Blog

Research notes, experiments, and technical explorations.

Series

Graph Signal Processing 2 parts · Understanding spectral methods on graphs

1. Spectral Analysis of Cora
2. Chebyshev Spectral Convolution

Graph Neural Networks 2 parts · Core concepts and analysis of GNNs

1. Revisiting Classical SSL Methods
2. Efficiency Comparison

All Posts

• Discovering Mechanisms in Tokenized Graph Transformers (Interactive Demo)

  Presented in ICML worhkshop on Mechanistic Interpretability 2026.

  July 2, 2026 · 8 min read

• On the feasibility of fidelity- for graph pruning

  Exploring whether GNN explanation methods can be repurposed for graph pruning

  August 3, 2024 · 10 min read

• Experiments on using Kolmogorov-Arnold Networks (KAN) on Graph Learning (Github link)

  Exploring the application of KAN architecture to graph neural networks

  March 29, 2024 · 1 min read

• Comparing the Efficiency of ChebConv, GCN, and SGC

  A comprehensive efficiency comparison of three popular spectral graph neural network models

  November 11, 2022 · 12 min read

• Understanding ChebConv with the NEIGHBORSMATCH problem

  Deep dive into Chebyshev spectral graph convolutions and analysis using the NEIGHBORSMATCH problem

  October 30, 2022 · 31 min read

• Dissecting Cora in the Spectral Domain

  Graph signal processing and Fourier analysis on the Cora dataset

  October 19, 2022 · 17 min read

• Revisiting Modern Benchmarks for Semi-suparvised Node Classification using Classical Methods

  Applying classical label propagation methods to modern graph benchmark datasets

  October 8, 2022 · 14 min read