amen zwa, esq.4d ago

Doug @dougmerritt, we were discussing the status of the #Kohonen #SOM research, last week. Here are a few recent (21st Century) publications on the topic that I like, either for their undergrad accessible styles or for their advanced research ideas.

Given that this is my favourite list, it skews heavily toward DSP and DIP. But then, Kohonen did design the SOM expressly for perceptual processing of auditory and visual signals.

The idea of implementing quantised SOMs on FPGAs intrigues me, at present.

• 2001 Kohonen—SOMs 3ed
• 2001 Kiang—Extending the Kohonen SOM for Cluster Analysis
• 2001 Villmann—Exts and Mods of SOM and Apps in Remote Sensing Image Analysis
• 2002 Seiffert—SOMs: Recent Advances and Apps
• 2003 Zherebtsov—Clustering Stock Portfolios
• 2004 Bação—Intro to SOM
• 2004 Mokriš—Decreasing the Feature Space Dim by SOMs
• 2005 Guthikonda—SOM
• 2005 Huang—Exploration of Dim Reduction for Text Visualisation
• 2007 Sharma—Image Comp and Feature Extr with NN
• 2007 Villmann—Class Imaging of Hyperspectral Satellite Remote Sensing Data Using FLSOM
• 2008 Sap—Overlapping Clusters
• 2008 Skupin— Intro: SOM
• 2008 Yin—SOMs: Background, Theories, Exts, and Apps
• 2009 Campoy—Dim Reduction by SOMs that Preserve Distances in Output Space
• 2010 Dvorský—Improvements Quality of SOMs Using Dim Reduction Methods
• 2012 Kohonen—Essentials of SOM
• 2012 Asan—An Intro to SOMs
• 2014 Kohonen—MATLAB Impl and Apps of SOM
• 2015 Abdelsamea—Image Feature Classification
• 2024 Linke—SOMson: Sonification of Multi-Dim Data in SOMs
• 2025 Malik—SOMs
• 2025 Nogales—SOMs as a Way to Evaluate Optimal Strategies for Balancing Binary Class Distribution

old school #AI

Show threadDougMerritt (log😅 = 💧log😄)

@AmenZwa
> Dim Reduction by SOMs that Preserve Distances in Output Space

That's a pretty important ability for some things.

Apr 11 at 1:08amWeb
Show threadamen zwa, esq.4d ago

@dougmerritt
Kohonen invented the SOM precisely for that purpose: unsupervised learning of a relations-preserving nonlinear mapping from a high-dimensional input vector space down to a 2D tonotopic map. He was inspired by how the primate auditory and visual systems perform such mappings, effortlessly.

The SOM is one of those Nature-inspired, simple, elegant, efficient, practical solutions—the engineering Holy Grail. You can see why I have been fascinated with this old school AI, for more than four decades.