arXiv subjects

Maia Fraser

Publications and source records attributed to Maia Fraser.

8 recordsLinked to original sources

arXiv2026

Manifold-Matching Autoencoders

We study a simple unsupervised regularization scheme for autoencoders called Manifold-Matching (MMAE): we align the pairwise distances in the latent space to those of the input data space by minimizing mean squared error. Because alignment occurs on pairwise distances rather than coordinates, it can also be extended to a lower-dimensional representation of the data, adding flexibility to the method. We find that this regularization outperforms similar methods on metrics based on preservation of nearest-neighbor distances and persistent homology-based measures. We also observe that MMAE provides a scalable approximation of Multi-Dimensional Scaling (MDS).

cs.LG↗

arXiv2023

Contact non-squeezing at large scale via generating functions

Using SFT techniques, Eliashberg, Kim and Polterovich (2006) proved that if $\pi R_2^2 \leq K \leq \pi R_1^2$ for some integer $K$ then there is no contact squeezing in $\mathbb{R}^{2n} \times S^1$ of the prequantization of the ball of radius $R_1$ into the prequantization of the ball of radius $R_2$. This result was extended to the case of balls of radius $R_1$ and $R_2$ with $1 \leq \pi R_2^2 \leq \pi R_1^2$ by Chiu (2017) and the first author (2016), using respectively microlocal sheaves and SFT. In the present article we recover this general contact non-squeezing theorem using generating functions, a classical method based on finite dimensional Morse theory. More precisely, we develop an equivariant version, with respect to a certain action of a finite cyclic group, of the generating function homology for domains of $\mathbb{R}^{2n} \times S^1$ defined by the second author (2011). A key role in the construction is played by translated chains of contactomorphisms, a generalization of translated points.

math.SG↗

arXiv2022

Dynamic programming with incomplete information to overcome navigational uncertainty in a nautical environment

Using a novel toy nautical navigation environment, we show that dynamic programming can be used when only incomplete information about a partially observed Markov decision process (POMDP) is known. By incorporating uncertainty into our model, we show that navigation policies can be constructed that maintain safety, outperforming the baseline performance of traditional dynamic programming for Markov decision processes (MDPs). Adding in controlled sensing methods, we show that these policies can also lower measurement costs at the same time.

math.OC↗

arXiv2016

On Sandon-type metrics for contactomorphism groups

For certain contact manifolds admitting a 1-periodic Reeb flow we construct a conjugation-invariant norm on the universal cover of the contactomorphism group. With respect to this norm the group admits a quasi-isometric monomorphism of the real line. The construction involves the partial order on contactomorphisms and symplectic intersections. This norm descends to a conjugation-invariant norm on the contactomorphism group. As a counterpoint, we discuss conditions under which conjugation-invariant norms for contactomorphisms are necessarily bounded.

math.SG↗

arXiv2016

Contact non-squeezing at large scale in ${\mathbb R}^{2n} \times S^1$

We define a $\mathbb{Z}_k$-equivariant version of the cylindrical contact homology used by Eliashberg-Kim-Polterovich (2006) to prove contact non-squeezing for prequantized integer-capacity balls $B(R) \times S^1 \subset \mathbb{R}^{2n} \times S^1$, $R \in \mathbb{N}$ and we use it to extend their result to all $R \geq 1$. Specifically we prove if $R \geq 1$ there is no $ψ\in \text{Cont}(\mathbb{R}^{2n} \times S^1)$, the group of compactly supported contactomorphisms of $\mathbb{R}^{2n} \times S^1$ which squeezes $\hat{B}(R) = B(R) \times S^1$ into itself, i.e. maps the closure of $\hat{B}(R)$ into $\hat{B}(R)$. A sheaf theoretic proof of non-existence of corresponding $ψ\in \text{Cont}_0(\mathbb{R}^{2n} \times S^1)$, the identity component of $\text{Cont}(\mathbb{R}^{2n} \times S^1)$, is due to Chiu (2014); it is not known if this is strictly weaker. Our construction has the advantage of retaining the contact homological viewpoint of Eliashberg-Kim-Polterovich and its potential for application in prequantizations of other Liouville manifolds. It makes use of the $\mathbb{Z}_k$-action generated by a vertical $1/k$-shift but can also be related, for prequantized balls, to the $\mathbb{Z}_k$-equivariant contact homology developed by Milin (2008) in her proof of orderability of lens spaces.

math.SG↗

arXiv2015

Contact spectral invariants and persistence

This sketch shows that the usual generating function based capacities have an interpretation in the language of persistent homology as persistences of certain homology classes in the persistence module formed by the corresponding generating function homology groups. This viewpoint suggests various new invariants, in particular a $\mathbb{Z}_k$-equivariant capacity which can be used to prove orderability of lens spaces, proved by Milin (2008) using contact homology and by Sandon (2010) using equivariant generating function homology. These are informal notes originally circulated in January 2014.

math.SG↗

arXiv2012

Local Routing in Graphs Embedded on Surfaces of Arbitrary Genus

We present a local routing algorithm which guarantees delivery in all connected graphs embedded on a known surface of genus $g$. The algorithm transports $O(g\log n)$ memory and finishes in time $O(g^2n^2)$, where $n$ is the size of the graph. It requires access to a homology basis for the surface. This algorithm, GFR, may be viewed as a suitable generalization of Face Routing (FR), the well-known algorithm for plane graphs, which we previously showed does {\it not} guarantee delivery in graphs embedded on positive genus surfaces. The problem for such surfaces is the potential presence of homologically non-trivial closed walks which may be traversed by the right-hand rule. We use an interesting mathematical property of homology bases (proven in Lemma \ref{lem:connectFaceBdr}) to show that such walks will not impede GFR. FR is at the base of most routing algorithms used in modern (2D) ad hoc networks: these algorithms all involve additional local techniques to deal with edge-crossings so FR may be applied. GFR should be viewed in the same light, as a base algorithm which could for example be tailored to sensor networks on surfaces in 3D. Currently there are no known efficient local, logarithmic memory algorithms for 3D ad hoc networks. From a theoretical point of view our work suggests that the efficiency advantages from which FR benefits are related to the codimension one nature of an embedded graph in a surface rather than the flatness of that surface (planarity).

cs.CG↗

arXiv2012

Persistent Homology of Filtered Covers

We prove an extension to the simplicial Nerve Lemma which establishes isomorphism of persistent homology groups, in the case where the covering spaces are filtered. While persistent homology is now widely used in topological data analysis, the usual Nerve Lemma does not provide isomorphism of persistent homology groups. Our argument involves some homological algebra: the key point being that although the maps produced in the standard proof of the Nerve Lemma do not commute as maps of chain complexes, the maps they induce on homology do.

math.AT↗