Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Quantization
Technique for reducing the precision of neural network weights and activations from 32 bits to 8 bits or less, significantly reducing memory and computational requirements.
Microcontroller
Compact integrated circuit containing processor, memory, and peripherals, optimized to operate with minimal power consumption and limited resources.
Edge AI
Artificial intelligence executed directly on edge devices, enabling local decisions without cloud connection dependency, reducing latency and bandwidth consumption.
TensorFlow Lite Micro
Google's framework specifically designed to run machine learning models on microcontrollers with less than 256KB of RAM and 1MB of storage.
Memory optimization
Set of techniques aimed at minimizing the memory footprint of models, including quantization, pruning, and compact network architectures adapted to MCU constraints.
On-device inference
Process of executing predictions directly on the embedded device, eliminating the need to transmit data to remote servers for processing.
Pruning
Technique of trimming non-critical neural connections in a network, reducing its complexity and size without significant loss of predictive performance.
Knowledge Distillation
Method of transferring knowledge from a large complex model (teacher) to a lightweight model (student) adapted to microcontroller resource constraints.
Neuromorphic Architecture
Design of circuits mimicking the structure and functioning of the biological brain, optimized for efficient processing with minimal energy consumption.
Model Compressor
Tool or algorithm that reduces the size of a machine learning model while preserving its predictive capabilities, essential for deployment on constrained devices.
Edge Impulse
Development platform specialized in creating, training, and deploying TinyML models on microcontrollers with an intuitive interface and automatic optimization.
Energy Consumption
Critical measurement in TinyML, aiming to minimize electrical consumption to enable years of autonomy on batteries or energy harvesting in IoT applications.
MCU
Compact microcontroller unit integrating processor, volatile and non-volatile memory, and communication interfaces in a single integrated circuit for embedded applications.
Real-time Processing
Capability of TinyML systems to provide predictive responses within constrained and predictable timeframes, essential for critical and interactive applications.
Embedded Intelligence
Integration of learning and inference capabilities directly into constrained electronic devices, creating autonomous and intelligent systems.
Lightweight Model
Neural network architecture specifically designed to minimize parameters and calculations while maintaining acceptable performance for deployment on microcontrollers.
Energy Harvesting
Technique for collecting environmental energy (light, vibration, thermal) to power TinyML devices, enabling near-unlimited autonomy without maintenance.
Embedded operating system
Real-time operating system optimized for microcontrollers, managing limited hardware resources and ensuring deterministic execution of TinyML tasks.