Searcharxiv⌕ Search

arXiv subjects

Yaoying Fu

Publications and source records attributed to Yaoying Fu.

4 recordsLinked to original sources

Canonical activation patterns and ReLU pattern varieties

A ReLU pattern variety records the polynomial relations among the coefficients of the affine maps associated with a prescribed set of activation patterns of a ReLU network. We study how the combinatorics of the canonical set of activation patterns of an unbiased ReLU network is reflected in the geometry of its pattern variety. In particular, we introduce the notion of pattern stability and characterize pattern stability and combinatorial stability for shallow networks. When the associated wall graph is connected and contains an edge of every neuron label, we resolve a problem of Alexandr and Montúfar by determining the defining ideal of the associated pattern variety explicitly. We establish a hierarchy of progressively weaker equivalence relations relating labeled combinatorial equivalence, equality and permutation equivalence of canonical pattern sets, and equivalences of the associated varieties. Finally, we relate functional dimension to the differential of the pattern map and show that it is bounded above by the dimension of the canonical pattern variety at every depth. For nondegenerate shallow networks, we give an exact formula for the functional dimension; for every shallow architecture, it equals the dimension of the canonical pattern variety on an open dense set of full Lebesgue measure.

math.CO↗

Toric geometry of ReLU neural networks

Given a continuous piecewise linear function $f:\mathbb{R}^{n_0} \to \mathbb{R}$ and a fixed architecture of feedforward ReLU neural networks, the exact function realization problem asks when a network of that architecture realizes $f$. In contrast, expressivity concerns characterizing all functions exactly realizable by a given architecture. To address these questions systematically, we establish a connection between toric geometry and ReLU neural networks, allowing tools from algebraic geometry to analyze the structures. Starting with an unbiased ReLU network with rational weights, we define the ReLU fan, the ReLU toric variety, and the ReLU Cartier divisor associated with the network. This work also relates the tropical and the toric geometry: the former captures combinatorial data of ReLU networks, while the latter recovers the underlying geometric structure. As an application, we obtain a complete characterization of the functions realizable by unbiased shallow ReLU networks with rational weights. The characterization is expressed in terms of intersection numbers of ReLU Cartier divisors with torus-invariant curves and leads to a decomposition of piecewise-linear functions into shallow and non-shallow components.

math.AG↗

Persistent Homology for Labeled Datasets: Gromov-Hausdorff Stability and Generalized Landscapes

Techniques from metric geometry have become fundamental tools in modern mathematical data science, providing principled methods for comparing datasets modeled as finite metric spaces. Two of the central tools in this area are the Gromov-Hausdorff distance and persistent homology, both of which yield isometry-invariant notions of distance between datasets. However, these frameworks do not account for categorical labels, which are intrinsic to many real-world datasets, such as labeled images, pre-clustered data, and semantically segmented shapes. In this paper, we introduce a general framework for labeled metric spaces and develop new notions of Gromov-Hausdorff distance and persistent homology which are adapted to this setting. Our main result shows that our persistent homology construction is stable with respect to our novel notion of Gromov-Hausdorff distance, extending a classic result in topological data analysis. To facilitate computation, we also introduce a labeled version of persistence landscapes and show that the landscape map is Lipschitz.

math.AT↗

Remarks on the distribution of colors in Gallai colorings

A Gallai coloring of a complete graph $K_n$ is an edge coloring without triangles colored with three different colors. A sequence $e_1\ge \dots \ge e_k$ of positive integers is an $(n,k)$-sequence if $\sum_{i=1}^k e_i=\binom{n}{2}$. An $(n,k)$-sequence is a G-sequence if there is a Gallai coloring of $K_n$ with $k$ colors such that there are $e_i$ edges of color $i$ for all $i,1\le i \le k$. Gyárfás, Pálvölgyi, Patkós and Wales proved that for any integer $k\ge 3$ there exists an integer $g(k)$ such that every $(n,k)$-sequence is a G-sequence if and only if $n\ge g(k)$. They showed that $g(3)=5, g(4)=8$ and $2k-2\le g(k)\le 8k^2+1$. We show that $g(5)=10$ and give almost matching lower and upper bounds for $g(k)$ by showing that with suitable constants $α,β>0$, $\frac{αk^{1.5}}{\ln k}\le g(k) \le βk^{1.5}$ for all sufficiently large $k$.

math.CO↗