Erase
PulseAugur coverage of Erase — every cluster mentioning Erase across labs, papers, and developer communities, ranked by signal.
2 day(s) with sentiment data
-
New defenses tackle prompt injection in LLMs with self-distillation and adversarial training · 2 sources tracked
Researchers have developed two novel methods to combat prompt injection attacks in large language models. The first, ERASE, uses on-policy self-distillation to replace malicious injected content in multimodal inputs (im…
-
New methods optimize visual tokenization for large vision-language models · 4 sources tracked
Researchers have developed several new methods to optimize visual tokenization for large vision-language models (VLMs), aiming to reduce computational overhead while maintaining accuracy. MixTok introduces a hybrid appr…
-
New ERASE framework enables AI model forgetting without weight changes
Researchers have introduced ERASE, a novel framework designed for on-the-go forgetting of private data within AI models without altering their weights. This method utilizes adversarial signal editing and class-condition…
-
New ERASE method speeds up recommendation system training by 9.51%
Researchers have developed a new training method called ERASE (EaRly bAckpropagation SchEdule) designed to accelerate the training of recommendation systems. ERASE reinterprets the detachment mechanism of Forward-Forwar…