Complex & Intelligent Systems, Volume 12, Issue 3, March 2026

2026-04-02 · Source: Computational Intelligence · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, short

Summary

The March 2026 issue of "Complex & Intelligent Systems" presents 32 research articles covering diverse topics in AI and complex systems. Key contributions include a predictive neural network for dynamic rescheduling in robotic cells and an analysis of 51% attack vulnerability in nascent blockchains. Several papers focus on efficient large language model (LLM) inference, such as a comparative analysis of QLoRA fine-tuning for Gemma 3, Qwen 3, and Llama 3.2 models on resource-constrained devices, and multi-tier dynamic storage of KV cache for LLM inference. Other research explores explainable multimodal hand gesture recognition, brain tumor detection from 3D MRI using kernelized fuzzy C means with recurrent convolutional networks, and sustainable and explainable multi-modal spectral fusion for image dehazing. The issue also features work on evolutionary multitasking algorithms, multi-vehicle motion planning, and radar object detection using Aquila-optimized capsule networks.

Key takeaway

For AI engineers and researchers working on resource-constrained applications, consider the comparative analysis of QLoRA fine-tuning for Gemma 3, Qwen 3, and Llama 3.2 models to optimize performance. Your choice of model and fine-tuning strategy can significantly impact efficiency and deployment feasibility, especially for tasks like multilingual spam detection. Evaluate the multi-tier dynamic storage of KV cache for LLM inference to further enhance efficiency under limited resources.

Key insights

The issue highlights advancements in AI, optimization, and security across diverse applications and system complexities.

Principles

• Efficiency is critical for AI deployment on constrained hardware.
• Explainability enhances trust in complex AI systems.
• Hybrid approaches often yield robust solutions.

Method

Methods include predictive neural networks for dynamic rescheduling, QLoRA fine-tuning for LLMs, kernelized fuzzy C means with recurrent convolutional networks for image analysis, and evolutionary multitasking algorithms.

In practice

• Apply QLoRA for efficient LLM deployment on edge devices.
• Utilize deep residual learning for multimodal gesture recognition.
• Employ Aquila-optimized capsule networks for radar object detection.

Topics

• Robotic Systems
• Large Language Models
• Blockchain Security
• Neural Networks
• Optimization Algorithms

Best for: NLP Engineer, Computer Vision Engineer, AI Scientist, Robotics Engineer, Research Scientist

Related on AIssential

• Multilingual Fine-Tuning via Localized Gradient Conflict Resolution — Localized gradient conflict resolution via parameter buckets significantly improves multilingual LLM fine-tuning efficiency and performance.
• Highly Efficient and Effective LLMs with Multi-Boolean Architectures — Direct finetuning LLMs in the Boolean domain with multi-kernel parameters enhances efficiency and performance.
• I Took a 397MB Model and Turned It Into a Customer Service Chatbot That Actually Works — Tiny, fine-tuned models can deliver effective, custom AI solutions for specific business needs at minimal cost.
• Are Large Language Models Economically Viable for Industry Deployment? — Industrial LLM deployment requires economic and operational metrics beyond accuracy, revealing smaller models often offer superior efficiency.
• This AI Paper Introduces TinyLoRA, A 13-Parameter Fine-Tuning Method That Reaches 91.8 Percent GSM8K on Qwen2.5-7B — TinyLoRA enables high LLM performance with only 13 parameters by leveraging reinforcement learning.
• Gemma 4 QAT models: Optimizing model compression for mobile and laptop efficiency — Quantization-Aware Training (QAT) significantly reduces LLM memory footprint while preserving quality for on-device deployment.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computational Intelligence.