Source author record

Morten Brun

Morten Brun appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

math.AT Machine Learning math.AC math.CO

Catalog footprint

What is connected

4works

4topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

Equivariant Structure on Smash Powers

We provide foundations for dealing with the equivariant structure of "smash powers" of commutative orthogonal ring spectra. The category of commutative orthogonal ring spectra $A$ is tensored over spaces $X$, so that $A \otimes X$ is a commutative orthogonal ring spectrum. If $X$ is a discrete space, this is literally the smash power of $A$ with itself indexed over $X$, and we keep this language also in the nondiscrete case. In particular $A \otimes S^1$ is a model for topological Hochschild homology. We provide a framework where a generalization of the cyclotomic structure of topological Hochschild homology is visible in a categorical framework, also for more general $G$ and $X$. Similar situations have been studied by others, e.g., in Hill, Hopkins and Ravenel's treatment of the norm construction and Brun, Carlsson, Dundas' covering homology. In the case of non-commutative $A$ and $X=S^1$, the situation is somewhat easier and has already been covered by Kro. We are motivated by applications to $G$ being a torus in order to study the iterated algebraic $K$-theory, and have to develop a categorical theory that in some ways goes beyond what has been done before. Most of the material appeared in the last author's thesis which was defended in 2011. We apologize for the delay.

preprint2021arXiv

Epistemic Uncertainty Quantification in Deep Learning Classification by the Delta Method

The Delta method is a classical procedure for quantifying epistemic uncertainty in statistical models, but its direct application to deep neural networks is prevented by the large number of parameters $P$. We propose a low cost variant of the Delta method applicable to $L_2$-regularized deep neural networks based on the top $K$ eigenpairs of the Fisher information matrix. We address efficient computation of full-rank approximate eigendecompositions in terms of either the exact inverse Hessian, the inverse outer-products of gradients approximation or the so-called Sandwich estimator. Moreover, we provide a bound on the approximation error for the uncertainty of the predictive class probabilities. We observe that when the smallest eigenvalue of the Fisher information matrix is near the $L_2$-regularization rate, the approximation error is close to zero even when $K\ll P$. A demonstration of the methodology is presented using a TensorFlow implementation, and we show that meaningful rankings of images based on predictive uncertainty can be obtained for two LeNet-based neural networks using the MNIST and CIFAR-10 datasets. Further, we observe that false positives have on average a higher predictive epistemic uncertainty than true positives. This suggests that there is supplementing information in the uncertainty measure not captured by the classification alone.

preprint2010arXiv

The Auslander-Reiten translate on monomial quotient rings

For a multidegree t in N^n, E.Miller has defined a category of positively t-determined modules over the polynomial ring S in n variables. We consider the Auslander-Reiten translate, Na_t, on the (derived) category of such modules. A monomial ideal I is positively t-determined if every generator x^a has a \leq t. We compute the multigraded cohomology- and betti spaces of Na_t^k(S/I) for every iterate k, and also the S-module structure of these cohomology modules. This comprehensively generalizes results of Hochster and Gräbe on local cohomology of Stanley-Reisner rings.

preprint2001arXiv

Filtered Topological Cyclic Homology and relative K-theory of nilpotent ideals

In this paper we examine certain filtrations of topological Hochschild homology and topological cyclic homology. As an example we show how the filtration with respect to a nilpotent ideal gives rise to an analog of a theorem of Goodwillie saying that rationally relative K-theory and relative cyclic homology agree. Our variation says that the p-torsion parts agree in a range of degrees. We use it to compute K_i(Z/p^m) for i < p-2.