Searcharxiv⌕ Search

arXiv subjects

Gwyneth Liu

Publications and source records attributed to Gwyneth Liu.

2 recordsLinked to original sources
arXiv2025

Community Recovery on Noisy Stochastic Block Models

We study the problem of community recovery in geometrically-noised stochastic block models (SBM). This work presents two primary contributions: (1) Motif--Attention Spectral Operator (MASO), an attention-based spectral operator that improves upon traditional spectral methods; and (2) Iterative Geometric Denoising (GeoDe), a configurable denoising algorithm that boosts spectral clustering performance. We demonstrate that the fusion of GeoDe+MASO significantly outperforms existing community detection methods on noisy SBMs. Furthermore, we show that using GeoDe+MASO as a denoising step improves belief propagation's community recovery by 79.7% on the Amazon Metadata dataset.

cs.SI↗
arXiv2025

NeuroADDA: Active Discriminative Domain Adaptation in Connectomic

Training segmentation models from scratch has been the standard approach for new electron microscopy connectomics datasets. However, leveraging pretrained models from existing datasets could improve efficiency and performance in constrained annotation budget. In this study, we investigate domain adaptation in connectomics by analyzing six major datasets spanning different organisms. We show that, Maximum Mean Discrepancy (MMD) between neuron image distributions serves as a reliable indicator of transferability, and identifies the optimal source domain for transfer learning. Building on this, we introduce NeuroADDA, a method that combines optimal domain selection with source-free active learning to effectively adapt pretrained backbones to a new dataset. NeuroADDA consistently outperforms training from scratch across diverse datasets and fine-tuning sample sizes, with the largest gain observed at $n=4$ samples with a 25-67\% reduction in Variation of Information. Finally, we show that our analysis of distributional differences among neuron images from multiple species in a learned feature space reveals that these domain "distances" correlate with phylogenetic distance among those species.

cs.CV↗