@misc{mcalister2024improvedrobustnesshyperparameterselection, title={Improved Robustness and Hyperparameter Selection in Modern Hopfield Networks}, author={Hayden McAlister and Anthony Robins and Lech Szymanski}, year={2024}, eprint={2407.08742}, archivePrefix={arXiv}, primaryClass={cs.NE}, url={https://arxiv.org/abs/2407.08742}, }

@article{10.1162/neco_a_01704, author = {McAlister, Hayden and Robins, Anthony and Szymanski, Lech}, title = {Prototype Analysis in Hopfield Networks with Hebbian Learning}, journal = {Neural Computation}, volume = {36}, number = {11}, pages = {2322-2364}, year = {2024}, abstract = {We discuss prototype formation in the Hopfield network. Typically, Hebbian learning with highly correlated states leads to degraded memory performance. We show that this type of learning can lead to prototype formation, where unlearned states emerge as representatives of large correlated subsets of states, alleviating capacity woes. This process has similarities to prototype learning in human cognition. We provide a substantial literature review of prototype learning in associative memories, covering contributions from psychology, statistical physics, and computer science. We analyze prototype formation from a theoretical perspective and derive a stability condition for these states based on the number of examples of the prototype presented for learning, the noise in those examples, and the number of nonexample states presented. The stability condition is used to construct a probability of stability for a prototype state as the factors of stability change. We also note similarities to traditional network analysis, allowing us to find a prototype capacity. We corroborate these expectations of prototype formation with experiments using a simple Hopfield network with standard Hebbian learning. We extend our experiments to a Hopfield network trained on data with multiple prototypes and find the network is capable of stabilizing multiple prototypes concurrently. We measure the basins of attraction of the multiple prototype states, finding attractor strength grows with the number of examples and the agreement of examples. We link the stability and dominance of prototype states to the energy profile of these states, particularly when comparing the profile shape to target states or other spurious states.}, issn = {0899-7667}, doi = {10.1162/neco_a_01704}, url = {https://doi.org/10.1162/neco_a_01704}, eprint = {https://direct.mit.edu/neco/article-pdf/doi/10.1162/neco\_a\_01704/2468185/neco\_a\_01704.pdf}, }

@article{xu.etal:2023, author={Xu, Haitao and Szymanski, Lech and McCane, Brendan}, journal={IEEE Transactions on Neural Networks and Learning Systems}, title={VASE: Variational Assorted Surprise Exploration for Reinforcement Learning}, year={2023}, volume={34}, number={3}, pages={1243--1252}, url={https://doi.org/10.1109/TNNLS.2021.3105140}, doi={10.1109/TNNLS.2021.3105140} }

@article{Szymanski.etal:2021, title = {Conceptual complexity of neural networks}, journal = {Neurocomputing}, volume = {469}, pages = {52-64}, year = {2022}, issn = {0925-2312}, doi = {https://doi.org/10.1016/j.neucom.2021.10.063}, url = {https://doi.org/10.1016/j.neucom.2021.10.063}, author = {Lech Szymanski and Brendan McCane and Craig Atkinson}, keywords = {deep learning, learning theory, complexity measures}, }

@INPROCEEDINGS{9653248, author={Szymanski, Lech and Lee, Michael}, booktitle={International Conference on Image and Vision Computing New Zealand (IVCNZ)}, title={Coarse facial feature detection in sheep}, year={2021}, volume={}, number={}, pages={1-6}, doi={10.1109/IVCNZ54163.2021.9653248}, url={https://doi.org/10.1109/IVCNZ54163.2021.9653248} }

@article{atkinson2020pseudo, title = "Pseudo-rehearsal: Achieving deep reinforcement learning without catastrophic forgetting", journal = "Neurocomputing", volume = "428", pages = "291 - 307", year = "2021", issn = "0925-2312", doi = "https://doi.org/10.1016/j.neucom.2020.11.050", url = "http://www.sciencedirect.com/science/article/pii/S0925231220318439", author = "Craig Atkinson and Brendan McCane and Lech Szymanski and Anthony Robins", }

@article{xu.etal:2020, title={MIME: Mutual Information Minimisation Exploration}, author={Haitao Xu and Brendan McCane and Lech Szymanski and Craig Atkinson}, journal={arXiv preprint arXiv:2001.05636}, url={https://arxiv.org/abs/2001.05636}, year={2020} }

@INPROCEEDINGS{vanSintAnnaland.etal2020, author={Y. v. S. Annaland and L. Szymanski and S. Mills}, booktitle={2020 35th International Conference on Image and Vision Computing New Zealand (IVCNZ)}, title={Predicting Cherry Quality Using Siamese Networks}, year={2020}, pages={1-6}, doi={10.1109/IVCNZ51579.2020.9290674}, url={https://doi.org/10.1109/IVCNZ51579.2020.9290674} }

@INPROCEEDINGS{Szymanski.etal2020a, author={L. Szymanski and M. Lee}, booktitle={2020 35th International Conference on Image and Vision Computing New Zealand (IVCNZ)}, title={Deep Sheep: kinship assignment in livestock from facial images}, year={2020}, pages={1-6}, doi={10.1109/IVCNZ51579.2020.9290558}, url={https://doi.org/10.1109/IVCNZ51579.2020.9290558} }

@article{lechszym.etal:2020, title={Switched linear projections for neural network interpretability}, author={Szymanski, Lech and McCane, Brendan and Atkinson, Craig}, journal={arXiv preprint arXiv:1909.11275}, url={https://arxiv.org/abs/1909.11275}, year={2020} }

@article{atkinson.etal:2019, title={GRIm-RePR: Prioritising Generating Important Features for Pseudo-Rehearsal}, author={Craig Atkinson and Brendan McCane and Lech Szymanski and Anthony Robins}, journal={arXiv preprint arXiv:1911.11988}, url={https://arxiv.org/abs/1911.11988}, year={2019} }

@inproceedings{xu2018twin, title={Twin Bounded Large Margin Distribution Machine}, author={Xu, Haitao and McCane, Brendan and Szymanski, Lech}, booktitle={Australasian Joint Conference on Artificial Intelligence}, pages={718--729}, year={2018}, url={https://link.springer.com/chapter/10.1007/978-3-030-03991-2_64}, organization={Springer} }

@article{atkinson2018pseudo-recursal, title={Pseudo-recursal: Solving the catastrophic forgetting problem in deep neural networks}, author={Atkinson, Craig and McCane, Brendan and Szymanski, Lech and Robins, Anthony}, journal={arXiv preprint arXiv:1802.03875}, url={https://arxiv.org/abs/1802.03875}, year={2018} }

@techreport{Szymanski.etal2012b, author = {Lech Szymanski and Chris Gorman and Alistair Knott and Brendan McCane and Martin Takac}, title = {On Learning Object Properties in Convolutional Neural Networks via an Inhibition of Return (IOR) Mechanism}, number = {OUCS-2018-04}, institution = {Department of Computer Science, University of Otago}, year = {2018}, url = {https://www.otago.ac.nz/computer-science/otago702113.pdf} }

@article{Szymanski.etal2018a, author = {Lech Szymanski and Brendan McCane and Michael Albert}, title = {The effect of the choice of neural network depth and breadth on the size of its hypothesis space}, journal = {CoRR}, volume = {abs/1806.02460}, year = {2018}, url = {http://arxiv.org/abs/1806.02460}, archivePrefix = {arXiv}, eprint = {1806.02460}, timestamp = {Mon, 13 Aug 2018 16:47:32 +0200}, biburl = {https://dblp.org/rec/bib/journals/corr/abs-1806-02460}, bibsource = {dblp computer science bibliography, https://dblp.org} }

@INPROCEEDINGS{Clark-Younger.etal:2018, author={H. Clark-Younger and S. Mills and L. Szymanski}, booktitle={2018 International Conference on Image and Vision Computing New Zealand (IVCNZ)}, title={Stacked Hourglass CNN for Handwritten Character Location}, year={2018}, pages={1-6}, url={https://doi.org/10.1109/IVCNZ.2018.8634694} }

@article{mccane.etal2017a, author = {Brendan McCane and Lech Szymanski}, title = {Efficiency of deep networks for radially symmetric functions}, journal = {Neurocomputing}, volume = {313}, pages = {119--124}, year = {2017}, url = {https://doi.org/10.1016/j.neucom.2018.06.003} }

@inproceedings{atkinson2017increasing, title={Increasing the accuracy of convolutional neural networks with progressive reinitialisation}, author={Atkinson, Craig and McCane, Brendan and Szymanski, Lech}, booktitle={2017 International Conference on Image and Vision Computing New Zealand (IVCNZ)}, pages={1--5}, year={2017}, url={https://doi.org/10.1109/IVCNZ.2017.8402457}, organization={IEEE} }

@article{Szymanski.etal:2017b, author = {Lech Szymanski and Brendan McCane and Wei Gao and Zhi{-}Hua Zhou}, title = {Effects of the optimisation of the margin distribution on generalisation in deep architectures}, journal = {CoRR}, volume = {abs/1704.05646}, year = {2017}, url = {http://arxiv.org/abs/1704.05646}, archivePrefix = {arXiv}, eprint = {1704.05646}, timestamp = {Mon, 13 Aug 2018 16:47:28 +0200}, biburl = {https://dblp.org/rec/bib/journals/corr/SzymanskiMGZ17}, bibsource = {dblp computer science bibliography, https://dblp.org} }

@INPROCEEDINGS{Szymanski.etal:2017a, author={L. Szymanski and S. Mills}, booktitle={2017 International Conference on Image and Vision Computing New Zealand (IVCNZ)}, title={CNN for historic handwritten document search}, year={2017}, pages={1-6}, url={https://doi.org/10.1109/IVCNZ.2017.8402461} }

@inproceedings{McCane.etal:2016a, author={B. McCane and L. Szymanski}, booktitle={2016 23rd International Conference on Pattern Recognition (ICPR)}, title={Deep networks are efficient for circular manifolds}, year={2016}, pages={3464-3469}, url={https://doi.org/10.1109/ICPR.2016.7900170} }

@article{Szymanski.etal:2014, author={L. Szymanski and B. McCane}, journal={IEEE Transactions on Neural Networks and Learning Systems}, title={Deep Networks are Effective Encoders of Periodicity}, year={2014}, volume={25}, number={10}, pages={1816-1827}, doi={10.1109/TNNLS.2013.2296046}, url={https://doi.org/10.1109/TNNLS.2013.2296046} }

@INPROCEEDINGS{Szymanski.etal:2013a, author={L. Szymanski and B. McCane}, booktitle={The 2013 International Joint Conference on Neural Networks (IJCNN)}, title={Learning in deep architectures with folding transformations}, year={2013}, pages={1-8}, url={https://doi.org/10.1109/IJCNN.2013.6706945} }

@INPROCEEDINGS{Martin.etal:2013a, author={S. Martin and L. Szymanski}, booktitle={2013 28th International Conference on Image and Vision Computing New Zealand (IVCNZ 2013)}, title={Singularity resolution for dimension reduction}, year={2013}, pages={19-24}, url={https://doi.org/10.1109/IVCNZ.2013.6726986} }

@INPROCEEDINGS{Szymanski.etal:2012b, author={L. Szymanski and B. McCane}, booktitle={The 2012 International Joint Conference on Neural Networks (IJCNN)}, title={Push-pull separability objective for supervised layer-wise training of neural networks}, year={2012}, pages={1-8}, url = {https://doi.org/10.1109/IJCNN.2012.6252366} }

@phdthesis{Szymanski:2012a, Author = {Lech Szymanski}, School = {University of Otago}, Title = {Deep architectures and classification by intermediary transformations}, Url = {http://hdl.handle.net/10523/2129}, Year = {2012} }

@INPROCEEDINGS{Szymanski.etal:2012a, author={L. Szymanski and B. McCane}, booktitle={The 2012 International Joint Conference on Neural Networks (IJCNN)}, title={Deep, super-narrow neural network is a universal classifier}, year={2012}, pages={1-8}, url={https://doi.org/10.1109/IJCNN.2012.6252513} }

@inproceedings{Szymanski.etal:2011c, Author = {Lech Szymanski and Brendan McCane}, Booktitle = {Proceedings of Image and Vision Computing New Zealand}, Pages = {449-452}, Title = {Visualising Kernel Spaces}, Year = {2011} }