AI-ordlista
Den kompletta ordlistan över AI
Split Learning
Collaborative learning approach where the neural network model is split between client and server, enabling learning without sharing raw data.
SplitNN
Split neural network architecture where initial layers run on the client and deeper layers on the server.
Cut Layer
Division point of the neural network model between client and server, determining which part of the computation is performed locally.
Client-side Model
Portion of the neural network model executed on the client device, processing data without transmitting it.
Server-side Model
Part of the neural network model hosted on the server, completing inference from received partial activations.
Vertical Split Learning
Variant where the model is split vertically between different layers, distributing computation between client and server.
Horizontal Split Learning
Approach where features are partitioned horizontally between different entities collaborating on the same model.
Model Partitioning
Process of strategically dividing an AI model into segments distributed across different infrastructures.
Privacy-Preserving ML
Set of techniques enabling machine learning without exposing users' sensitive data.
Communication Overhead
Cost in bandwidth and latency induced by exchanges between client and server in the distributed architecture.
Latency Optimization
Critical challenge in split learning aiming to minimize response delays in distributed systems.
Activation Smashing
Potential attack where intermediate activations are analyzed to reconstruct input data.
Collaborative Intelligence
Concept where multiple entities combine their computational resources for mutually beneficial learning.
Model Inference
Phase of using the trained model to make predictions, distributed in the context of split learning.