31,825 research outputs found
The Effect of Manifold Entanglement and Intrinsic Dimensionality on Learning
No full textWe empirically investigate the effect of class manifold entanglement and the intrinsic and extrinsic dimensionality of the data distribution on the sample complexity of supervised classification with deep ReLU networks. We separate the effect of entanglement and intrinsic dimensionality and show statistically for artificial and real-world image datasets that the intrinsic dimensionality and the entanglement have an interdependent effect on the sample complexity. Low levels of entanglement lead to low increases of the sample complexity when the intrinsic dimensionality is increased, while for high levels of entanglement the impact of the intrinsic dimensionality increases as well. Further, we show that in general the sample complexity is primarily due to the entanglement and only secondarily due to the intrinsic dimensionality of the data distribution
Michael Rodriguez interviews fiction writer Michael Kimball
No full textAuthor Michael Kimball talks about moving away from Michigan to become a successful writer, his education, the fiction reading series he has started in Baltimore, the life-story-on-postcard project, and his book "Dear everybody." Kimball is interviewed by Michigan State University Librarian Michael Rodriguez for the Michigan State University Libraries' Michigan Writers Series
Comparing Complexities of Decision Boundaries for Robust Training: A Universal Approach
No full textWe investigate the geometric complexity of decision boundaries for robust training compared to standard training. By considering the local geometry of nearest neighbour sets, we study them in a model-agnostic way and theoretically derive a lower-bound R∗ ∈ R on the perturbation magnitude δ ∈ R for which robust training provably requires a geometrically more complex decision boundary than accurate training. We show that state-of-the-art robust models learn more complex decision boundaries than their non-robust counterparts, confirming previous hypotheses. Then, we compute R∗ for common image benchmarks and find that it also empirically serves as an upper bound over which label noise is introduced. We demonstrate for deep neural network classifiers that perturbation magnitudes δ ≥ R∗ lead to reduced robustness and generalization performance. Therefore, R∗ bounds the maximum feasible perturbation magnitude for norm-bounded robust training and data augmentation. Finally, we show that R∗ < 0.5R for common benchmarks, where R is a distribution’s minimum nearest neighbour distance. Thus, we improve previous work on determining a distribution’s maximum robust radius
Relative Robustness of Quantized Neural Networks Against Adversarial Attacks
No full textNeural networks are increasingly being moved to edge computing devices and smart sensors, to reduce latency and save bandwidth. Neural network compression such as quantization is necessary to fit trained neural networks into these resource constrained devices. At the same time, their use in safety-critical applications raises the need to verify properties of neural networks. Adversarial perturbations have potential to be used as an attack mechanism on neural networks, leading to "obviously wrong" misclassification. SMT solvers have been proposed to formally prove robustness guarantees against such adversarial perturbations. We investigate how well these robustness guarantees are preserved when the precision of a neural network is quantized. We also evaluate how effectively adversarial attacks transfer to quantized neural networks. Our results show that quantized neural networks are generally robust relative to their full precision counterpart (98.6%-99.7%), and the transfer of adversarial attacks decreases to as low as 52.05% when the subtlety of perturbation increases. These results show that quantization introduces resilience against transfer of adversarial attacks whilst causing negligible loss of robustness
Michael Rodriguez interviews author Paul Clemens
No full textAuthor Paul Clemens talks about his book "Made in Detroit," the genre of memoir, and writing about race. Clemens is interviewed by Michigan State University Librarian Michael Rodriguez for the MSU Libraries' Michigan Writers Series. Held in the MSU Main Library
Evolving Ensembles:What Can We Learn from Biological Mutualisms?
No full textEnsembles are groups of classifiers which cooperate in order to reach a decision. Conventionally, the members of an ensemble are trained sequentially, and typically independently, and are not brought together until the final stages of ensemble generation. In this paper, we discuss the potential benefits of training classifiers together, so that they learn to interact at an early stage of their development. As a potential mechanism for achieving this, we consider the biological concept of mutualism, whereby cooperation emerges over the course of biological evolution. We also discuss potential mechanisms for implementing this approach within an evolutionary algorithm context
Michael Rodriguez interviews author Tom Springer
No full textAuthor Tom Springer is interviewed about his writing career and his newest book "Looking for hickories". Springer talks about his career following after earning an Environmental Journalism degree from Michigan State University. He calls his genre "creative non-fiction" and explains how he weaves his memories into his books about life in rural and wild Michigan. Part of the Michigan State University Libraries' Michigan Writers Series. Springer is interviewed by Librarian Michael Rodriguez
Michael Rodriguez interviews author Gary Gildner
No full textAuthor Gary Gildner explains why he left his tenured teaching position to move to Idaho to became a full-time writer of poetry. Gildner talks about donating his personal papers to Michigan State University Libraries' Special Collections, his writing style and how he approaches writing. Gildner is interviewed by MSU Librarian Michael Rodriguez for the MSU Libraries' Michigan Writer Series. Held at the MSU Main Library
Gold standard of UK degrees is lost in translation
No full textInflated marks, overworked staff and politically compromised courses are the price of exploiting offshore UK registered students, says Michael Day
Michael Rodriguez interviews historian and author Keith Widder
No full textHistorian and author Keith Widder talks about his move to Michigan from Wisconsin, his career as Curator of History for the Mackinac Island State Park Commission, his research interests, his book "Michigan Agricultural College", and his current projects. Widder is interviewed by Michigan State University Librarian Michael Rodriguez for the MSU Libraries' Michigan Writers Series. Held in the MSU Main Library
- …
