Peer-reviewed publications in machine learning, reinforcement learning, and probabilistic modeling.
State estimation in non-linear dynamical systems is challenging due to high computational costs and approximation errors. We propose an iterative algorithm based on approximate expectation propagation that provides accurate state estimates while being computationally efficient.
We propose copula flows, a method for generating synthetic data that preserves complex dependencies between variables. This approach is particularly useful for privacy-preserving machine learning applications.
Trial-and-error based reinforcement learning (RL) has seen rapid advancements in recent times, especially with the advent of deep neural networks. However, the majority of autonomous RL algorithms require a large number of interactions with the environment. We propose a model-based RL framework based on probabilistic Model Predictive Control (MPC) that achieves state-of-the-art data efficiency.
Multi-modal densities appear frequently in time series and practical applications. We devise a non-linear filtering algorithm where densities are represented by Gaussian mixture models, whose parameters are estimated in closed form. The resulting method exhibits superior performance on nonlinear benchmarks.
To engage in cooperative activities with human partners, robots have to possess basic interactive abilities and skills. We introduce Interaction Primitives, a representation that builds on dynamic motor primitives by maintaining a distribution over parameters to learn inherent correlations of cooperative activities.
For a complete list of my publications and citations, visit my Google Scholar profile .