Simple Convolutional

Simple convolutional neural networks (CNNs) remain a focus of research, despite the rise of more complex architectures, due to their efficiency and surprising effectiveness in various tasks. Current research investigates their representational power, exploring limitations in generalization and logical reasoning, and seeking to improve their performance through architectural modifications like novel attention mechanisms. This ongoing work aims to better understand the inner workings of these fundamental models, leading to improvements in interpretability and performance across diverse applications such as image segmentation and protein function analysis. The simplicity of CNNs makes them ideal for studying fundamental aspects of deep learning, offering insights applicable to more sophisticated networks.

Papers

July 31, 2024

Non-convolutional Graph Neural Networks
Yuanqing Wang, Kyunghyun Cho
\Sigma}{\Delta}$ Low Pas RNN Graph Level Classification RNN Behavior Simple Convolutional

June 1, 2024

Empirical influence functions to understand the logic of fine-tuning
Jordan K. Matelsky, Lyle Ungar, Konrad P. Kording
Neural Network Fine Tuning Logic Reasoning Semantic Distance Simple Convolutional

March 12, 2023

From Compass and Ruler to Convolution and Nonlinearity: On the Surprising Difficulty of Understanding a Simple CNN Solving a Simple Geometric Estimation Task
Thomas Dagès, Michael Lindenbaum, Alfred M. Bruckstein
Neural Network Activation Function Direct Convolution Nonlinear Model Difficulty Level Geometric Estimation Simple Convolutional

September 18, 2022

SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation
Meng-Hao Guo, Cheng-Ze Lu, Qibin Hou, Zhengning Liu, Ming-Ming Cheng, Shi-Min Hu
Semantic Segmentation Convolutional Attention Simple Convolutional

May 24, 2022

Learning multi-scale functional representations of proteins from single-cell microscopy data
Anastasia Razdaibiedina, Alexander Brechalov
LeArning Abstract Molecular Representation Single Cell Autoencoder Based Autoencoder Model Non Structural Protein Multi Scale Representation Protein Representation Simple Convolutional

May 19, 2022

Representation Power of Graph Neural Networks: Improved Expressivity via Algebraic Analysis
Charilaos I. Kanatsoulis, Alejandro Ribeiro
Graph Neural Network Behavior Expressivity Style Discriminative Representation Graph Shift Operator Graph Classification Datasets Simple Convolutional Non Determinsitic Algebraic

January 24, 2022

Paired Image to Image Translation for Strikethrough Removal From Handwritten Words
Raphaela Heil, Ekta Vats, Anders Hast
Image Translation Handwritten Text 3D Stroke Simple Convolutional

Simple Convolutional

Papers

Non-convolutional Graph Neural Networks

Empirical influence functions to understand the logic of fine-tuning

From Compass and Ruler to Convolution and Nonlinearity: On the Surprising Difficulty of Understanding a Simple CNN Solving a Simple Geometric Estimation Task

SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation

Learning multi-scale functional representations of proteins from single-cell microscopy data

Representation Power of Graph Neural Networks: Improved Expressivity via Algebraic Analysis

Paired Image to Image Translation for Strikethrough Removal From Handwritten Words