But when you add something new to an ML pipeline, there's new ways for adversaries to wreak havoc. We introduce a new type of data poisoning attack exploiting the dynamic nature of unlearn requests: a "camouflage" attack, which lies dormant until triggered by the adversary. 3/n
RESEARCH
-
Machine Unlearning Vulnerability Enables Poisoning Attacks
By
–
New paper led by Jimmy Di & Jack Douglas, co-advised with @AcharyaJayadev @ayush_sekhari
: "Hidden Poison: Machine Unlearning Enables Camouflaged Poisoning Attacks" The adaptive nature of machine unlearning opens a new attack vector for adversaries. https://
arxiv.org/abs/2212.10717 1/n -
Demis Hassabis on DeepMind Strategy and AI Publication Policy
By
–
Interesante entrevista a Demis Hassabis, CEO de DeepMind, sobre su trayectoria, visión, y opinión sobre la política de publicación de otras compañías. La mención a los "freeloaders", uff. https://
time.com/6246119/demis-
hassabis-deepmind-interview/
… -
Hallucinations of Generative AI: Intentional Design Features
By
–
#Hallucinations of #GenerativeAI – by design!
-
DeepMind Connects Gradient Meta-Learning with Convex Optimization
By
–
DeepMind Explores the Connection Between Gradient-Based Meta-Learning and Convex Optimization
-
DeepMind CEO hints at 2023 Sparrow beta release citing sources
By
–
CEO of Google's DeepMind @demishassabis hinted today at a 2023 private beta release of "Sparrow" (DeepMind's version of ChatGPT) that will be capable of citing sources for its responses.
-
Batch vs Epoch Level Logging Best Practices Training
By
–
AFAIK you pretty much always want batch-level logging for training and epoch-level for validation (and end-of-epoch training)?
-
DeepMind CEO Urges Caution on Mainstream AI
By
–
DeepMind’s CEO Helped Take #AI Mainstream. Now He’s Urging Caution https://
ti.me/3W6wDMr -
Collaborative Research on LLM Misuse and Mitigation Strategies
By
–
Collaborative research on LLM misuse & mitigation. Like for any new technology, important to maximize the upsides and mitigate the downsides: