We propose a new family of label randomizers for training regression models under the constraint of label differential privacy (DP).

Among all the contributing factors, the quality and selection of data is becoming increasingly recognized for its importance in training LLMs effectively.

An adversary may be able to "steal" the model by querying for counterfactuals and training a

In this paper, we introduce a new type of model backdoor: the privacy backdoor attack .

The rapid scaling of language models is motivating research using low-bitwidth quantization. In this work, we propose a novel binarization technique for Transformers applied to machine translation (BMT), the first of its kind.

AI systems can take harmful actions and are highly vulnerable to adversarial attacks.