Natan Katz
Biography
Natan Katz is a data scientist and researcher working at the intersection of machine learning, security, and algorithmic robustness. His study spans adversarial machine learning, model behavior analysis, and BNN. I am a cofounder of LuminAI security, which develops tools for malicious behavior and risks in open-source models In the lecture I will discuss theory of these attacks
Co-founder, Chief Scientist
? I am building a novel cyber tool for securing AI models using a statical algorithmic approach
? Fundraising
? Searching for potential design partners and setting PoCs
? Developing the core algorithmic engine
Research Interest
Cyber for AI, open-source models, activation analysis, LLM, VCs, investors,
PoC
Abstract
Open source models are extremely useful tools in contrast to API LLM such as OpenAI or Bedrock; open-source models allow users a control over the model weights and activations during inference, easy fine-tuning, and quick development of models for niche problems. However, open-source models are a source for supply chain attacks, particularly backdoors and adversarial examples. We will cover the mathematical concepts of some of the common attacks and present how LuminAI we handle these attacks using statistical tools such as Bayesian inference and BNN.
