On the explainability of max-plus neural networks

2026-05-05 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

A new study investigates the explainability of linear-min-max neural networks, a recently proposed model. These networks can be initially interpreted as k-medoids using the infinity norm for distance and are subsequently trained via subgradient descent. Despite being universal approximators, their decision process is traceable because a single most activated neuron dictates the output value. Leveraging this property, the researchers developed a "pixel fragility measure" to identify if changes to a single pixel can alter the classification output. Experiments conducted on the PneumoniaMnist dataset demonstrate that this explanation method for the neural network's output performs favorably compared to established techniques like SHAP and Integrated Gradient. This work will be presented at the IEEE International Symposium on Computer-Based Medical Systems (CBMS 2026) in June 2026.

Key takeaway

For Computer Vision Engineers developing explainable AI models, consider linear-min-max neural networks for their inherent traceability. This architecture allows you to pinpoint specific neuron activations for output decisions, potentially offering more direct and robust explanations than traditional methods like SHAP or Integrated Gradient, especially in critical applications like medical imaging diagnostics.

Key insights

Linear-min-max neural networks offer inherent explainability by design, outperforming SHAP and Integrated Gradient.

Principles

• Single neuron activation drives output
• Universal approximation with traceability

Method

The method involves interpreting network initialization as k-medoids, training with subgradient descent, and using a pixel fragility measure based on single neuron activation.

In practice

• Evaluate pixel fragility for classification
• Apply to medical image datasets

Topics

• Max-plus Neural Networks
• Model Explainability
• Pixel Fragility Measure
• SHAP
• Integrated Gradient

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Explainable AI in Production: A Neuro-Symbolic Model for Real-Time Fraud Detection — Integrating explanations directly into model architecture enables real-time, deterministic, and low-latency explainability.
• Applied Explainability for Large Language Models: A Comparative Study — Gradient-based explainability provides stable LLM insights, while attention is efficient but less aligned with predictions.
• OPTIMUS-Prime: Minimal and Sufficient Concept Explanations for Deep Vision Models — OPTIMUS-Prime provides minimal and sufficient concept-based visual explanations for deep vision models, grounded in prime implicant theory.
• Nobody Knows What’s Inside a Neural Network. A Few People Are Trying to Find Out. — A small group of researchers is undertaking "biology on language models," dissecting them like organisms to uncover their internal mechanisms, a process revealing unexpected findings.
• Language Model Circuits Are Sparse in the Neuron Basis — MLP activations, combined with RelP attribution, offer a sparse and faithful basis for language model circuit tracing, comparable to SAEs.
• Interpretable and Explainable Surrogate Modeling for Simulations: A State-of-the-Art Survey and Perspectives on Explainable AI for Decision-Making — Integrating XAI with surrogate models enhances transparency and diagnostic capabilities for complex simulations, bridging efficiency and interpretability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.