var bibbase_data = {"data":"\"Loading..\"\n\n
\n\n \n\n \n\n \n \n\n \n\n \n \n\n \n\n \n
\n generated by\n \n \"bibbase.org\"\n\n \n
\n \n\n
\n\n \n\n\n
\n\n Excellent! Next you can\n create a new website with this list, or\n embed it in an existing web page by copying & pasting\n any of the following snippets.\n\n
\n JavaScript\n (easiest)\n
\n \n <script src=\"https://bibbase.org/service/mendeley/20eea928-2566-3876-901d-9a50fe4a71d0?jsonp=1&theme=dividers&jsonp=1\"></script>\n \n
\n\n PHP\n
\n \n <?php\n $contents = file_get_contents(\"https://bibbase.org/service/mendeley/20eea928-2566-3876-901d-9a50fe4a71d0?jsonp=1&theme=dividers\");\n print_r($contents);\n ?>\n \n
\n\n iFrame\n (not recommended)\n
\n \n <iframe src=\"https://bibbase.org/service/mendeley/20eea928-2566-3876-901d-9a50fe4a71d0?jsonp=1&theme=dividers\"></iframe>\n \n
\n\n

\n For more details see the documention.\n

\n
\n
\n\n
\n\n This is a preview! To use this list on your own web site\n or create a new web site from it,\n create a free account. The file will be added\n and you will be able to edit it in the File Manager.\n We will show you instructions once you've created your account.\n
\n\n
\n\n

To the site owner:

\n\n

Action required! Mendeley is changing its\n API. In order to keep using Mendeley with BibBase past April\n 14th, you need to:\n

    \n
  1. renew the authorization for BibBase on Mendeley, and
    2. \n
  2. update the BibBase URL\n in your page the same way you did when you initially set up\n this page.\n
    3. \n
\n

\n\n

\n \n \n Fix it now\n

\n
\n\n
\n\n\n
\n \n \n
\n
\n  \n 2024\n \n \n (4)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Path signatures for diversity in probabilistic trajectory optimisation.\n \n \n \n \n\n\n \n Barcelos, L.; Lai, T.; Oliveira, R.; Borges, P.; and Ramos, F.\n\n\n \n\n\n\n The International Journal of Robotics Research. 2024.\n \n\n\n\n
\n\n\n\n \n \n \"PathWebsite\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{\n title = {Path signatures for diversity in probabilistic trajectory optimisation},\n type = {article},\n year = {2024},\n websites = {https://doi.org/10.1177/02783649241233300},\n id = {096a5896-bf74-3e2d-a7d0-bf6d5fbdbc0f},\n created = {2024-05-24T00:06:53.954Z},\n file_attached = {false},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2024-09-23T23:55:15.363Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {doi:10.1177/02783649241233300},\n source_type = {article},\n private_publication = {false},\n abstract = {Motion planning can be cast as a trajectory optimisation problem where a cost is minimised as a function of the trajectory being generated. In complex environments with several obstacles and complicated geometry, this optimisation problem is usually difficult to solve and prone to local minima. However, recent advancements in computing hardware allow for parallel trajectory optimisation where multiple solutions are obtained simultaneously, each initialised from a different starting point. Unfortunately, without a strategy preventing two solutions to collapse on each other, naive parallel optimisation can suffer from mode collapse diminishing the efficiency of the approach and the likelihood of finding a global solution. In this paper, we leverage on recent advances in the theory of rough paths to devise an algorithm for parallel trajectory optimisation that promotes diversity over the range of solutions, therefore avoiding mode collapses and achieving better global properties. Our approach builds on path signatures and Hilbert space representations of trajectories and connects parallel variational inference for trajectory estimation with diversity-promoting kernels. We empirically demonstrate that this strategy achieves lower average costs than competing alternatives on a range of problems, from 2D navigation to robotic manipulators operating in cluttered environments.},\n bibtype = {article},\n author = {Barcelos, Lucas and Lai, Tin and Oliveira, Rafael and Borges, Paulo and Ramos, Fabio},\n doi = {10.1177/02783649241233300},\n journal = {The International Journal of Robotics Research},\n keywords = {Stein's method,motion planning,path signatures,path-planning,variational inference}\n}
\n
\n\n\n
\n Motion planning can be cast as a trajectory optimisation problem where a cost is minimised as a function of the trajectory being generated. In complex environments with several obstacles and complicated geometry, this optimisation problem is usually difficult to solve and prone to local minima. However, recent advancements in computing hardware allow for parallel trajectory optimisation where multiple solutions are obtained simultaneously, each initialised from a different starting point. Unfortunately, without a strategy preventing two solutions to collapse on each other, naive parallel optimisation can suffer from mode collapse diminishing the efficiency of the approach and the likelihood of finding a global solution. In this paper, we leverage on recent advances in the theory of rough paths to devise an algorithm for parallel trajectory optimisation that promotes diversity over the range of solutions, therefore avoiding mode collapses and achieving better global properties. Our approach builds on path signatures and Hilbert space representations of trajectories and connects parallel variational inference for trajectory estimation with diversity-promoting kernels. We empirically demonstrate that this strategy achieves lower average costs than competing alternatives on a range of problems, from 2D navigation to robotic manipulators operating in cluttered environments.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Stein Random Feature Regression.\n \n \n \n\n\n \n Warren, H.; Oliveira, R.; and Ramos, F.\n\n\n \n\n\n\n In Conference on Uncertainty in Artificial Intelligence (UAI), 2024. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Stein Random Feature Regression},\n type = {inproceedings},\n year = {2024},\n city = {Barcelona, Spain},\n id = {71ab50bf-324e-3609-9ac6-779014bde674},\n created = {2024-09-23T23:55:14.928Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2024-09-23T23:55:21.568Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Warren2024},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Warren, Houston and Oliveira, Rafael and Ramos, Fabio},\n booktitle = {Conference on Uncertainty in Artificial Intelligence (UAI)},\n keywords = {Gaussian processes,Stein's method,variational inference}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Variational search distributions.\n \n \n \n \n\n\n \n Steinberg, D., M.; Oliveira, R.; Ong, C., S.; and Bonilla, E., V.\n\n\n \n\n\n\n arXiv e-prints. 2024.\n \n\n\n\n
\n\n\n\n \n \n \"VariationalWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{\n title = {Variational search distributions},\n type = {article},\n year = {2024},\n websites = {https://arxiv.org/abs/2409.06142},\n id = {2724850e-3962-33b7-ba62-e3441ac15821},\n created = {2024-10-03T01:55:31.685Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2025-02-17T02:54:02.921Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Steinberg2024},\n private_publication = {false},\n bibtype = {article},\n author = {Steinberg, Daniel M. and Oliveira, Rafael and Ong, Cheng Soon and Bonilla, Edwin V.},\n journal = {arXiv e-prints},\n keywords = {Bayesian optimisation,Gaussian processes,NTK,RKHS,approximate inference,approximation bounds,combinatorial optimisation,convergence analysis,neural networks,variational inference}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Bayesian Adaptive Calibration and Optimal Design.\n \n \n \n \n\n\n \n Oliveira, R.; Sejdinovic, D.; Howard, D.; and Bonilla, E., V.\n\n\n \n\n\n\n In 38th Conference on Neural Information Processing Systems (NeurIPS 2024), volume 37, of Advances in Neural Information Processing Systems, 2024. Courant Associates, Inc.\n \n\n\n\n
\n\n\n\n \n \n \"BayesianWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Bayesian Adaptive Calibration and Optimal Design},\n type = {inproceedings},\n year = {2024},\n volume = {37},\n websites = {https://arxiv.org/abs/2405.14440},\n publisher = {Courant Associates, Inc.},\n city = {Vancouver, Canada},\n series = {Advances in Neural Information Processing Systems},\n id = {8f6332d8-97a2-3794-b8dd-4437286d9b07},\n created = {2024-10-03T02:04:30.994Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2025-03-13T00:51:14.927Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2024},\n notes = {To appear},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Sejdinovic, Dino and Howard, David and Bonilla, Edwin V.},\n booktitle = {38th Conference on Neural Information Processing Systems (NeurIPS 2024)},\n keywords = {Bayesian inference,active learning,calibration,experimental design}\n}
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2022\n \n \n (5)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Bayesian optimization with informative parametric models via sequential Monte Carlo.\n \n \n \n \n\n\n \n Oliveira, R.; Scalzo, R.; Kohn, R.; Cripps, S.; Hardman, K.; Close, J.; Taghavi, N.; and Lemckert, C.\n\n\n \n\n\n\n Data-Centric Engineering, 3: e5. 3 2022.\n \n\n\n\n
\n\n\n\n \n \n \"BayesianWebsite\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{\n title = {Bayesian optimization with informative parametric models via sequential Monte Carlo},\n type = {article},\n year = {2022},\n keywords = {Bayesian optimisation,geoscience,sequential Monte Carlo},\n pages = {e5},\n volume = {3},\n websites = {https://www.cambridge.org/core/product/identifier/S2632673622000053/type/journal_article},\n month = {3},\n day = {8},\n id = {fc132933-021b-3eeb-84ee-959ecae6fdf8},\n created = {2022-03-13T03:39:20.297Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2022-05-12T08:01:33.395Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2022},\n private_publication = {false},\n abstract = {Bayesian optimization (BO) has been a successful approach to optimize expensive functions whose prior knowledge can be specified by means of a probabilistic model. Due to their expressiveness and tractable closed-form predictive distributions, Gaussian process (GP) surrogate models have been the default go-to choice when deriving BO frameworks. However, as nonparametric models, GPs offer very little in terms of interpretability and informative power when applied to model complex physical phenomena in scientific applications. In addition, the Gaussian assumption also limits the applicability of GPs to problems where the variables of interest may highly deviate from Gaussianity. In this article, we investigate an alternative modeling framework for BO which makes use of sequential Monte Carlo (SMC) to perform Bayesian inference with parametric models. We propose a BO algorithm to take advantage of SMC’s flexible posterior representations and provide methods to compensate for bias in the approximations and reduce particle degeneracy. Experimental results on simulated engineering applications in detecting water leaks and contaminant source localization are presented showing performance improvements over GP-based BO approaches.},\n bibtype = {article},\n author = {Oliveira, Rafael and Scalzo, Richard and Kohn, Robert and Cripps, Sally and Hardman, Kyle and Close, John and Taghavi, Nasrin and Lemckert, Charles},\n doi = {10.1017/dce.2022.5},\n journal = {Data-Centric Engineering}\n}
\n
\n\n\n
\n Bayesian optimization (BO) has been a successful approach to optimize expensive functions whose prior knowledge can be specified by means of a probabilistic model. Due to their expressiveness and tractable closed-form predictive distributions, Gaussian process (GP) surrogate models have been the default go-to choice when deriving BO frameworks. However, as nonparametric models, GPs offer very little in terms of interpretability and informative power when applied to model complex physical phenomena in scientific applications. In addition, the Gaussian assumption also limits the applicability of GPs to problems where the variables of interest may highly deviate from Gaussianity. In this article, we investigate an alternative modeling framework for BO which makes use of sequential Monte Carlo (SMC) to perform Bayesian inference with parametric models. We propose a BO algorithm to take advantage of SMC’s flexible posterior representations and provide methods to compensate for bias in the approximations and reduce particle degeneracy. Experimental results on simulated engineering applications in detecting water leaks and contaminant source localization are presented showing performance improvements over GP-based BO approaches.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Bayesian Optimisation for Robust Model Predictive Control under Model Parameter Uncertainty.\n \n \n \n \n\n\n \n Guzman, R.; Oliveira, R.; and Ramos, F.\n\n\n \n\n\n\n In 2022 International Conference on Robotics and Automation (ICRA), pages 5539-5545, 5 2022. IEEE\n \n\n\n\n
\n\n\n\n \n \n \"BayesianWebsite\n  \n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Bayesian Optimisation for Robust Model Predictive Control under Model Parameter Uncertainty},\n type = {inproceedings},\n year = {2022},\n pages = {5539-5545},\n websites = {https://ieeexplore.ieee.org/document/9812406/},\n month = {5},\n publisher = {IEEE},\n day = {23},\n id = {42942803-3066-35a3-8465-1aab700b403c},\n created = {2022-08-14T03:47:27.141Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2023-09-26T09:51:13.032Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Guzman2022},\n private_publication = {false},\n abstract = {We propose an adaptive optimisation approach for tuning stochastic model predictive control (MPC) hyper-parameters while jointly estimating probability distributions of the transition model parameters based on performance rewards. In particular, we develop a Bayesian optimisation (BO) algorithm with a heteroscedastic noise model to deal with varying noise across the MPC hyper-parameter and dynamics model parameter spaces. Typical homoscedastic noise models are unrealistic for tuning MPC since stochastic controllers are inherently noisy, and the level of noise is affected by their hyper-parameter settings. We evaluate the proposed optimisation algorithm in simulated control and robotics tasks where we jointly infer control and dynamics parameters. Experimental results demonstrate that our approach leads to higher cumulative rewards and more stable controllers.},\n bibtype = {inproceedings},\n author = {Guzman, Rel and Oliveira, Rafael and Ramos, Fabio},\n doi = {10.1109/ICRA46639.2022.9812406},\n booktitle = {2022 International Conference on Robotics and Automation (ICRA)},\n keywords = {Bayesian optimisation,control systems,robotics}\n}
\n
\n\n\n
\n We propose an adaptive optimisation approach for tuning stochastic model predictive control (MPC) hyper-parameters while jointly estimating probability distributions of the transition model parameters based on performance rewards. In particular, we develop a Bayesian optimisation (BO) algorithm with a heteroscedastic noise model to deal with varying noise across the MPC hyper-parameter and dynamics model parameter spaces. Typical homoscedastic noise models are unrealistic for tuning MPC since stochastic controllers are inherently noisy, and the level of noise is affected by their hyper-parameter settings. We evaluate the proposed optimisation algorithm in simulated control and robotics tasks where we jointly infer control and dynamics parameters. Experimental results demonstrate that our approach leads to higher cumulative rewards and more stable controllers.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Generalized Bayesian Quadrature with Spectral Kernels.\n \n \n \n\n\n \n Warren, H.; Oliveira, R.; and Ramos, F.\n\n\n \n\n\n\n In the 38th Conference on Uncertainty in Artificial Intelligence (UAI 2022), 2022. PMLR\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Generalized Bayesian Quadrature with Spectral Kernels},\n type = {inproceedings},\n year = {2022},\n publisher = {PMLR},\n city = {Eindhoven, The Netherlands},\n id = {985126c1-4383-3e4e-b9c1-c39a79f3a368},\n created = {2022-08-14T03:47:27.201Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2023-09-26T09:51:19.747Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Warren2022},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Warren, Houston and Oliveira, Rafael and Ramos, Fabio},\n booktitle = {the 38th Conference on Uncertainty in Artificial Intelligence (UAI 2022)},\n keywords = {Bayesian inference,Fourier features,Gaussian processes,approximation bounds,quadrature}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Batch Bayesian optimisation via density-ratio estimation with guarantees.\n \n \n \n \n\n\n \n Oliveira, R.; Tiao, L., C.; and Ramos, F.\n\n\n \n\n\n\n In 36th Conference on Neural Information Processing Systems (NeurIPS 2022), 2022. \n \n\n\n\n
\n\n\n\n \n \n \"BatchWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Batch Bayesian optimisation via density-ratio estimation with guarantees},\n type = {inproceedings},\n year = {2022},\n websites = {http://arxiv.org/abs/2209.10715},\n id = {5431d1d1-6c69-39e5-a90e-f6aafd3da0ce},\n created = {2022-10-12T17:26:09.800Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2024-12-06T01:57:14.396Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2022},\n private_publication = {false},\n abstract = {Bayesian optimisation (BO) algorithms have shown remarkable success in applications involving expensive black-box functions. Traditionally BO has been set as a sequential decision-making process which estimates the utility of query points via an acquisition function and a prior over functions, such as a Gaussian process. Recently, however, a reformulation of BO via density-ratio estimation (BORE) allowed reinterpreting the acquisition function as a probabilistic binary classifier, removing the need for an explicit prior over functions and increasing scalability. In this paper, we present a theoretical analysis of BORE's regret and an extension of the algorithm with improved uncertainty estimates. We also show that BORE can be naturally extended to a batch optimisation setting by recasting the problem as approximate Bayesian inference. The resulting algorithm comes equipped with theoretical performance guarantees and is assessed against other batch BO baselines in a series of experiments.},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Tiao, Louis C. and Ramos, Fabio},\n booktitle = {36th Conference on Neural Information Processing Systems (NeurIPS 2022)},\n keywords = {Bayesian optimisation,convergence analysis,density-ratio estimation,regret bounds}\n}
\n
\n\n\n
\n Bayesian optimisation (BO) algorithms have shown remarkable success in applications involving expensive black-box functions. Traditionally BO has been set as a sequential decision-making process which estimates the utility of query points via an acquisition function and a prior over functions, such as a Gaussian process. Recently, however, a reformulation of BO via density-ratio estimation (BORE) allowed reinterpreting the acquisition function as a probabilistic binary classifier, removing the need for an explicit prior over functions and increasing scalability. In this paper, we present a theoretical analysis of BORE's regret and an extension of the algorithm with improved uncertainty estimates. We also show that BORE can be naturally extended to a batch optimisation setting by recasting the problem as approximate Bayesian inference. The resulting algorithm comes equipped with theoretical performance guarantees and is assessed against other batch BO baselines in a series of experiments.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Value Function Approximations via Kernel Embeddings for No-Regret Reinforcement Learning.\n \n \n \n\n\n \n Chowdhury, S., R.; and Oliveira, R.\n\n\n \n\n\n\n In Asian Conference on Machine Learning, volume 189, pages 249-264, 2022. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Value Function Approximations via Kernel Embeddings for No-Regret Reinforcement Learning},\n type = {inproceedings},\n year = {2022},\n keywords = {RKHS,kernel embeddings,reinforcement learning},\n pages = {249-264},\n volume = {189},\n id = {406747aa-55c7-3c8a-9c40-a085c70b02b8},\n created = {2024-07-17T00:22:16.070Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2024-07-17T00:22:17.654Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Chowdhury2022},\n private_publication = {false},\n abstract = {We consider the regret minimization problem in reinforcement learning (RL) in the episodic setting. In many real-world RL environments, the state and action spaces are continuous or very large. Existing approaches establish regret guarantees by either a low-dimensional representation of the stochastic transition model or an approximation of the Q-functions. However, the understanding of function approximation schemes for state-value functions largely remains missing. In this paper, we propose an online model-based RL algorithm, namely the CME-RL, that learns embeddings of the state-transition distribution in a reproducing kernel Hilbert space while carefully balancing the exploitation-exploration tradeoff. We demonstrate the efficiency of_our algorithm by proving a frequentist (worst-case) regret bound that is of order Õ(HγN√N)1, where H is the episode length, N is the total number of time steps and γN is an information theoretic quantity relating the effective dimension of the state-action feature space. Our method bypasses the need for estimating transition probabilities and applies to any domain on which kernels can be defined. It also brings new insights into the general theory of kernel methods for approximate inference and RL regret minimization.},\n bibtype = {inproceedings},\n author = {Chowdhury, Sayak Ray and Oliveira, Rafael},\n booktitle = {Asian Conference on Machine Learning}\n}
\n
\n\n\n
\n We consider the regret minimization problem in reinforcement learning (RL) in the episodic setting. In many real-world RL environments, the state and action spaces are continuous or very large. Existing approaches establish regret guarantees by either a low-dimensional representation of the stochastic transition model or an approximation of the Q-functions. However, the understanding of function approximation schemes for state-value functions largely remains missing. In this paper, we propose an online model-based RL algorithm, namely the CME-RL, that learns embeddings of the state-transition distribution in a reproducing kernel Hilbert space while carefully balancing the exploitation-exploration tradeoff. We demonstrate the efficiency of_our algorithm by proving a frequentist (worst-case) regret bound that is of order Õ(HγN√N)1, where H is the episode length, N is the total number of time steps and γN is an information theoretic quantity relating the effective dimension of the state-action feature space. Our method bypasses the need for estimating transition probabilities and applies to any domain on which kernels can be defined. It also brings new insights into the general theory of kernel methods for approximate inference and RL regret minimization.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2021\n \n \n (4)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n Heteroscedastic Bayesian Optimisation for Stochastic Model Predictive Control.\n \n \n \n\n\n \n Guzman, R.; Oliveira, R.; and Ramos, F.\n\n\n \n\n\n\n IEEE Robotics and Automation Letters, 6(1): 56-63. 1 2021.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{\n title = {Heteroscedastic Bayesian Optimisation for Stochastic Model Predictive Control},\n type = {article},\n year = {2021},\n keywords = {Bayesian optimisation,control systems,model predictive control,motion planning,robotics},\n pages = {56-63},\n volume = {6},\n month = {1},\n publisher = {IEEE},\n id = {e25c9ba4-227e-341e-bd26-2a86fdef3e23},\n created = {2020-10-21T02:51:58.558Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2023-09-26T09:51:26.156Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Guzman2021},\n source_type = {article},\n private_publication = {false},\n abstract = {Model predictive control (MPC) has been successful in applications involving the control of complex physical systems. This class of controllers leverages the information provided by an approximate model of the system's dynamics to simulate the effect of control actions. MPC methods also present a few hyper-parameters which may require a relatively expensive tuning process by demanding interactions with the physical system. Therefore, we investigate fine-tuning MPC methods in the context of stochastic MPC, which presents extra challenges due to the randomness of the controller's actions. In these scenarios, performance outcomes present noise, which is not homogeneous across the domain of possible hyper-parameter settings, but which varies in an input-dependent way. To address these issues, we propose a Bayesian optimisation framework that accounts for heteroscedastic noise to tune hyper-parameters in control problems. Empirical results on benchmark continuous control tasks and a physical robot support the proposed framework's suitability relative to baselines, which do not take heteroscedasticity into account.},\n bibtype = {article},\n author = {Guzman, R and Oliveira, R and Ramos, F},\n doi = {10.1109/LRA.2020.3028830},\n journal = {IEEE Robotics and Automation Letters},\n number = {1}\n}
\n
\n\n\n
\n Model predictive control (MPC) has been successful in applications involving the control of complex physical systems. This class of controllers leverages the information provided by an approximate model of the system's dynamics to simulate the effect of control actions. MPC methods also present a few hyper-parameters which may require a relatively expensive tuning process by demanding interactions with the physical system. Therefore, we investigate fine-tuning MPC methods in the context of stochastic MPC, which presents extra challenges due to the randomness of the controller's actions. In these scenarios, performance outcomes present noise, which is not homogeneous across the domain of possible hyper-parameter settings, but which varies in an input-dependent way. To address these issues, we propose a Bayesian optimisation framework that accounts for heteroscedastic noise to tune hyper-parameters in control problems. Empirical results on benchmark continuous control tasks and a physical robot support the proposed framework's suitability relative to baselines, which do not take heteroscedasticity into account.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Non-Volume Preserving Hamiltonian Monte Carlo and No-U-Turn Samplers.\n \n \n \n\n\n \n Afshar, H., M.; Oliveira, R.; and Cripps, S.\n\n\n \n\n\n\n In Proceedings of the 24th International Conference on Artificial Intelligence and Statistics (AISTATS) 2021, volume 130, 2021. PMLR\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Non-Volume Preserving Hamiltonian Monte Carlo and No-U-Turn Samplers},\n type = {inproceedings},\n year = {2021},\n volume = {130},\n publisher = {PMLR},\n id = {158db3fa-b46a-38cf-89d2-4b629d6deceb},\n created = {2021-04-28T13:24:45.801Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2022-10-23T08:46:27.752Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Afshar2021},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Afshar, Hadi Mohasel and Oliveira, Rafael and Cripps, Sally},\n booktitle = {Proceedings of the 24th International Conference on Artificial Intelligence and Statistics (AISTATS) 2021},\n keywords = {Bayesian inference,Hamiltonian Monte Carlo,MCMC}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n No-Regret Approximate Inference via Bayesian Optimisation.\n \n \n \n\n\n \n Oliveira, R.; Ott, L.; and Ramos, F.\n\n\n \n\n\n\n In 37th Conference on Uncertainty in Artificial Intelligence (UAI 2021), 2021. PMLR\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {No-Regret Approximate Inference via Bayesian Optimisation},\n type = {inproceedings},\n year = {2021},\n publisher = {PMLR},\n id = {78882d95-882e-39da-8fe0-ac7b99047d76},\n created = {2021-07-27T08:59:11.454Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2022-08-21T04:45:03.930Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2021},\n notes = {If instead of log-likelihood we use likelihood directly, a noisy observation could be a class label directly in a classification problem. Then we can use a framework, like BORE.},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Ott, Lionel and Ramos, Fabio},\n booktitle = {37th Conference on Uncertainty in Artificial Intelligence (UAI 2021)},\n keywords = {Bayesian inference,Bayesian optimisation,Gaussian processes,RKHS,approximate inference,batch evaluations,regret bounds}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Dual Online Stein Variational Inference for Control and Dynamics.\n \n \n \n\n\n \n Barcelos, L.; Lambert, A.; Oliveira, R.; Borges, P.; Boots, B.; and Ramos, F.\n\n\n \n\n\n\n In Proceedings of Robotics: Science and Systems, 7 2021. \n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Dual Online Stein Variational Inference for Control and Dynamics},\n type = {inproceedings},\n year = {2021},\n month = {7},\n city = {Virtual},\n id = {5a0e9f55-fa0b-3869-b7d2-9db3648ef7be},\n created = {2021-08-11T06:35:03.538Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2022-10-23T08:46:28.380Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Barcelos-RSS-21},\n source_type = {inproceedings},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Barcelos, Lucas and Lambert, Alexander and Oliveira, Rafael and Borges, Paulo and Boots, Byron and Ramos, Fabio},\n doi = {10.15607/RSS.2021.XVII.068},\n booktitle = {Proceedings of Robotics: Science and Systems},\n keywords = {Stein's method,approximate inference,control systems,model predictive control,robotics,variational inference}\n}
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2020\n \n \n (6)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Bayesian Optimisation for Safe Navigation Under Localisation Uncertainty.\n \n \n \n \n\n\n \n Oliveira, R.; Ott, L.; Guizilini, V.; and Ramos, F.\n\n\n \n\n\n\n Robotics Research: The 18th International Symposium ISRR, pages 489-504. Amato, N., M.; Hager, G.; Thomas, S.; and Torres-Torriti, M., editor(s). Springer International Publishing, 2020.\n \n\n\n\n
\n\n\n\n \n \n \"RoboticsWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n\n\n\n
\n 
@inbook{\n type = {inbook},\n year = {2020},\n pages = {489-504},\n websites = {https://link.springer.com/chapter/10.1007/978-3-030-28619-4_37},\n publisher = {Springer International Publishing},\n city = {Cham},\n id = {c513ae20-a872-31f3-ac17-b5a45a0f5b3a},\n created = {2020-01-15T00:00:13.578Z},\n file_attached = {false},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-05-10T15:32:09.412Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {10.1007/978-3-030-28619-4_37},\n source_type = {inproceedings},\n private_publication = {false},\n abstract = {In outdoor environments, mobile robots are required to navigate through terrain with varying characteristics, some of which might significantly affect the integrity of the platform. Ideally, the robot should be able to identify areas that are safe for navigation based on its own percepts about the environment while avoiding damage to itself. Bayesian optimisation (BO) has been successfully applied to the task of learning a model of terrain traversability while guiding the robot through more traversable areas. An issue, however, is that localisation uncertainty can end up guiding the robot to unsafe areas and distort the model being learnt. In this paper, we address this problem and present a novel method that allows BO to consider localisation uncertainty by applying a Gaussian process model for uncertain inputs as a prior. We evaluate the proposed method in simulation and in experiments with a real robot navigating over rough terrain and compare it against standard BO methods.},\n bibtype = {inbook},\n author = {Oliveira, Rafael and Ott, Lionel and Guizilini, Vitor and Ramos, Fabio},\n editor = {Amato, Nancy M and Hager, Greg and Thomas, Shawna and Torres-Torriti, Miguel},\n chapter = {Bayesian Optimisation for Safe Navigation Under Localisation Uncertainty},\n title = {Robotics Research: The 18th International Symposium ISRR},\n keywords = {Bayesian optimisation,uncertain inputs}\n}
\n
\n\n\n
\n In outdoor environments, mobile robots are required to navigate through terrain with varying characteristics, some of which might significantly affect the integrity of the platform. Ideally, the robot should be able to identify areas that are safe for navigation based on its own percepts about the environment while avoiding damage to itself. Bayesian optimisation (BO) has been successfully applied to the task of learning a model of terrain traversability while guiding the robot through more traversable areas. An issue, however, is that localisation uncertainty can end up guiding the robot to unsafe areas and distort the model being learnt. In this paper, we address this problem and present a novel method that allows BO to consider localisation uncertainty by applying a Gaussian process model for uncertain inputs as a prior. We evaluate the proposed method in simulation and in experiments with a real robot navigating over rough terrain and compare it against standard BO methods.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n DISCO: Double likelihood-free Inference Stochastic Control.\n \n \n \n \n\n\n \n Barcelos, L.; Oliveira, R.; Possas, R.; Ott, L.; and Ramos, F.\n\n\n \n\n\n\n In 2020 IEEE International Conference on Robotics and Automation (ICRA), 2020. IEEE\n \n\n\n\n
\n\n\n\n \n \n \"DISCO:Website\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {DISCO: Double likelihood-free Inference Stochastic Control},\n type = {inproceedings},\n year = {2020},\n websites = {http://arxiv.org/abs/2002.07379},\n publisher = {IEEE},\n city = {Paris, France},\n id = {7d194e68-57cc-366a-a9c1-b56ee6959568},\n created = {2020-02-11T07:34:39.884Z},\n file_attached = {false},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-10-18T23:55:28.036Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Barcelos2020},\n private_publication = {false},\n abstract = {Accurate simulation of complex physical systems enables the development, testing, and certification of control strategies before they are deployed into the real systems. As simulators become more advanced, the analytical tractability of the differential equations and associated numerical solvers incorporated in the simulations diminishes, making them difficult to analyse. A potential solution is the use of probabilistic inference to assess the uncertainty of the simulation parameters given real observations of the system. Unfortunately the likelihood function required for inference is generally expensive to compute or totally intractable. In this paper we propose to leverage the power of modern simulators and recent techniques in Bayesian statistics for likelihood-free inference to design a control framework that is efficient and robust with respect to the uncertainty over simulation parameters. The posterior distribution over simulation parameters is propagated through a potentially non-analytical model of the system with the unscented transform, and a variant of the information theoretical model predictive control. This approach provides a more efficient way to evaluate trajectory roll outs than Monte Carlo sampling, reducing the online computation burden. Experiments show that the controller proposed attained superior performance and robustness on classical control and robotics tasks when compared to models not accounting for the uncertainty over model parameters.},\n bibtype = {inproceedings},\n author = {Barcelos, Lucas and Oliveira, Rafael and Possas, Rafael and Ott, Lionel and Ramos, Fabio},\n booktitle = {2020 IEEE International Conference on Robotics and Automation (ICRA)},\n keywords = {approximate inference,model predictive control}\n}
\n
\n\n\n
\n Accurate simulation of complex physical systems enables the development, testing, and certification of control strategies before they are deployed into the real systems. As simulators become more advanced, the analytical tractability of the differential equations and associated numerical solvers incorporated in the simulations diminishes, making them difficult to analyse. A potential solution is the use of probabilistic inference to assess the uncertainty of the simulation parameters given real observations of the system. Unfortunately the likelihood function required for inference is generally expensive to compute or totally intractable. In this paper we propose to leverage the power of modern simulators and recent techniques in Bayesian statistics for likelihood-free inference to design a control framework that is efficient and robust with respect to the uncertainty over simulation parameters. The posterior distribution over simulation parameters is propagated through a potentially non-analytical model of the system with the unscented transform, and a variant of the information theoretical model predictive control. This approach provides a more efficient way to evaluate trajectory roll outs than Monte Carlo sampling, reducing the online computation burden. Experiments show that the controller proposed attained superior performance and robustness on classical control and robotics tasks when compared to models not accounting for the uncertainty over model parameters.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Active Learning of Conditional Mean Embeddings via Bayesian Optimisation.\n \n \n \n \n\n\n \n Chowdhury, S., R.; Oliveira, R.; and Ramos, F.\n\n\n \n\n\n\n In Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence (UAI), 2020. PMLR volume 124\n \n\n\n\n
\n\n\n\n \n \n \"ActiveWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Active Learning of Conditional Mean Embeddings via Bayesian Optimisation},\n type = {inproceedings},\n year = {2020},\n websites = {http://auai.org/uai2020/accepted.php},\n publisher = {PMLR volume 124},\n city = {Toronto, Canada},\n id = {02c0806f-4084-3f0a-bee8-ec0278268799},\n created = {2020-07-01T14:29:49.080Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-07-30T04:56:03.766Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Chowdhury2020},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Chowdhury, Sayak Ray and Oliveira, Rafael and Ramos, Fabio},\n booktitle = {Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence (UAI)},\n keywords = {Bayesian optimisation,RKHS,approximate inference,kernel embeddings,regret bounds}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Sparse Spectrum Warped Input Measures for Nonstationary Kernel Learning.\n \n \n \n\n\n \n Tompkins, A.; Oliveira, R.; and Ramos, F.\n\n\n \n\n\n\n In 34th Conference on Neural Information Processing Systems (NeurIPS 2020), 2020. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Sparse Spectrum Warped Input Measures for Nonstationary Kernel Learning},\n type = {inproceedings},\n year = {2020},\n id = {42229615-87bc-3458-a05a-68a42bdbc510},\n created = {2020-10-18T23:41:29.267Z},\n file_attached = {false},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2021-04-07T04:12:05.125Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {tompkins2020sparse},\n source_type = {article},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Tompkins, Anthony and Oliveira, Rafael and Ramos, Fabio},\n booktitle = {34th Conference on Neural Information Processing Systems (NeurIPS 2020)},\n keywords = {Fourier features,Gaussian processes,non-stationary processes,sparse GPs,uncertain inputs}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Online BayesSim for combined simulator parameter inference and policy improvement.\n \n \n \n\n\n \n Possas, R.; Barcelos, L.; Oliveira, R.; Fox, D.; and Ramos, F.\n\n\n \n\n\n\n In IEEE International Conference on Intelligent Robots and Systems, pages 5445-5452, 2020. \n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n  \n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Online BayesSim for combined simulator parameter inference and policy improvement},\n type = {inproceedings},\n year = {2020},\n pages = {5445-5452},\n id = {d55ab184-03e1-3508-bc19-d6ec9733af6a},\n created = {2021-04-28T13:24:45.767Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2022-09-07T14:57:24.871Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Possas2020},\n private_publication = {false},\n abstract = {Recent advancements in Bayesian likelihood-free inference enables a probabilistic treatment for the problem of estimating simulation parameters and their uncertainty given sequences of observations. Domain randomization can be performed much more effectively when a posterior distribution provides the correct uncertainty over parameters in a simulated environment. In this paper, we study the integration of simulation parameter inference with both model-free reinforcement learning and model-based control in a novel sequential algorithm that alternates between learning a better estimation of parameters and improving the controller. This approach exploits the interdependence between the two problems to generate computational efficiencies and improved reliability when a black-box simulator is available. Experimental results suggest that both control strategies have better performance when compared to traditional domain randomization methods.},\n bibtype = {inproceedings},\n author = {Possas, Rafael and Barcelos, Lucas and Oliveira, Rafael and Fox, Dieter and Ramos, Fabio},\n doi = {10.1109/IROS45743.2020.9341401},\n booktitle = {IEEE International Conference on Intelligent Robots and Systems}\n}
\n
\n\n\n
\n Recent advancements in Bayesian likelihood-free inference enables a probabilistic treatment for the problem of estimating simulation parameters and their uncertainty given sequences of observations. Domain randomization can be performed much more effectively when a posterior distribution provides the correct uncertainty over parameters in a simulated environment. In this paper, we study the integration of simulation parameter inference with both model-free reinforcement learning and model-based control in a novel sequential algorithm that alternates between learning a better estimation of parameters and improving the controller. This approach exploits the interdependence between the two problems to generate computational efficiencies and improved reliability when a black-box simulator is available. Experimental results suggest that both control strategies have better performance when compared to traditional domain randomization methods.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n No-Regret reinforcement learning with value function approximation: a kernel embedding approach.\n \n \n \n\n\n \n Chowdhury, S., R.; and Oliveira, R.\n\n\n \n\n\n\n arXiv. 2020.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@article{\n title = {No-Regret reinforcement learning with value function approximation: a kernel embedding approach},\n type = {article},\n year = {2020},\n keywords = {Kernel mean embedding,Model-based RL,Value function approximation},\n id = {3b618603-efbe-36ec-9e86-1ef672971abc},\n created = {2021-04-28T13:24:45.964Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2023-09-26T09:51:20.409Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Chowdhury2020a},\n private_publication = {false},\n abstract = {We consider the regret minimization problem in reinforcement learning (RL) in the episodic setting. In many real-world RL environments, the state and action spaces are continuous or very large. Existing approaches establish regret guarantees by either a low-dimensional representation of the stochastic transition model or an approximation of the Q-functions. However, the understanding of function approximation schemes for state-value functions largely remains missing. In this paper, we propose an online model-based RL algorithm, namely the CME-RL, that learns representations of transition distributions as embeddings in a reproducing kernel Hilbert space while carefully balancing the exploitation-exploration tradeoff. We demonstrate the efficiency of our algorithm by proving a frequentist (worst-case) regret bound that is of order Oõ(HγN√N)1, where H is the episode length, N is the total number of time steps and γN is an information theoretic quantity relating the effective dimension of the state-action feature space. Our method bypasses the need for estimating transition probabilities and applies to any domain on which kernels can be defined. It also brings new insights into the general theory of kernel methods for approximate inference and RL regret minimization.},\n bibtype = {article},\n author = {Chowdhury, Sayak Ray and Oliveira, Rafael},\n journal = {arXiv}\n}
\n
\n\n\n
\n We consider the regret minimization problem in reinforcement learning (RL) in the episodic setting. In many real-world RL environments, the state and action spaces are continuous or very large. Existing approaches establish regret guarantees by either a low-dimensional representation of the stochastic transition model or an approximation of the Q-functions. However, the understanding of function approximation schemes for state-value functions largely remains missing. In this paper, we propose an online model-based RL algorithm, namely the CME-RL, that learns representations of transition distributions as embeddings in a reproducing kernel Hilbert space while carefully balancing the exploitation-exploration tradeoff. We demonstrate the efficiency of our algorithm by proving a frequentist (worst-case) regret bound that is of order Oõ(HγN√N)1, where H is the episode length, N is the total number of time steps and γN is an information theoretic quantity relating the effective dimension of the state-action feature space. Our method bypasses the need for estimating transition probabilities and applies to any domain on which kernels can be defined. It also brings new insights into the general theory of kernel methods for approximate inference and RL regret minimization.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2019\n \n \n (2)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Bayesian optimisation under uncertain inputs.\n \n \n \n \n\n\n \n Oliveira, R.; Ott, L.; and Ramos, F.\n\n\n \n\n\n\n In 22nd International Conference on Artificial Intelligence and Statistics (AISTATS), 2019. PMLR\n \n\n\n\n
\n\n\n\n \n \n \"BayesianWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Bayesian optimisation under uncertain inputs},\n type = {inproceedings},\n year = {2019},\n websites = {http://proceedings.mlr.press/v89/oliveira19a.html},\n publisher = {PMLR},\n city = {Naha, Okinawa, Japan},\n id = {7a75af4c-61a2-3bd1-a415-edae7cd72bd4},\n created = {2019-04-03T03:21:05.581Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-05-10T15:32:09.470Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2019},\n private_publication = {false},\n abstract = {Bayesian optimisation (BO) has been a successful approach to optimise functions which are expensive to evaluate and whose observations are noisy. Classical BO algorithms, however, do not account for errors about the location where observations are taken, which is a common issue in problems with physical components. In these cases, the estimation of the actual query location is also subject to uncertainty. In this context, we propose an upper confidence bound (UCB) algorithm for BO problems where both the outcome of a query and the true query location are uncertain. The algorithm employs a Gaussian process model that takes probability distributions as inputs. Theoretical results are provided for both the proposed algorithm and a conventional UCB approach within the uncertain-inputs setting. Finally, we evaluate each method's performance experimentally, comparing them to other input noise aware BO approaches on simulated scenarios involving synthetic and real data.},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Ott, Lionel and Ramos, Fabio},\n booktitle = {22nd International Conference on Artificial Intelligence and Statistics (AISTATS)},\n keywords = {Bayesian optimisation,Gaussian processes,RKHS,regret bounds,uncertain inputs}\n}
\n
\n\n\n
\n Bayesian optimisation (BO) has been a successful approach to optimise functions which are expensive to evaluate and whose observations are noisy. Classical BO algorithms, however, do not account for errors about the location where observations are taken, which is a common issue in problems with physical components. In these cases, the estimation of the actual query location is also subject to uncertainty. In this context, we propose an upper confidence bound (UCB) algorithm for BO problems where both the outcome of a query and the true query location are uncertain. The algorithm employs a Gaussian process model that takes probability distributions as inputs. Theoretical results are provided for both the proposed algorithm and a conventional UCB approach within the uncertain-inputs setting. Finally, we evaluate each method's performance experimentally, comparing them to other input noise aware BO approaches on simulated scenarios involving synthetic and real data.\n
\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n \n Distributional Bayesian optimisation for variational inference on black-box simulators.\n \n \n \n \n\n\n \n Oliveira, R.; Ott, L.; and Ramos, F.\n\n\n \n\n\n\n In 2nd Symposium on Advances in Approximate Bayesian Inference, 2019. \n \n\n\n\n
\n\n\n\n \n \n \"DistributionalWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Distributional Bayesian optimisation for variational inference on black-box simulators},\n type = {inproceedings},\n year = {2019},\n websites = {https://openreview.net/forum?id=HJxvcJhVYS},\n city = {Vancouver, Canada},\n id = {0f594471-54d2-30ba-9a7f-891f9cdc43da},\n created = {2019-11-29T07:27:31.445Z},\n file_attached = {false},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2024-11-22T03:13:43.883Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2019},\n private_publication = {false},\n abstract = {Inverse problems are ubiquitous in natural sciences and refer to the challenging task of inferring complex and potentially multi-modal posterior distributions over hidden param- eters given a set of observations. Typically, a model of the physical process in the form of differential equations is available but leads to intractable inference over its parameters. While the forward propagation of parameters through the model simulates the evolution of the system, the inverse problem of finding the parameters given the sequence of states is not unique. In this work, we propose a generalisation of the Bayesian optimisation framework to approximate inference of simulator parameters. To this end, we devise distributional Bayesian optimisation as an algorithm that optimises a black-box function with respect to distribution inputs. In experiments, we demonstrate the method in inference tasks with a reinforcement learning environment.},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Ott, Lionel and Ramos, Fabio},\n booktitle = {2nd Symposium on Advances in Approximate Bayesian Inference},\n keywords = {Bayesian optimisation,Fourier features,Gaussian processes,batch evaluations,variational inference}\n}
\n
\n\n\n
\n Inverse problems are ubiquitous in natural sciences and refer to the challenging task of inferring complex and potentially multi-modal posterior distributions over hidden param- eters given a set of observations. Typically, a model of the physical process in the form of differential equations is available but leads to intractable inference over its parameters. While the forward propagation of parameters through the model simulates the evolution of the system, the inverse problem of finding the parameters given the sequence of states is not unique. In this work, we propose a generalisation of the Bayesian optimisation framework to approximate inference of simulator parameters. To this end, we devise distributional Bayesian optimisation as an algorithm that optimises a black-box function with respect to distribution inputs. In experiments, we demonstrate the method in inference tasks with a reinforcement learning environment.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2018\n \n \n (2)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Learning to Race through Coordinate Descent Bayesian Optimisation.\n \n \n \n \n\n\n \n Oliveira, R.; Rocha, F., H., M.; Ott, L.; Guizilini, V.; Ramos, F.; and Grassi, V.\n\n\n \n\n\n\n In IEEE International Conference on Robotics and Automation (ICRA), pages 6431-6438, 2018. \n \n\n\n\n
\n\n\n\n \n \n \"LearningWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Learning to Race through Coordinate Descent Bayesian Optimisation},\n type = {inproceedings},\n year = {2018},\n pages = {6431-6438},\n websites = {https://ieeexplore.ieee.org/document/8460735},\n city = {Brisbane, Australia},\n id = {9ad78292-6903-3460-af5d-8a220366bfbb},\n created = {2018-03-07T04:09:24.356Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-07-01T16:04:20.142Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2018},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Rocha, Fernando H M and Ott, Lionel and Guizilini, Vitor and Ramos, Fabio and Grassi, Valdir},\n booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},\n keywords = {Bayesian optimisation,Brazilian institution,high-dimensional models}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Leveraging Localisation Information into Bayesian Optimisation for Planning in Robotics.\n \n \n \n\n\n \n Oliveira, R.; Ott, L.; Guizilini, V.; and Ramos, F.\n\n\n \n\n\n\n In ICRA Workshop on Informative Path Planning and Adaptive Sampling (WIPPAS), 2018. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Leveraging Localisation Information into Bayesian Optimisation for Planning in Robotics},\n type = {inproceedings},\n year = {2018},\n city = {Brisbane, Australia},\n id = {a2124f37-2f83-34c6-80d3-2a05f6b7babe},\n created = {2018-05-30T12:21:23.975Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-05-10T15:33:29.423Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2018},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Ott, Lionel and Guizilini, Vitor and Ramos, Fabio},\n booktitle = {ICRA Workshop on Informative Path Planning and Adaptive Sampling (WIPPAS)},\n keywords = {Bayesian optimisation,Gaussian processes,field robotics,mobile robotics,regret bounds,uncertain inputs}\n}
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2017\n \n \n (2)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n \n Bayesian Optimisation for Safe Navigation under Localisation Uncertainty.\n \n \n \n \n\n\n \n Oliveira, R.; Ott, L.; Guizlini, V.; and Ramos, F.\n\n\n \n\n\n\n In 18th International Symposium on Robotics Research (ISRR), 2017. \n \n\n\n\n
\n\n\n\n \n \n \"BayesianWebsite\n  \n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Bayesian Optimisation for Safe Navigation under Localisation Uncertainty},\n type = {inproceedings},\n year = {2017},\n websites = {https://arxiv.org/abs/1709.02169},\n city = {Puerto Varas, Chile},\n id = {96834f69-77e8-3ec6-b947-7174c9b40bf1},\n created = {2020-05-10T15:32:08.122Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-08-19T05:31:00.697Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2017},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Ott, Lionel and Guizlini, Vitor and Ramos, Fabio},\n booktitle = {18th International Symposium on Robotics Research (ISRR)},\n keywords = {Bayesian optimisation,Gaussian processes,mobile robotics,robotics,uncertain inputs}\n}
\n
\n\n\n\n
\n\n\n
\n \n\n \n \n \n \n \n Bayesian optimisation for safe navigation under localisation uncertainty.\n \n \n \n\n\n \n Oliveira, R.; Ott, L.; Guizilini, V.; and Ramos, F.\n\n\n \n\n\n\n 2017.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n  \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n\n\n\n
\n 
@misc{\n title = {Bayesian optimisation for safe navigation under localisation uncertainty},\n type = {misc},\n year = {2017},\n source = {arXiv},\n id = {8cbef92d-a44c-377c-b7bb-0264d294715e},\n created = {2020-10-27T23:59:00.000Z},\n file_attached = {false},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2022-03-22T07:06:03.764Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {false},\n hidden = {false},\n citation_key = {Oliveira2017},\n private_publication = {true},\n abstract = {Copyright © 2017, arXiv, All rights reserved. In outdoor environments, mobile robots are required to navigate through terrain with varying characteristics, some of which might significantly affect the integrity of the platform. Ideally, the robot should be able to identify areas that are safe for navigation based on its own percepts about the environment while avoiding damage to itself. Bayesian optimisation (BO) has been successfully applied to the task of learning a model of terrain traversability while guiding the robot through more traversable areas. An issue, however, is that localisation uncertainty can end up guiding the robot to unsafe areas and distort the model being learnt. In this paper, we address this problem and present a novel method that allows BO to consider localisation uncertainty by applying a Gaussian process model for uncertain inputs as a prior. We evaluate the proposed method in simulation and in experiments with a real robot navigating over rough terrain and compare it against standard BO methods.},\n bibtype = {misc},\n author = {Oliveira, R. and Ott, L. and Guizilini, V. and Ramos, F.}\n}
\n
\n\n\n
\n Copyright © 2017, arXiv, All rights reserved. In outdoor environments, mobile robots are required to navigate through terrain with varying characteristics, some of which might significantly affect the integrity of the platform. Ideally, the robot should be able to identify areas that are safe for navigation based on its own percepts about the environment while avoiding damage to itself. Bayesian optimisation (BO) has been successfully applied to the task of learning a model of terrain traversability while guiding the robot through more traversable areas. An issue, however, is that localisation uncertainty can end up guiding the robot to unsafe areas and distort the model being learnt. In this paper, we address this problem and present a novel method that allows BO to consider localisation uncertainty by applying a Gaussian process model for uncertain inputs as a prior. We evaluate the proposed method in simulation and in experiments with a real robot navigating over rough terrain and compare it against standard BO methods.\n
\n\n\n
\n\n\n\n\n\n
\n
\n\n
\n
\n  \n 2016\n \n \n (1)\n \n \n
\n
\n \n \n
\n \n\n \n \n \n \n \n Active Perception for Modelling Energy Consumption in Off-Road Navigation.\n \n \n \n\n\n \n Oliveira, R.; Ott, L.; and Ramos, F.\n\n\n \n\n\n\n In Australasian Conference on Robotics and Automation (ACRA), 2016. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n  \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n
\n 
@inproceedings{\n title = {Active Perception for Modelling Energy Consumption in Off-Road Navigation},\n type = {inproceedings},\n year = {2016},\n city = {Brisbane, Australia},\n id = {9aa0729a-1bf5-3fc0-891a-f807f47f7473},\n created = {2017-06-15T11:02:31.934Z},\n file_attached = {true},\n profile_id = {20eea928-2566-3876-901d-9a50fe4a71d0},\n last_modified = {2020-05-10T15:33:29.425Z},\n read = {true},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Oliveira2016},\n private_publication = {false},\n bibtype = {inproceedings},\n author = {Oliveira, Rafael and Ott, Lionel and Ramos, Fabio},\n booktitle = {Australasian Conference on Robotics and Automation (ACRA)},\n keywords = {Bayesian optimisation,Gaussian processes,energy-cost estimation,mobile robotics,path-planning,terrain modelling,traversability estimation}\n}
\n
\n\n\n\n
\n\n\n\n\n\n
\n
\n\n\n\n\n
\n\n\n \n\n \n \n \n \n\n
\n"}; document.write(bibbase_data.data);