Search

Deep reinforcement learning (DRL) is poised to revolutionize the field of artificial intelligence (AI) and represents a step toward building autonomous systems with a higherlevel understanding of the visual world. Currently, deep learning is enabling reinforcement learning (RL) to scale to problems that were previously intractable, such as learning to play video games directly from pixels. DRL algorithms are also applied to robotics, allowing control policies for robots to be learned directly from camera inputs in the real world. In this survey, we begin with an introduction to the general field of RL, then progress to the main streams of value-based and policy-based methods. Our survey will cover central algorithms in deep RL, including the deep Q-network (DQN), trust region policy optimization (TRPO), and asynchronous advantage actor critic. In parallel, we highlight the unique advantages of deep neural networks, focusing on visual understanding via RL. To conclude, we describe several current areas of research within the field.

Most of existing models for facial behavior analysis rely on generic classifiers, which fail to generalize well to previously unseen data. This is because of inherent differences in source (training) and target (test) data, mainly caused by variation in subjects' facial morphology, camera views, and so on. All of these account for different contexts in which target and source data are recorded, and thus, may adversely affect the performance of the models learned solely from source data. In this paper, we exploit the notion of domain adaptation and propose a data efficient approach to adapt already learned classifiers to new unseen contexts. Specifically, we build upon the probabilistic framework of Gaussian processes (GPs), and introduce domain-specific GP experts (e.g., for each subject). The model adaptation is facilitated in a probabilistic fashion, by conditioning the target expert on the predictions from multiple source experts. We further exploit the predictive variance of each expert to define an optimal weighting during inference. We evaluate the proposed model on three publicly available data sets for multi-class (MultiPIE) and multi-label (DISFA, FERA2015) facial expression analysis by performing adaptation of two contextual factors: 'where' (view) and 'who' (subject). In our experiments, the proposed approach consistently outperforms: 1) both source and target classifiers, while using a small number of target examples during the adaptation and 2) related state-of-the-art approaches for supervised domain adaptation.

Bayesian optimization (BO) is a powerful approach for seeking the global optimum of expensive black-box functions and has proven successful for fine tuning hyper-parameters of machine learning models. However, BO is practically limited to optimizing 10–20 parameters. To scale BO to high dimensions, we usually make structural assumptions on the decomposition of the objective and/or exploit the intrinsic lower dimensionality of the problem, e.g. by using linear projections. We could achieve a higher compression rate with nonlinear projections, but learning these nonlinear embeddings typically requires much data. This contradicts the BO objective of a relatively small evaluation budget. To address this challenge, we propose to learn a low-dimensional feature space jointly with (a) the response surface and (b) a reconstruction mapping. Our approach allows for optimization of BO’s acquisition function in the lower-dimensional subspace, which significantly simplifies the optimization problem. We reconstruct the original parameter space from the lower-dimensional subspace for evaluating the black-box function. For meaningful exploration, we solve a constrained optimization problem.

Model discrimination identifies a mathematical model that usefully explains and predicts a given system’s behaviour. Researchers will often have several models, i.e. hypotheses, about an underlying system mechanism, but insufficient experimental data to discriminate between the models, i.e. discard inaccurate models. Given rival mathematical models and an initial experimental data set, optimal design of experiments suggests maximally informative experimental observations that maximise a design criterion weighted by prediction uncertainty. The model uncertainty requires gradients, which may not be readily available for black-box models. This paper (i) proposes a new design criterion using the Jensen-Renyi divergence, and (ii) develops a novel method replacing black-box models with Gaussian process surrogates. Using the surrogates, we marginalise out the model parameters with approximate inference. Results show these contributions working well for both classical and new test instances. We also (iii) introduce and discuss GPdoemd, the open-source implementation of the Gaussian process surrogate method.

Designing gaits and corresponding control policies is a key challenge in robot locomotion. Even with a viable controller parametrization, finding near-optimal parameters can be daunting. Typically, this kind of parameter optimization requires specific expert knowledge and extensive robot experiments. Automatic black-box gait optimization methods greatly reduce the need for human expertise and time-consuming design processes. Many different approaches for automatic gait optimization have been suggested to date. However, no extensive comparison among them has yet been performed. In this article, we thoroughly discuss multiple automatic optimization methods in the context of gait optimization. We extensively evaluate Bayesian optimization, a model-based approach to black-box optimization under uncertainty, on both simulated problems and real robots. This evaluation demonstrates that Bayesian optimization is particularly suited for robotic applications, where it is crucial to find a good set of gait parameters in a small number of experiments.

We describe the Customer LifeTime Value (CLTV) prediction system deployed at ASOS.com, a global online fashion retailer. CLTV prediction is an important problem in e-commerce where an accurate estimate of future value allows retailers to effectively allocate marketing spend, identify and nurture high value customers and mitigate exposure to losses. The system at ASOS provides daily estimates of the future value of every customer and is one of the cornerstones of the personalised shopping experience. The state of the art in this domain uses large numbers of handcrafted features and ensemble regressors to forecast value, predict churn and evaluate customer loyalty. Recently, domains including language, vision and speech have shown dramatic advances by replacing handcrafted features with features that are learned automatically from data. We detail the system deployed at ASOS and show that learning feature representations is a promising extension to the state of the art in CLTV modelling. We propose a novel way to generate embeddings of customers, which addresses the issue of the ever changing product catalogue and obtain a significant improvement over an exhaustive set of handcrafted features., Comment: 10 pages, 11 figures

Twitter provides an open and rich source of data for studying human behaviour at scale and is widely used in social and network sciences. However, a major criticism of Twitter data is that demographic information is largely absent. Enhancing Twitter data with user ages would advance our ability to study social network structures, information flows and the spread of contagions. Approaches toward age detection of Twitter users typically focus on specific properties of tweets, e.g., linguistic features, which are language dependent. In this paper, we devise a language-independent methodology for determining the age of Twitter users from data that is native to the Twitter ecosystem. The key idea is to use a Bayesian framework to generalise ground-truth age information from a few Twitter users to the entire network based on what/whom they follow. Our approach scales to inferring the age of 700 million Twitter accounts with high accuracy.

Intention inference can be an essential step toward efficient human–robot interaction. For this purpose, we propose the Intention-Driven Dynamics Model (IDDM) to probabilistically model the generative process of movements that are directed by the intention. The IDDM allows the intention to be inferred from observed movements using Bayes’ theorem. The IDDM simultaneously finds a latent state representation of noisy and high-dimensional observations, and models the intention-driven dynamics in the latent states. As most robotics applications are subject to real-time constraints, we develop an efficient online algorithm that allows for real-time intention inference. Two human–robot interaction scenarios, i.e. target prediction for robot table tennis and action recognition for interactive humanoid robots, are used to evaluate the performance of our inference algorithm. In both intention inference tasks, the proposed algorithm achieves substantial improvements over support vector machines and Gaussian processes.

We consider the problem of area coverage for robot teams operating under resource constraints, while modeling spatio-temporal environmental phenomena. The aim of the mobile robot team is to avoid exhaustive search and only visit the most important locations that can improve the prediction accuracy of a spatio-temporal model. We use a Gaussian Process (GP) to model spatially varying and temporally evolving dynamics of the target phenomenon. Each robot of the team is allocated a dedicated search area wherein the robot autonomously optimizes its prediction accuracy. We present this as a Decentralized Computation and Centralized Data Fusion approach wherein the trajectory sampled by the robot is generated using our proposed Resource-Constrained Decentralized Active Sensing (RC-DAS). Since each robot possesses its own independent prediction model, at the end of robot's mission time, we fuse all the prediction models from all robots to have a global model of the spatio-temporal phenomenon. Previously, all robots and GPs needed to be synchronized, such that the GPs can be jointly trained. However, doing so defeats the purpose of a fully decentralized mobile robot team. Thus, we allow the robots to independently gather new measurements and update their model parameters irrespective of other members of the team. To evaluate the performance of our model, we compare the trajectory traced by the robot using active and passive (e.g., nearest neighbor selection) sensing. We compare the performance and cost incurred by a resource constrained optimization with the unconstrained entropy maximization version.

Social and Affective Robotics is a growing multidisciplinary field encompassing computer science, engineering, psychology, education, and many other disciplines. It explores how social and affective factors influence interactions between humans and robots, and how affect and social signals can be sensed and integrated into the design, implementation, and evaluation of robots. With talks by renowned researchers in this area, Social and Affective Robotics Tutorial will help both new and experienced researchers to identify trends, concepts, methodologies and applications in this field, identified as a technological megatrend driving the fourth industrial revolution.

For a broad range of research and practical applications it is important to understand the allegiances, communities and structure of key players in society. One promising direction towards extracting this information is to exploit the rich relational data in digital social networks (the social graph). As global social networks (e.g., Facebook and Twitter) are very large, most approaches make use of distributed computing systems for this purpose. Distributing graph processing requires solving many difficult engineering problems, which has lead some researchers to look at single-machine solutions that are faster and easier to maintain. In this article, we present an approach for analyzing full social networks on a standard laptop, allowing for interactive exploration of the communities in the locality of a set of user specified query vertices. The key idea is that the aggregate actions of large numbers of users can be compressed into a data structure that encapsulates the edge weights between vertices in a derived graph. Local communities can be constructed by selecting vertices that are connected to the query vertices with high edge weights in the derived graph. This compression is robust to noise and allows for interactive queries of local communities in real-time, which we define to be less than the average human reaction time of 0.25s. We achieve single-machine real-time performance by compressing the neighborhood of each vertex using minhash signatures and facilitate rapid queries through Locality Sensitive Hashing. These techniques reduce query times from hours using industrial desktop machines operating on the full graph to milliseconds on standard laptops. Our method allows exploration of strongly associated regions (i.e., communities) of large graphs in real-time on a laptop. It has been deployed in software that is actively used by social network analysts and offers another channel for media owners to monetize their data, helping them to continue to provide free services that are valued by billions of people globally.

Autonomous learning has been a promising direction in control and robotics for more than a decade since data-driven learning allows to reduce the amount of engineering knowledge, which is otherwise required. However, autonomous reinforcement learning (RL) approaches typically require many interactions with the system to learn controllers, which is a practical limitation in real systems, such as robots, where many interactions can be impractical and time consuming. To address this problem, current learning approaches typically require task-specific knowledge in form of expert demonstrations, realistic simulators, pre-shaped policies, or specific knowledge about the underlying dynamics. In this article, we follow a different approach and speed up learning by extracting more information from data. In particular, we learn a probabilistic, non-parametric Gaussian process transition model of the system. By explicitly incorporating model uncertainty into long-term planning and controller learning our approach reduces the effects of model errors, a key problem in model-based learning. Compared to state-of-the art RL our model-based policy search method achieves an unprecedented speed of learning. We demonstrate its applicability to autonomous learning in real robot and control tasks., 20 pages, 29 figures; fixed a typo in equation on page 8

International audience; — In whole-body control, joint torques and external forces need to be estimated accurately. In principle, this can be done through pervasive joint-torque sensing and accurate system identification. However, these sensors are expensive and may not be integrated in all links. Moreover, the exact position of the contact must be known for a precise estimation. If contacts occur on the whole body, tactile sensors can estimate the contact location, but this requires a kinematic spatial calibration, which is prone to errors. Accumulating errors may have dramatic effects on the system identification. As an alternative to classical model-based approaches we propose a data-driven mixture-of-experts learning approach using Gaus-sian processes. This model predicts joint torques directly from raw data of tactile and force/torque sensors. We compare our approach to an analytic model-based approach on real world data recorded from the humanoid iCub. We show that the learned model accurately predicts the joint torques resulting from contact forces, is robust to changes in the environment and outperforms existing dynamic models that use of force/ torque sensor data.

© 2014 Elsevier B.V.In robotics, lower-level controllers are typically used to make the robot solve a specific task in a fixed context. For example, the lower-level controller can encode a hitting movement while the context defines the target coordinates to hit. However, in many learning problems the context may change between task executions. To adapt the policy to a new context, we utilize a hierarchical approach by learning an upper-level policy that generalizes the lower-level controllers to new contexts. A common approach to learn such upper-level policies is to use policy search. However, the majority of current contextual policy search approaches are model-free and require a high number of interactions with the robot and its environment. Model-based approaches are known to significantly reduce the amount of robot experiments, however, current model-based techniques cannot be applied straightforwardly to the problem of learning contextual upper-level policies. They rely on specific parametrizations of the policy and the reward function, which are often unrealistic in the contextual policy search formulation. In this paper, we propose a novel model-based contextual policy search algorithm that is able to generalize lower-level controllers, and is data-efficient. Our approach is based on learned probabilistic forward models and information theoretic policy search. Unlike current algorithms, our method does not require any assumption on the parametrization of the policy or the reward function. We show on complex simulated robotic tasks and in a real robot experiment that the proposed learning framework speeds up the learning process by up to two orders of magnitude in comparison to existing methods, while learning high quality policies.

26.03.14 KB. Ok toadd working paper to spiral, Off-the-shelf Gaussian Process (GP) covariance functions encode smoothness assumptions on the structure of the function to be modeled. To model complex and non-differentiable functions, these smoothness assumptions are often too restrictive. One way to alleviate this limitation is to find a different representation of the data by introducing a feature space. This feature space is often learned in an unsupervised way, which might lead to data representations that are not useful for the overall regression task. In this paper, we propose Manifold Gaussian Processes, a novel supervised method that jointly learns a transformation of the data into a feature space and a GP regression from the feature space to observed space. The Manifold GP is a full GP and allows to learn data representations, which are useful for the overall regression task. As a proof-of-concept, we evaluate our approach on complex non-smooth functions where standard GPs perform poorly, such as step functions and robotics tasks with contacts.

One of the most elegant ways of teaching new skills to robots is to provide demonstrations of a task and let the robot imitate this behavior. Such imitation learning is a non-trivial task: Different anatomies of robot and teacher, and reduced robustness towards changes in the control task are two major difficulties in imitation learning. We present an imitation-learning approach to efficiently learn a task from expert demonstrations. Instead of finding policies indirectly, either via state-action mappings (behavioral cloning), or cost function learning (inverse reinforcement learning), our goal is to find policies directly such that predicted trajectories match observed ones. To achieve this aim, we model the trajectory of the teacher and the predicted robot trajectory by means of probability distributions. We match these distributions by minimizing their Kullback-Leibler divergence. In this paper, we propose to learn probabilistic forward models to compute a probability distribution over trajectories. We compare our approach to model-based reinforcement learning methods with hand-crafted cost functions. Finally, we evaluate our method with experiments on a real compliant robot.

Deep learning has proven to be beneficial for complex tasks such as classifying images. However, this approach has been mostly applied to static datasets. The analysis of non-stationary (e.g., concept drift) streams of data involves specific issues connected with the temporal and changing nature of the data. In this paper, we propose a proof-of-concept method, called Adaptive Deep Belief Networks, of how deep learning can be generalized to learn online from changing streams of data. We do so by exploiting the generative properties of the model to incrementally re-train the Deep Belief Network whenever new data are collected. This approach eliminates the need to store past observations and, therefore, requires only constant memory consumption. Hence, our approach can be valuable for life-long learning from non-stationary data streams. © 2012 Springer-Verlag.

This paper presents a novel approach for automatic optimisation of reconfigurable design parameters based on knowledge transfer. The key idea is to make use of insights derived from optimising related designs to benefit future optimisations. We show how to use designs targeting one device to speed up optimisation of another device. The proposed approach is evaluated based on various applications including computational finance and seismic imaging. It is capable of achieving up to 35% reduction in optimisation time in producing designs with similar performance, compared to alternative optimisation methods.

We present a novel approach for supervised domain adaptation that is based upon the probabilistic framework of Gaussian processes (GPs). Specifically, we introduce domain-specific GPs as local experts for facial expression classification from face images. The adaptation of the classifier is facilitated in probabilistic fashion by conditioning the target expert on multiple source experts. Furthermore, in contrast to existing adaptation approaches, we also learn a target expert from available target data solely. Then, a single and confident classifier is obtained by combining the predictions from multiple experts based on their confidence. Learning of the model is efficient and requires no retraining/ reweighting of the source classifiers. We evaluate the proposed approach on two publicly available datasets for multi-class (MultiPIE) and multi-label (DISFA) facial expression classification. To this end, we perform adaptation of two contextual factors: 'where' (view) and 'who' (subject). We show in our experiments that the proposed approach consistently outperforms both source and target classifiers, while using as few as 30 target examples. It also outperforms the state-of-the-art approaches for supervised domain adaptation.