Noel E. O'Connor
21 Papers
1 Citations
Noel E. O'Connor is an academic researcher. The author has contributed to research in topics: Computer science & Engineering. The author has an hindex of 1, co-authored 15 publications.
Chat about Author
Papers
Dexterous robotic manipulation using deep reinforcement learning and knowledge transfer for complex sparse reward‐based tasks
Qiang Wang,Francisco Roldán Sánchez,Robert J. McCarthy,David Cordova Bulens,Kevin McGuinness,Noel E. O'Connor,M. Wuthrich,Felix Widmaier,Stefan Bauer,Stephen J. Redmond +9 more
TL;DR: A novel use of a Knowledge Transfer (KT) technique that allows the strategies learned by the agent in the original task to be transferred to this task (where orientation matters), which shows good generalisation properties and could be applied to any actor-critic learning algorithm.
Is your noise correction noisy? PLS: Robustness to label noise with two stage detection
Paul Albert,Eric Arazo,Tarun Kirshna,Noel E. O'Connor,Kevin McGuinness +4 more
- 10 Oct 2022
TL;DR: The pseudo-loss is proposed, a simple metric that is strongly correlated with pseudo-label correctness on noisy samples and dynamically down weight under-confident pseudo-labels throughout training to avoid con-rmation bias and improve the network accuracy.
6
Behaviour Discriminator: A Simple Data Filtering Method to Improve Offline Policy Learning
Qing Wang,Robert McCarthy,David Cordova Bulens,Kevin McGuinness,Noel E. O'Connor,Francisco Roldán Sánchez,Stephen J. Redmond +6 more
TL;DR: In this paper , the authors proposed a behaviour discriminator (BD) concept, a novel and simple data filtering approach based on semi-supervised learning, which can accurately discern expert data from a mixed-quality dataset.
Joint one‐sided synthetic unpaired image translation and segmentation for colorectal cancer prevention
TL;DR: This article proposed CUT-seg, a joint training approach where a segmentation model and a generative model are jointly trained to produce realistic images while learning to segment polyps.
3
Learning Saliency From Fixations
Yasser Abdelaziz Dahou Djilali,Kevin McGuinness,Noel E. O'Connor +2 more
- 23 Nov 2023
TL;DR: SalTR is a novel saliency prediction approach that leverages parallel decoding in transformers to learn saliency solely from fixation maps. It directly predicts fixations points using a global loss and cross-attention over image features. SalTR achieves comparable performance to state-of-the-art approaches on Salicon and MIT300 benchmarks.
2