-network to choose a specific evolutionary operator based on feedback it receives from the environment during optimization. The algorithm's structure contains five single-objective evolutionary algorithm operators. This single-objective structure is transformed into a multi-objective one using the R2 indicator. This indicator serves two purposes within our structure: first, it renders the algorithm multi-objective, and second, provides a means to evaluate each algorithm's performance in each generation to facilitate constructing the reinforcement learning-based reward function. The proposed R2-reinforcement learning multi-objective evolutionary algorithm (R2-RLMOEA) is compared with six other multi-objective algorithms that are based on R2 indicators. These six algorithms include the operators used in R2-RLMOEA as well as an R2 indicator-based algorithm that randomly selects operators during optimization. We benchmark performance using the CEC09 functions, with performance measured by inverted generational distance and spacing. The R2-RLMOEA algorithm outperforms all other algorithms with strong statistical significance (p<0.001) when compared with the average spacing metric across all ten benchmarks.

Policy gradient methods in actor-critic reinforcement learning (RL) have become perhaps the most promising approaches to solving continuous optimal control problems. However, the trial-and-error nature of RL and the inherent randomness associated with solution approximations cause variations in the learned optimal values and policies. This has significantly hindered their successful deployment in real life applications where control responses need to meet dynamic performance criteria deterministically. Here we propose a novel phased actor in actor-critic (PAAC) method, aiming at improving policy gradient estimation and thus the quality of the control policy. Specifically, PAAC accounts for both $Q$ value and TD error in its actor update. We prove qualitative properties of PAAC for learning convergence of the value and policy, solution optimality, and stability of system dynamics. Additionally, we show variance reduction in policy gradient estimation. PAAC performance is systematically and quantitatively evaluated in this study using DeepMind Control Suite (DMC). Results show that PAAC leads to significant performance improvement measured by total cost, learning variance, robustness, learning speed and success rate. As PAAC can be piggybacked onto general policy gradient learning frameworks, we select well-known methods such as direct heuristic dynamic programming (dHDP), deep deterministic policy gradient (DDPG) and their variants to demonstrate the effectiveness of PAAC. Consequently we provide a unified view on these related policy gradient algorithms.

Multi-object transport using multi-robot systems has the potential for diverse practical applications such as delivery services owing to its efficient individual and scalable cooperative transport. However, allocating transportation tasks of objects with unknown weights remains challenging. Moreover, the presence of infeasible tasks (untransportable objects) can lead to robot stoppage (deadlock). This paper proposes a framework for dynamic task allocation that involves storing task experiences for each task in a scalable manner with respect to the number of robots. First, these experiences are broadcasted from the cloud server to the entire robot system. Subsequently, each robot learns the exclusion levels for each task based on those task experiences, enabling it to exclude infeasible tasks and reset its task priorities. Finally, individual transportation, cooperative transportation, and the temporary exclusion of tasks considered infeasible are achieved. The scalability and versatility of the proposed method were confirmed through numerical experiments with an increased number of robots and objects, including unlearned weight objects. The effectiveness of the temporary deadlock avoidance was also confirmed by introducing additional robots within an episode. The proposed method enables the implementation of task allocation strategies that are feasible for different numbers of robots and various transport tasks without prior consideration of feasibility.

The success of collaboration between humans and robots in shared environments relies on the robot's real-time adaptation to human motion. Specifically, in Social Navigation, the agent should be close enough to assist but ready to back up to let the human move freely, avoiding collisions. Human trajectories emerge as crucial cues in Social Navigation, but they are partially observable from the robot's egocentric view and computationally complex to process. We propose the first Social Dynamics Adaptation model (SDA) based on the robot's state-action history to infer the social dynamics. We propose a two-stage Reinforcement Learning framework: the first learns to encode the human trajectories into social dynamics and learns a motion policy conditioned on this encoded information, the current status, and the previous action. Here, the trajectories are fully visible, i.e., assumed as privileged information. In the second stage, the trained policy operates without direct access to trajectories. Instead, the model infers the social dynamics solely from the history of previous actions and statuses in real-time. Tested on the novel Habitat 3.0 platform, SDA sets a novel state of the art (SoA) performance in finding and following humans.

Building embodied AI systems that can follow arbitrary language instructions in any 3D environment is a key challenge for creating general AI. Accomplishing this goal requires learning to ground language in perception and embodied actions, in order to accomplish complex tasks. The Scalable, Instructable, Multiworld Agent (SIMA) project tackles this by training agents to follow free-form instructions across a diverse range of virtual 3D environments, including curated research environments as well as open-ended, commercial video games. Our goal is to develop an instructable agent that can accomplish anything a human can do in any simulated 3D environment. Our approach focuses on language-driven generality while imposing minimal assumptions. Our agents interact with environments in real-time using a generic, human-like interface: the inputs are image observations and language instructions and the outputs are keyboard-and-mouse actions. This general approach is challenging, but it allows agents to ground language across many visually complex and semantically rich environments while also allowing us to readily run agents in new environments. In this paper we describe our motivation and goal, the initial progress we have made, and promising preliminary results on several diverse research environments and a variety of commercial video games.

Advances in computational power and AI have increased interest in reinforcement learning approaches to inventory management. This paper provides a theoretical foundation for these approaches and investigates the benefits of restricting to policy structures that are well-established by decades of inventory theory. In particular, we prove generalization guarantees for learning several well-known classes of inventory policies, including base-stock and (s, S) policies, by leveraging the celebrated Vapnik-Chervonenkis (VC) theory. We apply the concepts of the Pseudo-dimension and Fat-shattering dimension from VC theory to determine the generalizability of inventory policies, that is, the difference between an inventory policy's performance on training data and its expected performance on unseen data. We focus on a classical setting without contexts, but allow for an arbitrary distribution over demand sequences and do not make any assumptions such as independence over time. We corroborate our supervised learning results using numerical simulations. Managerially, our theory and simulations translate to the following insights. First, there is a principle of "learning less is more" in inventory management: depending on the amount of data available, it may be beneficial to restrict oneself to a simpler, albeit suboptimal, class of inventory policies to minimize overfitting errors. Second, the number of parameters in a policy class may not be the correct measure of overfitting error: in fact, the class of policies defined by T time-varying base-stock levels exhibits a generalization error comparable to that of the two-parameter (s, S) policy class. Finally, our research suggests situations in which it could be beneficial to incorporate the concepts of base-stock and inventory position into black-box learning machines, instead of having these machines directly learn the order quantity actions.

Portfolio optimization tasks describe sequential decision problems in which the investor's wealth is distributed across a set of assets. Allocation constraints are used to enforce minimal or maximal investments into particular subsets of assets to control for objectives such as limiting the portfolio's exposure to a certain sector due to environmental concerns. Although methods for constrained Reinforcement Learning (CRL) can optimize policies while considering allocation constraints, it can be observed that these general methods yield suboptimal results. In this paper, we propose a novel approach to handle allocation constraints based on a decomposition of the constraint action space into a set of unconstrained allocation problems. In particular, we examine this approach for the case of two constraints. For example, an investor may wish to invest at least a certain percentage of the portfolio into green technologies while limiting the investment in the fossil energy sector. We show that the action space of the task is equivalent to the decomposed action space, and introduce a new reinforcement learning (RL) approach CAOSD, which is built on top of the decomposition. The experimental evaluation on real-world Nasdaq-100 data demonstrates that our approach consistently outperforms state-of-the-art CRL benchmarks for portfolio optimization.

We present the first study on provably efficient randomized exploration in cooperative multi-agent reinforcement learning (MARL). We propose a unified algorithm framework for randomized exploration in parallel Markov Decision Processes (MDPs), and two Thompson Sampling (TS)-type algorithms, CoopTS-PHE and CoopTS-LMC, incorporating the perturbed-history exploration (PHE) strategy and the Langevin Monte Carlo exploration (LMC) strategy respectively, which are flexible in design and easy to implement in practice. For a special class of parallel MDPs where the transition is (approximately) linear, we theoretically prove that both CoopTS-PHE and CoopTS-LMC achieve a $\widetilde{\mathcal{O}}(d^{3/2}H^2\sqrt{MK})$ regret bound with communication complexity $\widetilde{\mathcal{O}}(dHM^2)$, where $d$ is the feature dimension, $H$ is the horizon length, $M$ is the number of agents, and $K$ is the number of episodes. This is the first theoretical result for randomized exploration in cooperative MARL. We evaluate our proposed method on multiple parallel RL environments, including a deep exploration problem (\textit{i.e.,} $N$-chain), a video game, and a real-world problem in energy systems. Our experimental results support that our framework can achieve better performance, even under conditions of misspecified transition models. Additionally, we establish a connection between our unified framework and the practical application of federated learning.