Summary of recent scientific articles

Introduction

The 2025–2026 period has seen a significant acceleration in the adoption of AI solutions for industry. Three themes dominate recent research: Industrial IoT (IIoT), Machine Learning (ML), and AI Privacy. This article synthesizes the most recent publications and connects them to the work of Giacomo Veneri, a pioneer in these fields.

1. Industrial IoT: The New Frontier of Security

Latest Research (2025–2026)

A fundamental paper published in Scientific Reports (Nature) presents a zero-trust digital twin framework for IIoT that combines:

  • Anomaly detection using deep learning
  • Differential privacy (ε = 25)
  • Lightweight blockchain logging (SHA-256)
  • Accuracy: 89–91% with MLP and CNN-BiLSTM
  • Inference latency: ~1 second for 500 samples

REF: https://www.nature.com/articles/s41598-026-42041-w

DVACNN-Fed introduces a federated learning approach for NIDS (Network Intrusion Detection System) that:

2. Industrial Machine Learning: From Theory to Production

Latest Research (2025–2026)

RC-NF (Robot-Conditioned Normalizing Flow) is a method for real-time anomaly detection in robotic manipulation:

  • Improves the robustness of Vision-Language-Action models in OOD (Out-of-Distribution) environments
  • Trained only with positive samples (unsupervised)
  • Latency < 100 ms for OOD signals
  • Benchmark: LIBERO-Anomaly-10

REF: https://arxiv.org/abs/2603.11106

Soft Thinking introduces a continuous conceptual space of probability-weighted token embedding mixtures:

  • Improves pass@1 accuracy by up to 2.48 points
  • Reduces token usage by up to 22.4% compared to Chain-of-Thought
  • Enhances the interpretability of outputs

REF: https://arxiv.org/abs/2505.15778

See also

Our book “Hands-On Industrial Internet of Things” (Packt Publishing, 2018, 2nd edition 2024), with 63 citations. The book is a reference for building IIoT infrastructure using Industry 4.0 standards.

REF: https://amzn.asia/d/09fUEj6J