Offline model learning for planning is a branch of machine learning that trains agents to perform actions in an unknown environment using a fixed batch of previously collected experiences. The limited size of the data set hinders the estimate of the Value function of the relative Markov Decision Process (MDP), bounding the performance of the obtained policy in the real world. In this context, recent works showed that planning with a discount factor lower than the one used during the evaluation phase yields more performing policies. However, the optimal discount factor is finally chosen by cross-validation. Our aim is to show that looking for a sub-optimal solution of a Bayesian MDP might lead to better performances with respect to the current baselines that work in the offline setting. Hence, we propose Exploitation vs Caution (EvC), an algorithm that automatically selects the policy that solves a Risk-sensitive Bayesian MDP in a set of policies obtained by solving several MDPs characterized by different discount factors and transition dynamics. On one hand, the Bayesian formalism elegantly includes model uncertainty and on another hand the introduction of a risk-sensitive utility function guarantees robustness. We evaluated the proposed approach in different discrete simple environments offering a fair variety of MDP classes. We also compared the obtained results with state-of-the-art offline learning for planning baselines such as MOPO and MOReL. In the tested scenarios EvC is more robust than the said approaches suggesting that sub-optimally solving an Offline Risk-sensitive Bayesian MDP (ORBMDP) could define a sound framework for planning under model uncertainty.

Mobility on demand (MoD) systems show great promise in realizing flexible and efficient urban transportation. However, significant technical challenges arise from operational decision making associated with MoD vehicle dispatch and fleet rebalancing. For this reason, operators tend to employ simplified algorithms that have been demonstrated to work well in a particular setting. To help bridge the gap between novel and existing methods, we propose a modular framework for fleet rebalancing based on model-free reinforcement learning (RL) that can leverage an existing dispatch method to minimize system cost. In particular, by treating dispatch as part of the environment dynamics, a centralized agent can learn to intermittently direct the dispatcher to reposition free vehicles and mitigate against fleet imbalance. We formulate RL state and action spaces as distributions over a grid partitioning of the operating area, making the framework scalable and avoiding the complexities associated with multiagent RL. Numerical experiments, using real-world trip and network data, demonstrate that this approach has several distinct advantages over baseline methods including: improved system cost; high degree of adaptability to the selected dispatch method; and the ability to perform scale-invariant transfer learning between problem instances with similar vehicle and request distributions.

We introduce AndroidEnv, an open-source platform for Reinforcement Learning (RL) research built on top of the Android ecosystem. AndroidEnv allows RL agents to interact with a wide variety of apps and services commonly used by humans through a universal touchscreen interface. Since agents train on a realistic simulation of an Android device, they have the potential to be deployed on real devices. In this report, we give an overview of the environment, highlighting the significant features it provides for research, and we present an empirical evaluation of some popular reinforcement learning agents on a set of tasks built on this platform.

We propose a novel framework to bootstrap the reputation of on-demand service compositions. On-demand compositions are usually context-aware and have little or no direct consumer feedback. The reputation bootstrapping of single or atomic services does not consider the topology of the composition and relationships among reputation-related factors. We apply Conditional Preference Networks (CP-nets) of reputation-related factors for component services in a composition. The reputation of a composite service is bootstrapped by the composition of CP-nets. We consider the history of invocation among component services to determine reputation-interdependence in a composition. The composition rules are constructed using the composition topology and four types of reputation-influence among component services. A heuristic-based Q-learning approach is proposed to select the optimal set of reputation-related CP-nets. Experimental results prove the efficiency of the proposed approach.

I recently started an AI-focused educational newsletter, that already has over 80,000 subscribers. TheSequence is a no-BS (meaning no hype, no news etc) ML-oriented newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers and concepts. That paradigm assumes that the target knowledge is already embedded in the dataset and doesn't require any further clarifications but that rarely resembles how humans learn. When presented with a new subject, we are constantly forced to ask questions and clarifications about it.

Transfer learning can speed up training in machine learning and is regularly used in classification tasks. It reuses prior knowledge from other tasks to pre-train networks for new tasks. In reinforcement learning, learning actions for a behavior policy that can be applied to new environments is still a challenge, especially for tasks that involve much planning. Sokoban is a challenging puzzle game. It has been used widely as a benchmark in planning-based reinforcement learning. In this paper, we show how prior knowledge improves learning in Sokoban tasks. We find that reusing feature representations learned previously can accelerate learning new, more complex, instances. In effect, we show how curriculum learning, from simple to complex tasks, works in Sokoban. Furthermore, feature representations learned in simpler instances are more general, and thus lead to positive transfers towards more complex tasks, but not vice versa. We have also studied which part of the knowledge is most important for transfer to succeed, and identify which layers should be used for pre-training.

We consider the problem of modeling trajectories of drivers in a road network from the perspective of inverse reinforcement learning. As rational agents, drivers are trying to maximize some reward function unknown to an external observer as they make up their trajectories. We apply the concept of random utility from microeconomic theory to model the unknown reward function as a function of observable features plus an error term which represents features known only to the driver. We develop a parameterized generative model for the trajectories based on a random utility Markov decision process formulation of drivers decisions. We show that maximum entropy inverse reinforcement learning is a particular case of our proposed formulation when we assume a Gumbel density function for the unobserved reward error terms. We illustrate Bayesian inference on model parameters through a case study with real trajectory data from a large city obtained from sensors placed on sparsely distributed points on the street network.

Intelligent behaviour in the physical world exhibits structure at multiple spatial and temporal scales. Although movements are ultimately executed at the level of instantaneous muscle tensions or joint torques, they must be selected to serve goals defined on much longer timescales, and in terms of relations that extend far beyond the body itself, ultimately involving coordination with other agents. Recent research in artificial intelligence has shown the promise of learning-based approaches to the respective problems of complex movement, longer-term planning and multi-agent coordination. However, there is limited research aimed at their integration. We study this problem by training teams of physically simulated humanoid avatars to play football in a realistic virtual environment. We develop a method that combines imitation learning, single- and multi-agent reinforcement learning and population-based training, and makes use of transferable representations of behaviour for decision making at different levels of abstraction. In a sequence of stages, players first learn to control a fully articulated body to perform realistic, human-like movements such as running and turning; they then acquire mid-level football skills such as dribbling and shooting; finally, they develop awareness of others and play as a team, bridging the gap between low-level motor control at a timescale of milliseconds, and coordinated goal-directed behaviour as a team at the timescale of tens of seconds. We investigate the emergence of behaviours at different levels of abstraction, as well as the representations that underlie these behaviours using several analysis techniques, including statistics from real-world sports analytics. Our work constitutes a complete demonstration of integrated decision-making at multiple scales in a physically embodied multi-agent setting. See project video at https://youtu.be/KHMwq9pv7mg.

In partially observable reinforcement learning, offline training gives access to latent information which is not available during online training and/or execution, such as the system state. Asymmetric actor-critic methods exploit such information by training a history-based policy via a state-based critic. However, many asymmetric methods lack theoretical foundation, and are only evaluated on limited domains. We examine the theory of asymmetric actor-critic methods which use state-based critics, and expose fundamental issues which undermine the validity of a common variant, and its ability to address high partial observability. We propose an unbiased asymmetric actor-critic variant which is able to exploit state information while remaining theoretically sound, maintaining the validity of the policy gradient theorem, and introducing no bias and relatively low variance into the training process. An empirical evaluation performed on domains which exhibit significant partial observability confirms our analysis, and shows the unbiased asymmetric actor-critic converges to better policies and/or faster than symmetric actor-critic and standard asymmetric actor-critic baselines.

Reinforcement learning (RL) has been used successfully for solving tasks which have a well defined reward function – think AlphaZero for Go, OpenAI Five for Dota, or AlphaStar for StarCraft. However, in many practical situations you don't have a well defined reward function. Even a task as seemingly straightforward as cleaning a room has many subtle cases: should a business card with a piece of gum be thrown away as trash, or might it have sentimental value? Should the clothes on the floor be washed, or returned to the closet? Where are notebooks supposed to be stored?