Traffic optimization challenges, such as load balancing, flow scheduling, and improving packet delivery time, are difficult online decision-making problems in wide area networks (WAN). Complex heuristics are needed for instance to find optimal paths that improve packet delivery time and minimize interruptions which may be caused by link failures or congestion. The recent success of reinforcement learning (RL) algorithms can provide useful solutions to build better robust systems that learn from experience in model-free settings. In this work, we consider a path optimization problem, specifically for packet routing, in large complex networks. We develop and evaluate a model-free approach, applying multi-agent meta reinforcement learning (MAMRL) that can determine the next-hop of each packet to get it delivered to its destination with minimum time overall. Specifically, we propose to leverage and compare deep policy optimization RL algorithms for enabling distributed model-free control in communication networks and present a novel meta-learning-based framework, MAMRL, for enabling quick adaptation to topology changes. To evaluate the proposed framework, we simulate with various WAN topologies. Our extensive packet-level simulation results show that compared to classical shortest path and traditional reinforcement learning approaches, MAMRL significantly reduces the average packet delivery time even when network demand increases; and compared to a non-meta deep policy optimization algorithm, our results show the reduction of packet loss in much fewer episodes when link failures occur while offering comparable average packet delivery time.
In this study, we present a meta-learning model to adapt the predictions of the network's capacity between viewers who participate in a live video streaming event. We propose the MELANIE model, where an event is formulated as a Markov Decision Process, performing meta-learning on reinforcement learning tasks. By considering a new event as a task, we design an actor-critic learning scheme to compute the optimal policy on estimating the viewers' high-bandwidth connections. To ensure fast adaptation to new connections or changes among viewers during an event, we implement a prioritized replay memory buffer based on the Kullback-Leibler divergence of the reward/throughput of the viewers' connections. Moreover, we adopt a model-agnostic meta-learning framework to generate a global model from past events. As viewers scarcely participate in several events, the challenge resides on how to account for the low structural similarity of different events. To combat this issue, we design a graph signature buffer to calculate the structural similarities of several streaming events and adjust the training of the global model accordingly. We evaluate the proposed model on the link weight prediction task on three real-world datasets of live video streaming events. Our experiments demonstrate the effectiveness of our proposed model, with an average relative gain of 25% against state-of-the-art strategies. For reproduction purposes, our evaluation datasets and implementation are publicly available at https://github.com/stefanosantaris/melanie
Traffic signal control is of critical importance for the effective use of transportation infrastructures. The rapid increase of vehicle traffic and changes in traffic patterns make traffic signal control more and more challenging. Reinforcement Learning (RL)-based algorithms have demonstrated their potential in dealing with traffic signal control. However, most existing solutions require a large amount of training data, which is unacceptable for many real-world scenarios. This paper proposes a novel model-based meta-reinforcement learning framework (ModelLight) for traffic signal control. Within ModelLight, an ensemble of models for road intersections and the optimization-based meta-learning method are used to improve the data efficiency of an RL-based traffic light control method. Experiments on real-world datasets demonstrate that ModelLight can outperform state-of-the-art traffic light control algorithms while substantially reducing the number of required interactions with the real-world environment.
Ad hoc teamwork problem describes situations where an agent has to cooperate with previously unseen agents to achieve a common goal. For an agent to be successful in these scenarios, it has to have a suitable cooperative skill. One could implement cooperative skills into an agent by using domain knowledge to design the agent's behavior. However, in complex domains, domain knowledge might not be available. Therefore, it is worthwhile to explore how to directly learn cooperative skills from data. In this work, we apply meta-reinforcement learning (meta-RL) formulation in the context of the ad hoc teamwork problem. Our empirical results show that such a method could produce robust cooperative agents in two cooperative environments with different cooperative circumstances: social compliance and language interpretation. (This is a full paper of the extended abstract version.)
We introduce AlphaD3M, an automatic machine learning (AutoML) system based on meta reinforcement learning using sequence models with self play. AlphaD3M is based on edit operations performed over machine learning pipeline primitives providing explainability. We compare AlphaD3M with state-of-the-art AutoML systems: Autosklearn, Autostacker, and TPOT, on OpenML datasets. AlphaD3M achieves competitive performance while being an order of magnitude faster, reducing computation time from hours to minutes, and is explainable by design.
Meta-reinforcement learning (meta-RL) algorithms enable agents to adapt quickly to tasks from few samples in dynamic environments. Such a feat is achieved through dynamic representations in an agent's policy network (obtained via reasoning about task context, model parameter updates, or both). However, obtaining rich dynamic representations for fast adaptation beyond simple benchmark problems is challenging due to the burden placed on the policy network to accommodate different policies. This paper addresses the challenge by introducing neuromodulation as a modular component to augment a standard policy network that regulates neuronal activities in order to produce efficient dynamic representations for task adaptation. The proposed extension to the policy network is evaluated across multiple discrete and continuous control environments of increasing complexity. To prove the generality and benefits of the extension in meta-RL, the neuromodulated network was applied to two state-of-the-art meta-RL algorithms (CAVIA and PEARL). The result demonstrates that meta-RL augmented with neuromodulation produces significantly better result and richer dynamic representations in comparison to the baselines.
Hierarchical reinforcement learning (HRL) has seen widespread interest as an approach to tractable learning of complex modular behaviors. However, existing work either assume access to expert-constructed hierarchies, or use hierarchy-learning heuristics with no provable guarantees. To address this gap, we analyze HRL in the meta-RL setting, where a learner learns latent hierarchical structure during meta-training for use in a downstream task. We consider a tabular setting where natural hierarchical structure is embedded in the transition dynamics. Analogous to supervised meta-learning theory, we provide "diversity conditions" which, together with a tractable optimism-based algorithm, guarantee sample-efficient recovery of this natural hierarchy. Furthermore, we provide regret bounds on a learner using the recovered hierarchy to solve a meta-test task. Our bounds incorporate common notions in HRL literature such as temporal and state/action abstractions, suggesting that our setting and analysis capture important features of HRL in practice.
Meta reinforcement learning (RL) attempts to discover new RL algorithms automatically from environment interaction. In so-called black-box approaches, the policy and the learning algorithm are jointly represented by a single neural network. These methods are very flexible, but they tend to underperform in terms of generalisation to new, unseen environments. In this paper, we explore the role of symmetries in meta-generalisation. We show that a recent successful meta RL approach that meta-learns an objective for backpropagation-based learning exhibits certain symmetries (specifically the reuse of the learning rule, and invariance to input and output permutations) that are not present in typical black-box meta RL systems. We hypothesise that these symmetries can play an important role in meta-generalisation. Building off recent work in black-box supervised meta learning, we develop a black-box meta RL system that exhibits these same symmetries. We show through careful experimentation that incorporating these symmetries can lead to algorithms with a greater ability to generalise to unseen action & observation spaces, tasks, and environments.
Meta-reinforcement learning (meta-RL) algorithms allow for agents to learn new behaviors from small amounts of experience, mitigating the sample inefficiency problem in RL. However, while meta-RL agents can adapt quickly to new tasks at test time after experiencing only a few trajectories, the meta-training process is still sample-inefficient. Prior works have found that in the multi-task RL setting, relabeling past transitions and thus sharing experience among tasks can improve sample efficiency and asymptotic performance. We apply this idea to the meta-RL setting and devise a new relabeling method called Hindsight Foresight Relabeling (HFR). We construct a relabeling distribution using the combination of "hindsight", which is used to relabel trajectories using reward functions from the training task distribution, and "foresight", which takes the relabeled trajectories and computes the utility of each trajectory for each task. HFR is easy to implement and readily compatible with existing meta-RL algorithms. We find that HFR improves performance when compared to other relabeling methods on a variety of meta-RL tasks.
We apply a reinforcement meta-learning framework to optimize an integrated and adaptive guidance and flight control system for an air-to-air missile. The system is implemented as a policy that maps navigation system outputs directly to commanded rates of change for the missile's control surface deflections. The system induces intercept trajectories against a maneuvering target that satisfy control constraints on fin deflection angles, and path constraints on look angle and load. We test the optimized system in a six degrees-of-freedom simulator that includes a non-linear radome model and a strapdown seeker model, and demonstrate that the system adapts to both a large flight envelope and off-nominal flight conditions including perturbation of aerodynamic coefficient parameters and center of pressure locations, and flexible body dynamics. Moreover, we find that the system is robust to the parasitic attitude loop induced by radome refraction and imperfect seeker stabilization. We compare our system's performance to a longitudinal model of proportional navigation coupled with a three loop autopilot, and find that our system outperforms this benchmark by a large margin. Additional experiments investigate the impact of removing the recurrent layer from the policy and value function networks, performance with an infrared seeker, and flexible body dynamics.
While conventional reinforcement learning focuses on designing agents that can perform one task, meta-learning aims, instead, to solve the problem of designing agents that can generalize to different tasks (e.g., environments, obstacles, and goals) that were not considered during the design or the training of these agents. In this spirit, in this paper, we consider the problem of training a provably safe Neural Network (NN) controller for uncertain nonlinear dynamical systems that can generalize to new tasks that were not present in the training data while preserving strong safety guarantees. Our approach is to learn a set of NN controllers during the training phase. When the task becomes available at runtime, our framework will carefully select a subset of these NN controllers and compose them to form the final NN controller. Critical to our approach is the ability to compute a finite-state abstraction of the nonlinear dynamical system. This abstract model captures the behavior of the closed-loop system under all possible NN weights, and is used to train the NNs and compose them when the task becomes available. We provide theoretical guarantees that govern the correctness of the resulting NN. We evaluated our approach on the problem of controlling a wheeled robot in cluttered environments that were not present in the training data.
Meta Reinforcement Learning (Meta-RL) has seen substantial advancements recently. In particular, off-policy methods were developed to improve the data efficiency of Meta-RL techniques. \textit{Probabilistic embeddings for actor-critic RL} (PEARL) is a leading approach for multi-MDP adaptation problems. A major drawback of many existing Meta-RL methods, including PEARL, is that they do not explicitly consider the safety of the prior policy when it is exposed to a new task for the first time. Safety is essential for many real-world applications, including field robots and Autonomous Vehicles (AVs). In this paper, we develop the PEARL PLUS (PEARL$^+$) algorithm, which optimizes the policy for both prior (pre-adaptation) safety and posterior (after-adaptation) performance. Building on top of PEARL, our proposed PEARL$^+$ algorithm introduces a prior regularization term in the reward function and a new Q-network for recovering the state-action value under prior context assumptions, to improve the robustness to task distribution shift and safety of the trained network exposed to a new task for the first time. The performance of PEARL$^+$ is validated by solving three safety-critical problems related to robots and AVs, including two MuJoCo benchmark problems. From the simulation experiments, we show that safety of the prior policy is significantly improved and more robust to task distribution shift compared to PEARL.
The performance of many medical image analysis tasks are strongly associated with image data quality. When developing modern deep learning algorithms, rather than relying on subjective (human-based) image quality assessment (IQA), task amenability potentially provides an objective measure of task-specific image quality. To predict task amenability, an IQA agent is trained using reinforcement learning (RL) with a simultaneously optimised task predictor, such as a classification or segmentation neural network. In this work, we develop transfer learning or adaptation strategies to increase the adaptability of both the IQA agent and the task predictor so that they are less dependent on high-quality, expert-labelled training data. The proposed transfer learning strategy re-formulates the original RL problem for task amenability in a meta-reinforcement learning (meta-RL) framework. The resulting algorithm facilitates efficient adaptation of the agent to different definitions of image quality, each with its own Markov decision process environment including different images, labels and an adaptable task predictor. Our work demonstrates that the IQA agents pre-trained on non-expert task labels can be adapted to predict task amenability as defined by expert task labels, using only a small set of expert labels. Using 6644 clinical ultrasound images from 249 prostate cancer patients, our results for image classification and segmentation tasks show that the proposed IQA method can be adapted using data with as few as respective 19.7% and 29.6% expert-reviewed consensus labels and still achieve comparable IQA and task performance, which would otherwise require a training dataset with 100% expert labels.
Recent state-of-the-art artificial agents lack the ability to adapt rapidly to new tasks, as they are trained exclusively for specific objectives and require massive amounts of interaction to learn new skills. Meta-reinforcement learning (meta-RL) addresses this challenge by leveraging knowledge learned from training tasks to perform well in previously unseen tasks. However, current meta-RL approaches limit themselves to narrow parametric task distributions, ignoring qualitative differences between tasks that occur in the real world. In this paper, we introduce TIGR, a Task-Inference-based meta-RL algorithm using Gaussian mixture models (GMM) and gated Recurrent units, designed for tasks in non-parametric environments. We employ a generative model involving a GMM to capture the multi-modality of the tasks. We decouple the policy training from the task-inference learning and efficiently train the inference mechanism on the basis of an unsupervised reconstruction objective. We provide a benchmark with qualitatively distinct tasks based on the half-cheetah environment and demonstrate the superior performance of TIGR compared to state-of-the-art meta-RL approaches in terms of sample efficiency (3-10 times faster), asymptotic performance, and applicability in non-parametric environments with zero-shot adaptation.
Animals have remarkable abilities to adapt locomotion to different terrains and tasks. However, robots trained by means of reinforcement learning are typically able to solve only a single task and a transferred policy is usually inferior to that trained from scratch. In this work, we demonstrate that meta-reinforcement learning can be used to successfully train a robot capable to solve a wide range of locomotion tasks. The performance of the meta-trained robot is similar to that of a robot that is trained on a single task.
Meta-reinforcement learning (RL) methods can meta-train policies that adapt to new tasks with orders of magnitude less data than standard RL, but meta-training itself is costly and time-consuming. If we can meta-train on offline data, then we can reuse the same static dataset, labeled once with rewards for different tasks, to meta-train policies that adapt to a variety of new tasks at meta-test time. Although this capability would make meta-RL a practical tool for real-world use, offline meta-RL presents additional challenges beyond online meta-RL or standard offline RL settings. Meta-RL learns an exploration strategy that collects data for adapting, and also meta-trains a policy that quickly adapts to data from a new task. Since this policy was meta-trained on a fixed, offline dataset, it might behave unpredictably when adapting to data collected by the learned exploration strategy, which differs systematically from the offline data and thus induces distributional shift. We propose a hybrid offline meta-RL algorithm, which uses offline data with rewards to meta-train an adaptive policy, and then collects additional unsupervised online data, without any reward labels to bridge this distribution shift. By not requiring reward labels for online collection, this data can be much cheaper to collect. We compare our method to prior work on offline meta-RL on simulated robot locomotion and manipulation tasks and find that using additional unsupervised online data collection leads to a dramatic improvement in the adaptive capabilities of the meta-trained policies, matching the performance of fully online meta-RL on a range of challenging domains that require generalization to new tasks.
In Meta-Reinforcement Learning (meta-RL) an agent is trained on a set of tasks to prepare for and learn faster in new, unseen, but related tasks. The training tasks are usually hand-crafted to be representative of the expected distribution of test tasks and hence all used in training. We show that given a set of training tasks, learning can be both faster and more effective (leading to better performance in the test tasks), if the training tasks are appropriately selected. We propose a task selection algorithm, Information-Theoretic Task Selection (ITTS), based on information theory, which optimizes the set of tasks used for training in meta-RL, irrespectively of how they are generated. The algorithm establishes which training tasks are both sufficiently relevant for the test tasks, and different enough from one another. We reproduce different meta-RL experiments from the literature and show that ITTS improves the final performance in all of them.
Model-agnostic meta-reinforcement learning requires estimating the Hessian matrix of value functions. This is challenging from an implementation perspective, as repeatedly differentiating policy gradient estimates may lead to biased Hessian estimates. In this work, we provide a unifying framework for estimating higher-order derivatives of value functions, based on off-policy evaluation. Our framework interprets a number of prior approaches as special cases and elucidates the bias and variance trade-off of Hessian estimates. This framework also opens the door to a new family of estimates, which can be easily implemented with auto-differentiation libraries, and lead to performance gains in practice.
The generalization ability of most meta-reinforcement learning (meta-RL) methods is largely limited to test tasks that are sampled from the same distribution used to sample training tasks. To overcome the limitation, we propose Latent Dynamics Mixture (LDM) that trains a reinforcement learning agent with imaginary tasks generated from mixtures of learned latent dynamics. By training a policy on mixture tasks along with original training tasks, LDM allows the agent to prepare for unseen test tasks during training and prevents the agent from overfitting the training tasks. LDM significantly outperforms standard meta-RL methods in test returns on the gridworld navigation and MuJoCo tasks where we strictly separate the training task distribution and the test task distribution.
Many real-world domains are subject to a structured non-stationarity which affects the agent's goals and the environmental dynamics. Meta-reinforcement learning (RL) has been shown successful for training agents that quickly adapt to related tasks. However, most of the existing meta-RL algorithms for non-stationary domains either make strong assumptions on the task generation process or require sampling from it at training time. In this paper, we propose a novel algorithm (TRIO) that optimizes for the future by explicitly tracking the task evolution through time. At training time, TRIO learns a variational module to quickly identify latent parameters from experience samples. This module is learned jointly with an optimal exploration policy that takes task uncertainty into account. At test time, TRIO tracks the evolution of the latent parameters online, hence reducing the uncertainty over future tasks and obtaining fast adaptation through the meta-learned policy. Unlike most existing methods, TRIO does not assume Markovian task-evolution processes, it does not require information about the non-stationarity at training time, and it captures complex changes undergoing in the environment. We evaluate our algorithm on different simulated problems and show it outperforms competitive baselines.