B
Benjamin Rosman
Researcher at University of the Witwatersrand
Publications - 110
Citations - 1164
Benjamin Rosman is an academic researcher from University of the Witwatersrand. The author has contributed to research in topics: Reinforcement learning & Computer science. The author has an hindex of 15, co-authored 88 publications receiving 792 citations. Previous affiliations of Benjamin Rosman include University of Edinburgh & Council of Scientific and Industrial Research.
Papers
More filters
Journal ArticleDOI
Learning spatial relationships between objects
TL;DR: An algorithm which redescribes a scene in terms of a layered representation, from labeled point clouds of the objects in the scene, which provides symbolic meaning to the inter-object relationships useful for subsequent commonsense reasoning and decision making is presented.
Journal ArticleDOI
Bayesian policy reuse
TL;DR: The problem of policy reuse is formalised and an algorithm for efficiently responding to a novel task instance by reusing a policy from this library of existing policies, where the choice is based on observed ‘signals’ which correlate to policy performance is presented.
Proceedings ArticleDOI
Fingerprint minutiae extraction using deep learning
TL;DR: A deep neural network is proposed — MENet, for Minutiae Extraction Network — to learn a data-driven representation of minutiae points and established a voting scheme to construct training data, and it is shown that MENet performs favourably in comparisons against existingminutiae extractors.
Proceedings ArticleDOI
Nonparametric Bayesian reward segmentation for skill discovery using inverse reinforcement learning
TL;DR: This work uses a Bayesian nonparametric approach to propose skill segmentations and maximum entropy inverse reinforcement learning to infer reward functions from the segments, and produces a set of Markov Decision Processes that best describe the input trajectories.
Proceedings Article
Belief reward shaping in reinforcement learning
Ofir Marom,Benjamin Rosman +1 more
TL;DR: This work presents a novel Bayesian reward shaping framework that augments the reward distribution with prior beliefs that decay with experience and proves that under suitable conditions a Markov decision process augmented with this framework is consistent with the optimal policy of the original MDP when using the Q-learning algorithm.