There is considerable interest in designing meta-reinforcement learning (meta-RL) algorithms, which enable autonomous agents to adapt new tasks from small amount of experience. In meta-RL, the specification (such as reward function) of current task is hidden from the agent. In addition, states are hidden within each task owing to sensor noise or limitations in realistic environments. Therefore, the meta-RL agent faces the challenge of specifying both the hidden task and states based on small amount of experience. To address this, we propose estimating disentangled belief about task and states, leveraging an inductive bias that the task and states can be regarded as global and local features of each task. Specifically, we train a hierarchical state-space model (HSSM) parameterized by deep neural networks as an environment model, whose global and local latent variables correspond to task and states, respectively. Because the HSSM does not allow analytical computation of posterior distribution, i.e., belief, we employ amortized inference to approximate it. After the belief is obtained, we can augment observations of a model-free policy with the belief to efficiently train the policy. Moreover, because task and state information are factorized and interpretable, the downstream policy training is facilitated compared with the prior methods that did not consider the hierarchical nature. Empirical validations on a GridWorld environment confirm that the HSSM can separate the hidden task and states information. Then, we compare the meta-RL agent with the HSSM to prior meta-RL methods in MuJoCo environments, and confirm that our agent requires less training data and reaches higher final performance.
Our team is proposing to run a full-scale energy demand response experiment in an office building. Although this is an exciting endeavor which will provide value to the community, collecting training data for the reinforcement learning agent is costly and will be limited. In this work, we apply a meta-learning architecture to warm start the experiment with simulated tasks, to increase sample efficiency. We present results that demonstrate a similar a step up in complexity still corresponds with better learning.
Recurrent meta reinforcement learning (meta-RL) agents are agents that employ a recurrent neural network (RNN) for the purpose of "learning a learning algorithm". After being trained on a pre-specified task distribution, the learned weights of the agent's RNN are said to implement an efficient learning algorithm through their activity dynamics, which allows the agent to quickly solve new tasks sampled from the same distribution. However, due to the black-box nature of these agents, the way in which they work is not yet fully understood. In this study, we shed light on the internal working mechanisms of these agents by reformulating the meta-RL problem using the Partially Observable Markov Decision Process (POMDP) framework. We hypothesize that the learned activity dynamics is acting as belief states for such agents. Several illustrative experiments suggest that this hypothesis is true, and that recurrent meta-RL agents can be viewed as agents that learn to act optimally in partially observable environments consisting of multiple related tasks. This view helps in understanding their failure cases and some interesting model-based results reported in the literature.
Meta Reinforcement Learning (MRL) enables an agent to learn from a limited number of past trajectories and extrapolate to a new task. In this paper, we attempt to improve the robustness of MRL. We build upon model-agnostic meta-learning (MAML) and propose a novel method to generate adversarial samples for MRL by using Generative Adversarial Network (GAN). That allows us to enhance the robustness of MRL to adversal attacks by leveraging these attacks during meta training process.
It is essential for an automated vehicle in the field to perform discretionary lane changes with appropriate roadmanship - driving safely and efficiently without annoying or endangering other road users - under a wide range of traffic cultures and driving conditions. While deep reinforcement learning methods have excelled in recent years and been applied to automated vehicle driving policy, there are concerns about their capability to quickly adapt to unseen traffic with new environment dynamics. We formulate this challenge as a multi-Markov Decision Processes (MDPs) adaptation problem and developed Meta Reinforcement Learning (MRL) driving policies to showcase their quick learning capability. Two types of distribution variation in environments were designed and simulated to validate the fast adaptation capability of resulting MRL driving policies which significantly outperform a baseline RL.
Meta-learning is a branch of machine learning which aims to quickly adapt models, such as neural networks, to perform new tasks by learning an underlying structure across related tasks. In essence, models are being trained to learn new tasks effectively rather than master a single task. Meta-learning is appealing for process control applications because the perturbations to a process required to train an AI controller can be costly and unsafe. Additionally, the dynamics and control objectives are similar across many different processes, so it is feasible to create a generalizable controller through meta-learning capable of quickly adapting to different systems. In this work, we construct a deep reinforcement learning (DRL) based controller and meta-train the controller using a latent context variable through a separate embedding neural network. We test our meta-algorithm on its ability to adapt to new process dynamics as well as different control objectives on the same process. In both cases, our meta-learning algorithm adapts very quickly to new tasks, outperforming a regular DRL controller trained from scratch. Meta-learning appears to be a promising approach for constructing more intelligent and sample-efficient controllers.
Model agnostic meta-learning (MAML) is a popular state-of-the-art meta-learning algorithm that provides good weight initialization of a model given a variety of learning tasks. The model initialized by provided weight can be fine-tuned to an unseen task despite only using a small amount of samples and within a few adaptation steps. MAML is simple and versatile but requires costly learning rate tuning and careful design of the task distribution which affects its scalability and generalization. This paper proposes a more robust MAML based on an adaptive learning scheme and a prioritization task buffer(PTB) referred to as Robust MAML (RMAML) for improving scalability of training process and alleviating the problem of distribution mismatch. RMAML uses gradient-based hyper-parameter optimization to automatically find the optimal learning rate and uses the PTB to gradually adjust train-ing task distribution toward testing task distribution over the course of training. Experimental results on meta reinforcement learning environments demonstrate a substantial performance gain as well as being less sensitive to hyper-parameter choice and robust to distribution mismatch.
Meta-reinforcement learning typically requires orders of magnitude more samples than single task reinforcement learning methods. This is because meta-training needs to deal with more diverse distributions and train extra components such as context encoders. To address this, we propose a novel self-supervised learning task, which we named Trajectory Contrastive Learning (TCL), to improve meta-training. TCL adopts contrastive learning and trains a context encoder to predict whether two transition windows are sampled from the same trajectory. TCL leverages the natural hierarchical structure of context-based meta-RL and makes minimal assumptions, allowing it to be generally applicable to context-based meta-RL algorithms. It accelerates the training of context encoders and improves meta-training overall. Experiments show that TCL performs better or comparably than a strong meta-RL baseline in most of the environments on both meta-RL MuJoCo (5 of 6) and Meta-World benchmarks (44 out of 50).
In this paper, the problem of enhancing the quality of virtual reality (VR) services is studied for an indoor terahertz (THz)/visible light communication (VLC) wireless network. In the studied model, small base stations (SBSs) transmit high-quality VR images to VR users over THz bands and light-emitting diodes (LEDs) provide accurate indoor positioning services for them using VLC. Here, VR users move in real time and their movement patterns change over time according to their applications, where both THz and VLC links can be blocked by the bodies of VR users. To control the energy consumption of the studied THz/VLC wireless VR network, VLC access points (VAPs) must be selectively turned on so as to ensure accurate and extensive positioning for VR users. Based on the user positions, each SBS must generate corresponding VR images and establish THz links without body blockage to transmit the VR content. The problem is formulated as an optimization problem whose goal is to maximize the reliability of the VR network by selecting the appropriate VAPs to be turned on and controlling the user association with SBSs. To solve this problem, a policy gradient-based reinforcement learning (RL) algorithm that adopts a meta-learning approach is proposed. The proposed meta policy gradient (MPG) algorithm enables the trained policy to quickly adapt to new user movement patterns. In order to solve the problem of maximizing the average number of successfully served users for VR scenarios with a large number of users, a dual method based MPG algorithm (D-MPG) with a low complexity is proposed. Simulation results demonstrate that, compared to the trust region policy optimization algorithm (TRPO), the proposed MPG and D-MPG algorithms yield up to 26.8% and 21.9% improvement in the reliability as well as 81.2% and 87.5% gains in the convergence speed, respectively.
Reinforcement learning is a promising paradigm for solving sequential decision-making problems, but low data efficiency and weak generalization across tasks are bottlenecks in real-world applications. Model-based meta reinforcement learning addresses these issues by learning dynamics and leveraging knowledge from prior experience. In this paper, we take a closer look at this framework, and propose a new Thompson-sampling based approach that consists of a new model to identify task dynamics together with an amortized policy optimization step. We show that our model, called a graph structured surrogate model (GSSM), outperforms state-of-the-art methods in predicting environment dynamics. Additionally, our approach is able to obtain high returns, while allowing fast execution during deployment by avoiding test time policy gradient optimization.
There has been rapidly growing interest in meta-learning as a method for increasing the flexibility and sample efficiency of reinforcement learning. One problem in this area of research, however, has been a scarcity of adequate benchmark tasks. In general, the structure underlying past benchmarks has either been too simple to be inherently interesting, or too ill-defined to support principled analysis. In the present work, we introduce a new benchmark for meta-RL research, emphasizing transparency and potential for in-depth analysis as well as structural richness. Alchemy is a 3D video game, implemented in Unity, which involves a latent causal structure that is resampled procedurally from episode to episode, affording structure learning, online inference, hypothesis testing and action sequencing based on abstract domain knowledge. We evaluate a pair of powerful RL agents on Alchemy and present an in-depth analysis of one of these agents. Results clearly indicate a frank and specific failure of meta-learning, providing validation for Alchemy as a challenging benchmark for meta-RL. Concurrent with this report, we are releasing Alchemy as public resource, together with a suite of analysis tools and sample agent trajectories.
This work developed a meta-learning approach that adapts the control policy on the fly to different changing conditions for robust locomotion. The proposed method constantly updates the interaction model, samples feasible sequences of actions of estimated the state-action trajectories, and then applies the optimal actions to maximize the reward. To achieve online model adaptation, our proposed method learns different latent vectors of each training condition, which are selected online given the newly collected data. Our work designs appropriate state space and reward functions, and optimizes feasible actions in an MPC fashion which are then sampled directly in the joint space considering constraints, hence requiring no prior design of specific walking gaits. We further demonstrate the robot's capability of detecting unexpected changes during interaction and adapting control policies quickly. The extensive validation on the SpotMicro robot in a physics simulation shows adaptive and robust locomotion skills under varying ground friction, external pushes, and different robot models including hardware faults and changes.
As power systems are undergoing a significant transformation with more uncertainties, less inertia and closer to operation limits, there is increasing risk of large outages. Thus, there is an imperative need to enhance grid emergency control to maintain system reliability and security. Towards this end, great progress has been made in developing deep reinforcement learning (DRL) based grid control solutions in recent years. However, existing DRL-based solutions have two main limitations: 1) they cannot handle well with a wide range of grid operation conditions, system parameters, and contingencies; 2) they generally lack the ability to fast adapt to new grid operation conditions, system parameters, and contingencies, limiting their applicability for real-world applications. In this paper, we mitigate these limitations by developing a novel deep meta reinforcement learning (DMRL) algorithm. The DMRL combines the meta strategy optimization together with DRL, and trains policies modulated by a latent space that can quickly adapt to new scenarios. We test the developed DMRL algorithm on the IEEE 300-bus system. We demonstrate fast adaptation of the meta-trained DRL polices with latent variables to new operating conditions and scenarios using the proposed method and achieve superior performance compared to the state-of-the-art DRL and model predictive control (MPC) methods.
This paper introduces Fast Linearized Adaptive Policy (FLAP), a new meta-reinforcement learning (meta-RL) method that is able to extrapolate well to out-of-distribution tasks without the need to reuse data from training, and adapt almost instantaneously with the need of only a few samples during testing. FLAP builds upon the idea of learning a shared linear representation of the policy so that when adapting to a new task, it suffices to predict a set of linear weights. A separate adapter network is trained simultaneously with the policy such that during adaptation, we can directly use the adapter network to predict these linear weights instead of updating a meta-policy via gradient descent, such as in prior meta-RL methods like MAML, to obtain the new policy. The application of the separate feed-forward network not only speeds up the adaptation run-time significantly, but also generalizes extremely well to very different tasks that prior Meta-RL methods fail to generalize to. Experiments on standard continuous-control meta-RL benchmarks show FLAP presenting significantly stronger performance on out-of-distribution tasks with up to double the average return and up to 8X faster adaptation run-time speeds when compared to prior methods.
Meta-reinforcement learning (RL) addresses the problem of sample inefficiency in deep RL by using experience obtained in past tasks for a new task to be solved. However, most meta-RL methods require partially or fully on-policy data, i.e., they cannot reuse the data collected by past policies, which hinders the improvement of sample efficiency. To alleviate this problem, we propose a novel off-policy meta-RL method, embedding learning and evaluation of uncertainty (ELUE). An ELUE agent is characterized by the learning of a feature embedding space shared among tasks. It learns a belief model over the embedding space and a belief-conditional policy and Q-function. Then, for a new task, it collects data by the pretrained policy, and updates its belief based on the belief model. Thanks to the belief update, the performance can be improved with a small amount of data. In addition, it updates the parameters of the neural networks to adjust the pretrained relationships when there are enough data. We demonstrate that ELUE outperforms state-of-the-art meta RL methods through experiments on meta-RL benchmarks.
Meta Reinforcement Learning (RL) methods focus on automating the design of RL algorithms that generalize to a wide range of environments. The framework introduced in (Anonymous, 2020) addresses the problem by representing different RL algorithms as Directed Acyclic Graphs (DAGs), and using an evolutionary meta learner to modify these graphs and find good agent update rules. While the search language used to generate graphs in the paper serves to represent numerous already-existing RL algorithms (e.g., DQN, DDQN), it has limitations when it comes to representing Policy Gradient algorithms. In this work we try to close this gap by extending the original search language and proposing graphs for five different Policy Gradient algorithms: VPG, PPO, DDPG, TD3, and SAC.
This paper discusses an Enhanced Model-Agnostic Meta-Learning (E-MAML) algorithm that generates fast convergence of the policy function from a small number of training examples when applied to new learning tasks. Built on top of Model-Agnostic Meta-Learning (MAML), E-MAML maintains a set of policy parameters learned in the environment for previous tasks. We apply E-MAML to developing reinforcement learning (RL)-based online fault tolerant control schemes for dynamic systems. The enhancement is applied when a new fault occurs, to re-initialize the parameters of a new RL policy that achieves faster adaption with a small number of samples of system behavior with the new fault. This replaces the random task sampling step in MAML. Instead, it exploits the extant previously generated experiences of the controller. The enhancement is sampled to maximally span the parameter space to facilitate adaption to the new fault. We demonstrate the performance of our approach combining E-MAML with proximal policy optimization (PPO) on the well-known cart pole example, and then on the fuel transfer system of an aircraft.
Reinforcement learning (RL) enjoyed significant progress over the last years. One of the most important steps forward was the wide application of neural networks. However, architectures of these neural networks are typically constructed manually. In this work, we study recently proposed neural architecture search (NAS) methods for optimizing the architecture of RL agents. We carry out experiments on the Atari benchmark and conclude that modern NAS methods find architectures of RL agents outperforming a manually selected one.
We are interested in learning models of non-stationary environments, which can be framed as a multi-task learning problem. Model-free reinforcement learning algorithms can achieve good asymptotic performance in multi-task learning at a cost of extensive sampling, due to their approach, which requires learning from scratch. While model-based approaches are among the most data efficient learning algorithms, they still struggle with complex tasks and model uncertainties. Meta-reinforcement learning addresses the efficiency and generalization challenges on multi task learning by quickly leveraging the meta-prior policy for a new task. In this paper, we propose a meta-reinforcement learning approach to learn the dynamic model of a non-stationary environment to be used for meta-policy optimization later. Due to the sample efficiency of model-based learning methods, we are able to simultaneously train both the meta-model of the non-stationary environment and the meta-policy until dynamic model convergence. Then, the meta-learned dynamic model of the environment will generate simulated data for meta-policy optimization. Our experiment demonstrates that our proposed method can meta-learn the policy in a non-stationary environment with the data efficiency of model-based learning approaches while achieving the high asymptotic performance of model-free meta-reinforcement learning.
Humans can quickly associate stimuli to solve problems in novel contexts. Our novel neural network model learns state representations of facts that can be composed to perform such associative inference. To this end, we augment the LSTM model with an associative memory, dubbed Fast Weight Memory (FWM). Through differentiable operations at every step of a given input sequence, the LSTM updates and maintains compositional associations stored in the rapidly changing FWM weights. Our model is trained end-to-end by gradient descent and yields excellent performance on compositional language reasoning problems, meta-reinforcement-learning for POMDPs, and small-scale word-level language modelling.