Peter Vrancx

John McLeod, Hrvoje Stojic, Vincent Adam et al.

In the past decade, model-free reinforcement learning (RL) has provided solutions to challenging domains such as robotics. Model-based RL shows the prospect of being more sample-efficient than model-free methods in terms of agent-environment interactions, because the model enables to extrapolate to unseen situations. In the more recent past, model-based methods have shown superior results compared to model-free methods in some challenging domains with non-linear state transitions. At the same time, it has become apparent that RL is not market-ready yet and that many real-world applications are going to require model-based approaches, because model-free methods are too sample-inefficient and show poor performance in early stages of training. The latter is particularly important in industry, e.g. in production systems that directly impact a company's revenue. This demonstrates the necessity for a toolbox to push the boundaries for model-based RL. While there is a plethora of toolboxes for model-free RL, model-based RL has received little attention in terms of toolbox development. Bellman aims to fill this gap and introduces the first thoroughly designed and tested model-based RL toolbox using state-of-the-art software engineering practices. Our modular approach enables to combine a wide range of environment models with generic model-based agent classes that recover state-of-the-art algorithms. We also provide an experiment harness to compare both model-free and model-based agents in a systematic fashion w.r.t. user-defined evaluation metrics (e.g. cumulative reward). This paves the way for new research directions, e.g. investigating uncertainty-aware environment models that are not necessarily neural-network-based, or developing algorithms to solve industrially-motivated benchmarks that share characteristics with real-world problems.

Marcin B. Tomczak, Sergio Valcarcel Macua, Enrique Munoz de Cote et al.

Recent policy optimization approaches have achieved substantial empirical success by constructing surrogate optimization objectives. The Approximate Policy Iteration objective (Schulman et al., 2015a; Kakade and Langford, 2002) has become a standard optimization target for reinforcement learning problems. Using this objective in practice requires an estimator of the advantage function. Policy optimization methods such as those proposed in Schulman et al. (2015b) estimate the advantages using a parametric critic. In this work we establish conditions under which the parametric approximation of the critic does not introduce bias to the updates of surrogate objective. These results hold for a general class of parametric policies, including deep neural networks. We obtain a result analogous to the compatible features derived for the original Policy Gradient Theorem (Sutton et al., 1999). As a result, we also identify a previously unknown bias that current state-of-the-art policy optimization algorithms (Schulman et al., 2015a, 2017) have introduced by not employing these compatible features.

Marcin B. Tomczak, Dongho Kim, Peter Vrancx et al.

Recent policy optimization approaches (Schulman et al., 2015a; 2017) have achieved substantial empirical successes by constructing new proxy optimization objectives. These proxy objectives allow stable and low variance policy learning, but require small policy updates to ensure that the proxy objective remains an accurate approximation of the target policy value. In this paper we derive an alternative objective that obtains the value of the target policy by applying importance sampling (IS). However, the basic importance sampled objective is not suitable for policy optimization, as it incurs too high variance in policy updates. We therefore introduce an approximation that allows us to directly trade-off the bias of approximation with the variance in policy updates. We show that our approximation unifies previously developed approaches and allows us to interpolate between them. We develop a practical algorithm by optimizing the introduced objective with proximal policy optimization techniques (Schulman et al., 2017). We also provide a theoretical analysis of the introduced policy optimization objective demonstrating bias-variance trade-off. We empirically demonstrate that the resulting algorithm improves upon state of the art on-policy policy optimization on continuous control benchmarks.

Janith C. Petangoda, Sergio Pascual-Diaz, Vincent Adam et al.

We propose a novel framework for multi-task reinforcement learning (MTRL). Using a variational inference formulation, we learn policies that generalize across both changing dynamics and goals. The resulting policies are parametrized by shared parameters that allow for transfer between different dynamics and goal conditions, and by task-specific latent-space embeddings that allow for specialization to particular tasks. We show how the latent-spaces enable generalization to unseen dynamics and goals conditions. Additionally, policies equipped with such embeddings serve as a space of skills (or options) for hierarchical reinforcement learning. Since we can change task dynamics and goals independently, we name our framework Disentangled Skill Embeddings (DSE).

Felix Leibfried, Peter Vrancx

This paper proposes a new optimization objective for value-based deep reinforcement learning. We extend conventional Deep Q-Networks (DQNs) by adding a model-learning component yielding a transcoder network. The prediction errors for the model are included in the basic DQN loss as additional regularizers. This augmented objective leads to a richer training signal that provides feedback at every time step. Moreover, because learning an environment model shares a common structure with the RL problem, we hypothesize that the resulting objective improves both sample efficiency and performance. We empirically confirm our hypothesis on a range of 20 games from the Atari benchmark attaining superior results over vanilla DQN without model-based regularization.

Garrett Andersen, Peter Vrancx, Haitham Bou-Ammar

Hierarchical learning (HL) is key to solving complex sequential decision problems with long horizons and sparse rewards. It allows learning agents to break-up large problems into smaller, more manageable subtasks. A common approach to HL, is to provide the agent with a number of high-level skills that solve small parts of the overall problem. A major open question, however, is how to identify a suitable set of reusable skills. We propose a principled approach that uses human demonstrations to infer a set of subgoals based on changes in the demonstration dynamics. Using these subgoals, we decompose the learning problem into an abstract high-level representation and a set of low-level subtasks. The abstract description captures the overall problem structure, while subtasks capture desired skills. We demonstrate that we can jointly optimize over both levels of learning. We show that the resulting method significantly outperforms previous baselines on two challenging problems: the Atari 2600 game Montezuma's Revenge, and a simulated robotics problem moving the ant robot through a maze.

Anna Harutyunyan, Peter Vrancx, Pierre-Luc Bacon et al.

A temporally abstract action, or an option, is specified by a policy and a termination condition: the policy guides option behavior, and the termination condition roughly determines its length. Generally, learning with longer options (like learning with multi-step returns) is known to be more efficient. However, if the option set for the task is not ideal, and cannot express the primitive optimal policy exactly, shorter options offer more flexibility and can yield a better solution. Thus, the termination condition puts learning efficiency at odds with solution quality. We propose to resolve this dilemma by decoupling the behavior and target terminations, just like it is done with policies in off-policy learning. To this end, we give a new algorithm, Q(β), that learns the solution with respect to any termination condition, regardless of how the options actually terminate. We derive Q(β) by casting learning with options into a common framework with well-studied multi-step off-policy learning. We validate our algorithm empirically, and show that it holds up to its motivating claims.

Denis Steckelmacher, Diederik M. Roijers, Anna Harutyunyan et al.

Many real-world reinforcement learning problems have a hierarchical nature, and often exhibit some degree of partial observability. While hierarchy and partial observability are usually tackled separately (for instance by combining recurrent neural networks and options), we show that addressing both problems simultaneously is simpler and more efficient in many cases. More specifically, we make the initiation set of options conditional on the previously-executed option, and show that options with such Option-Observation Initiation Sets (OOIs) are at least as expressive as Finite State Controllers (FSCs), a state-of-the-art approach for learning in POMDPs. OOIs are easy to design based on an intuitive description of the task, lead to explainable policies and keep the top-level and option policies memoryless. Our experiments show that OOIs allow agents to learn optimal policies in challenging POMDPs, while being much more sample-efficient than a recurrent neural network over options.

Jesus Lago, Fjo De Ridder, Peter Vrancx et al.

Motivated by the increasing integration among electricity markets, in this paper we propose two different methods to incorporate market integration in electricity price forecasting and to improve the predictive performance. First, we propose a deep neural network that considers features from connected markets to improve the predictive accuracy in a local market. To measure the importance of these features, we propose a novel feature selection algorithm that, by using Bayesian optimization and functional analysis of variance, evaluates the effect of the features on the algorithm performance. In addition, using market integration, we propose a second model that, by simultaneously predicting prices from two markets, improves the forecasting accuracy even further. As a case study, we consider the electricity market in Belgium and the improvements in forecasting accuracy when using various French electricity features. We show that the two proposed models lead to improvements that are statistically significant. Particularly, due to market integration, the predictive accuracy is improved from 15.7% to 12.5% sMAPE (symmetric mean absolute percentage error). In addition, we show that the proposed feature selection algorithm is able to perform a correct assessment, i.e. to discard the irrelevant features.

Frederik Ruelens, Bert J. Claessens, Peter Vrancx et al.

This paper considers a demand response agent that must find a near-optimal sequence of decisions based on sparse observations of its environment. Extracting a relevant set of features from these observations is a challenging task and may require substantial domain knowledge. One way to tackle this problem is to store sequences of past observations and actions in the state vector, making it high dimensional, and apply techniques from deep learning. This paper investigates the capabilities of different deep learning techniques, such as convolutional neural networks and recurrent neural networks, to extract relevant features for finding near-optimal policies for a residential heating system and electric water heater that are hindered by sparse observations. Our simulation results indicate that in this specific scenario, feeding sequences of time-series to an LSTM network, which is a specific type of recurrent neural network, achieved a higher performance than stacking these time-series in the input of a convolutional neural network or deep neural network.

Roxana Rădulescu, Peter Vrancx, Ann Nowé

Congestion problems are omnipresent in today's complex networks and represent a challenge in many research domains. In the context of Multi-agent Reinforcement Learning (MARL), approaches like difference rewards and resource abstraction have shown promising results in tackling such problems. Resource abstraction was shown to be an ideal candidate for solving large-scale resource allocation problems in a fully decentralized manner. However, its performance and applicability strongly depends on some, until now, undocumented assumptions. Two of the main congestion benchmark problems considered in the literature are: the Beach Problem Domain and the Traffic Lane Domain. In both settings the highest system utility is achieved when overcrowding one resource and keeping the rest at optimum capacity. We analyse how abstract grouping can promote this behaviour and how feasible it is to apply this approach in a real-world domain (i.e., what assumptions need to be satisfied and what knowledge is necessary). We introduce a new test problem, the Road Network Domain (RND), where the resources are no longer independent, but rather part of a network (e.g., road network), thus choosing one path will also impact the load on other paths having common road segments. We demonstrate the application of state-of-the-art MARL methods for this new congestion model and analyse their performance. RND allows us to highlight an important limitation of resource abstraction and show that the difference rewards approach manages to better capture and inform the agents about the dynamics of the environment.

Bert J. Claessens, Peter Vrancx, Frederik Ruelens

Direct load control of a heterogeneous cluster of residential demand flexibility sources is a high-dimensional control problem with partial observability. This work proposes a novel approach that uses a convolutional neural network to extract hidden state-time features to mitigate the curse of partial observability. More specific, a convolutional neural network is used as a function approximator to estimate the state-action value function or Q-function in the supervised learning step of fitted Q-iteration. The approach is evaluated in a qualitative simulation, comprising a cluster of thermostatically controlled loads that only share their air temperature, whilst their envelope temperature remains hidden. The simulation results show that the presented approach is able to capture the underlying hidden features and successfully reduce the electricity cost the cluster.

Denis Steckelmacher, Peter Vrancx

This paper explores the performance of fitted neural Q iteration for reinforcement learning in several partially observable environments, using three recurrent neural network architectures: Long Short-Term Memory, Gated Recurrent Unit and MUT1, a recurrent neural architecture evolved from a pool of several thousands candidate architectures. A variant of fitted Q iteration, based on Advantage values instead of Q values, is also explored. The results show that GRU performs significantly better than LSTM and MUT1 for most of the problems considered, requiring less training episodes and less CPU time before learning a very good policy. Advantage learning also tends to produce better results.

Anna Harutyunyan, Tim Brys, Peter Vrancx et al.

Potential-based reward shaping (PBRS) is an effective and popular technique to speed up reinforcement learning by leveraging domain knowledge. While PBRS is proven to always preserve optimal policies, its effect on learning speed is determined by the quality of its potential function, which, in turn, depends on both the underlying heuristic and the scale. Knowing which heuristic will prove effective requires testing the options beforehand, and determining the appropriate scale requires tuning, both of which introduce additional sample complexity. We formulate a PBRS framework that reduces learning speed, but does not incur extra sample complexity. For this, we propose to simultaneously learn an ensemble of policies, shaped w.r.t. many heuristics and on a range of scales. The target policy is then obtained by voting. The ensemble needs to be able to efficiently and reliably learn off-policy: requirements fulfilled by the recent Horde architecture, which we take as our basis. We demonstrate empirically that (1) our ensemble policy outperforms both the base policy, and its single-heuristic components, and (2) an ensemble over a general range of scales performs at least as well as one with optimally tuned components.

Anna Harutyunyan, Tim Brys, Peter Vrancx et al.

Recent advances of gradient temporal-difference methods allow to learn off-policy multiple value functions in parallel with- out sacrificing convergence guarantees or computational efficiency. This opens up new possibilities for sound ensemble techniques in reinforcement learning. In this work we propose learning an ensemble of policies related through potential-based shaping rewards. The ensemble induces a combination policy by using a voting mechanism on its components. Learning happens in real time, and we empirically show the combination policy to outperform the individual policies of the ensemble.