arXiv subjects

Alvin Jin

Publications and source records attributed to Alvin Jin.

6 recordsLinked to original sources

arXiv2026

Topological Simplification in Predictive Coding Networks

We study the topology of learned representations in predictive coding networks (PCNs), a neuro-inspired bidirectional architecture, using a quantitative layer-wise persistent homology analysis. We train well-performing PCNs on a synthetic classification dataset ($\geq 99.9\%$ test accuracy) and on MNIST ($\geq 95\%$ test accuracy), and measure how topological features change across layers for different architectures and activation functions. We find that smaller PCNs collapse connected components across layers earlier than larger models (Spearman $\unicode{x1D70C} \in [0.72, 0.79]$ across activations), with model size measured as the sum of hidden-layer widths. We also observe a strong negative correlation ($\unicode{x1D70C} = -0.58$) between the depth at which simplification occurs and reconstruction error; i.e., architectures that simplify later reconstruct better. Finally, a seed-level bootstrap comparison across architectures and activations shows that PCNs consistently collapse connected components later than matched MLPs, with an average difference of $3.6$ layers. These results suggest that persistent homology offers a useful quantitative lens on the compression--reconstruction tradeoff in PCNs, and that both model capacity and the recurrent, bidirectional dynamics of predictive coding inference shape when this tradeoff is resolved across layers.

cs.LG↗

arXiv2026

Hierarchical Synthetic Tabular Data Generation: A Hybrid Top-Down and Bottom-Up Framework

Existing approaches for synthetic tabular data generation are based on either purely generative models or LLMs, both of which struggle with data heterogeneity, logical consistency, rare-event coverage, and robustness in low-data regimes. In this paper, we propose a hierarchical hybrid top-down and bottom-up (H-TDBU) framework that decouples semantic structures from stochastic texture. In the top-down path, structure-driven logical constraints and cross-modal alignment rules are constructed, while in the bottom-up path, lightweight tabular generators are used to learn local statistical patterns from real data. The two paths are consolidated in a unified synthesis engine with an iterative feedback loop. We evaluate the framework on weak multimodal financial benchmarks combining tabular and sentiment-text data. Experimental results show that our H-TDBU approach improves train-synthetic-test-real performance over neural baseline methods while preserving semantic consistency. Our results suggest that hierarchical rule-guided synthesis provides an effective mechanism for combining controllability, semantic coherence, and statistical fidelity in synthetic data generation.

cs.LG↗

arXiv2022

Decomposing filtered chain complexes: geometry behind barcoding algorithms

In Topological Data Analysis, filtered chain complexes enter the persistence pipeline between the initial filtering of data and the final persistence invariants extraction. It is known that they admit a tame class of indecomposables, called interval spheres. In this paper, we provide an algorithm to decompose filtered chain complexes into such interval spheres. This algorithm provides geometric insights into various aspects of the standard persistence algorithm and two of its run-time optimizations. Moreover, since it works for any filtered chain complexes, our algorithm can be applied in more general cases. As an application, we show how to decompose filtered kernels with it.

math.AT↗

arXiv2021

Realisations of posets and tameness

We introduce a construction called realisation which transforms posets into posets. We show that realisations share several key features with upper semilattices. For example, we define local dimensions of points in a poset and show that these numbers for realisations behave in a similar way as they do for upper semilattices. Furthermore, similarly to upper semilattices, realisations have well-behaved discrete approximations which are suitable for capturing homological properties of functors indexed by them. These discretisations are convenient and effective for describing tameness of functors. Homotopical and homological properties of tame functors, particularly those indexed by realisations, are discussed, with emphasis on the use of Koszul complexes to compute Betti diagrams of minimal free resolutions of tame functors indexed by upper semilattices and realisations.

math.AT↗

arXiv2020

Homotopical decompositions of simplicial and Vietoris Rips complexes

Motivated by applications in Topological Data Analysis, we consider decompositions of a simplicial complex induced by a cover of its vertices. We study how the homotopy type of such decompositions approximates the homotopy of the simplicial complex itself. The difference between the simplicial complex and such an approximation is quantitatively measured by means of the so called obstruction complexes. Our general machinery is then specialized to clique complexes, Vietoris-Rips complexes and Vietoris-Rips complexes of metric gluings. For the latter we give metric conditions which allow to recover the first and zero-th homology of the gluing from the respective homologies of the components.

math.AT↗

arXiv2018

The rigid-flexible value for symplectic embeddings of four-dimensional ellipsoids into polydiscs

We consider the embedding function $c_b(a)$ describing the problem of symplectically embedding an ellipsoid $E(1,a)$ into the smallest scaling of the polydisc $P(1,b)$. Previous work suggests that determining the entirety of $c_b(a)$ for all $b$ is difficult, as infinite staircases can appear for many sequences of irrational $b$. In contrast, we show that for every polydisc $P(1,b)$ with $b>2$, there is an explicit formula for the minimum $a$ such that the embedding problem is determined only by volume. That is, when the ellipsoid is sufficiently stretched, there is a symplectic embedding of $E(1,a)$ fully filling an appropriately scaled polydisc $P(\lambda,\lambda b)$. Denoted $RF(b)$, this rigid-flexible ($RF$) value is piecewise smooth with a discrete set of discontinuities for $b>2$. At the same time, by exhibiting a sequence of obstructive classes for $b_n = \frac{n+1}{n}$ at $a=8$, we show % that $c_{b_n}(8)$ is above the volume constraint. So, in combination with the Frenkel-M\"{u}ller result, it follows that $RF$ is also discontinuous at $b=1$.

math.SG↗