People

Denoising autoencoders (DAEs) have proven useful for unsupervised representation learning, but a thorough theoretical understanding is still lacking of how the input noise influences learning. Here we develop theory for how noise influences learning in DAEs. By focusing on linear DAEs, we are able to derive analytic expressions that exactly describe their learning dynamics. We verify our theoretical predictions with simulations as well as experiments on MNIST and CIFAR-10. The theory illustrates how, when tuned correctly, noise allows DAEs to ignore low variance directions in the inputs while learning to reconstruct them. Furthermore, in a comparison of the learning dynamics of DAEs to standard regularised autoencoders, we show that noise has a similar regularisation effect to weight decay, but with faster training dynamics. We also show that our theoretical predictions approximate learning dynamics on real-world data and qualitatively match observed dynamics in nonlinear DAEs.

@{202,
  author = {Arnu Pretorius, Steve Kroon, H. Kamper},
  title = {Learning Dynamics of Linear Denoising Autoencoders},
  abstract = {Denoising autoencoders (DAEs) have proven useful for unsupervised representation learning, but a thorough theoretical understanding is still lacking of how the input noise influences learning. Here we develop theory for how noise influences learning in DAEs. By focusing on linear DAEs, we are able to derive analytic expressions that exactly describe their learning dynamics. We verify our theoretical predictions with simulations as well as experiments on MNIST and CIFAR-10. The theory illustrates how, when tuned correctly, noise allows DAEs to ignore low variance directions in the inputs while learning to reconstruct them. Furthermore, in a comparison of the learning dynamics of DAEs to standard regularised autoencoders, we show that noise has a similar regularisation effect to weight decay, but with faster training dynamics. We also show that our theoretical predictions approximate learning dynamics on real-world data and qualitatively match observed dynamics in nonlinear DAEs.},
  year = {2018},
  journal = {35th International Conference on Machine Learning},
  pages = {4141-4150},
  month = {10/07-15/07},
  publisher = {Proceedings of Machine Learning Research (PMLR)},
  address = {Sweden},
  isbn = {1938-7228},
}

No Abstract

@article{152,
  author = {Steve Kroon, A. Heavens, Y. Fantaye, E. Sellentin, H. Eggers, Z. Hosenie, A. Mootoovaloo},
  title = {No evidence for extensions to the standard cosmological model},
  abstract = {No Abstract},
  year = {2017},
  journal = {Physical Review Letters},
  volume = {119},
  pages = {101301-101305},
  issue = {2017},
  publisher = {American Physical Society},
  url = {https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.119.101301},
}

Reinforcement Learning (RL) is a powerful technique to develop intelligent agents in the field of Artificial Intelligence (AI). This paper proposes a new RL algorithm called the Temporal-Difference value iteration algorithm with state-value functions and presents applications of this algorithm to the decision-making problems challenged in the RoboCup Small Size League (SSL) domain. Six scenarios were defined to develop shooting skills for an SSL soccer robot in various situations using the proposed algorithm. Furthermore, an Artificial Neural Network (ANN) model, namely Multi-Layer Perceptron (MLP) was used as a function approximator in each application. The experimental results showed that the proposed RL algorithm had effectively trained the RL agent to acquire good shooting skills. The RL agent showed good performance under specified experimental conditions.

@article{151,
  author = {Steve Kroon, M. Yoon, J. Bekker},
  title = {New reinforcement learning algorithm for robot soccer},
  abstract = {Reinforcement Learning (RL) is a powerful technique to develop intelligent agents in the field of Artificial Intelligence (AI). This paper proposes a new RL algorithm called the Temporal-Difference value iteration algorithm with state-value functions and presents applications of this algorithm to the decision-making problems challenged in the RoboCup Small Size League (SSL) domain. Six scenarios were defined to develop shooting skills for an SSL soccer robot in various situations using the proposed algorithm. Furthermore, an Artificial Neural Network (ANN) model, namely Multi-Layer Perceptron (MLP) was used as a function approximator in each application. The experimental results showed that the proposed RL algorithm had effectively trained the RL agent to acquire good shooting skills. The RL agent showed good performance under specified experimental conditions.},
  year = {2017},
  journal = {Orion},
  volume = {33},
  pages = {1-20},
  issue = {1},
  publisher = {Operations Research Society of South Africa (ORSSA)},
  address = {South Africa},
  isbn = {2224-0004 (online)},
  url = {http://orion.journals.ac.za/pub/article/view/542},
}

No Abstract

@{143,
  author = {Steve Kroon, PB Le Roux, Willem Bester},
  title = {DSaaS:  A cloud Service for Persistent Data Structures},
  abstract = {No Abstract},
  year = {2016},
  journal = {CLOSER, 6th International Conference on Cloud Computing and Services Science},
  pages = {37-48},
  month = {23/04-25/04},
  address = {Portugal},
  isbn = {978-989-758-182-3},
}

No Abstract

@{129,
  author = {Steve Kroon, S. Nienaber, M.J. Booysen},
  title = {A Comparison of Low-Cost Monocular Vision Techniques for Pothole Distance Estimation},
  abstract = {No Abstract},
  year = {2015},
  journal = {IEEE Symposium Series on Computational Intelligence: IEEE Symposium on Computational Intelligence in Vehicles and Transportation Systems},
  pages = {419-426},
  month = {08/12-10/12},
}