What Is a Backdoor Attack on an AI Model?
A backdoor attack makes an AI model behave normally most of the time but produce an attacker-chosen result when a hidden trigger appears.
Don't Just Read it – Explore It
A backdoor attack makes an AI model behave normally most of the time but produce an attacker-chosen result when a hidden trigger appears.
Data poisoning deliberately contaminates training data so an AI model learns distorted patterns, weaker performance, or attacker-chosen behavior.
Secure aggregation lets a server learn the combined value of many client updates without learning the individual update submitted by each client.
Task arithmetic treats the parameter change created by fine-tuning as a vector that can be added, removed, scaled, or combined with other task vectors.
Federated averaging trains a shared model by aggregating updates produced locally on many clients rather than centralizing their raw training data.
Model merging combines parameters or parameter changes from multiple trained models to create a new model without ordinary joint retraining.
Ring all-reduce arranges workers in a logical circle and circulates data chunks so reduction and redistribution use network bandwidth efficiently.
Expert parallelism distributes the expert subnetworks of a mixture-of-experts model across devices and routes tokens to the selected experts.
All-reduce combines values from every training worker with an operation such as a sum and returns the combined result to every worker.
Activation checkpointing saves only selected intermediate states during training and recomputes the others during backpropagation to reduce memory use.
Mixed-precision training uses lower-precision arithmetic where it is safe and higher precision where numerical range or stability is needed.
AI agents are systems in which a model can choose steps and use tools to pursue a goal, adapting its plan as results come back.