About the Speaker
The paper discusses about reversing the deep learning AI model revealing its architecture, critical hyperparametser that can be exploited by malicious actor. This reversing is beyond finding password, keys and buffer overflows. The paper will discuss detail analysis of reversing the model from different models viz. googlenet, llama etc and various formats such as hd5, onnx and bin. The parameters discussed after reversing are related to tensors in deep learning models viz. sparsity of matrix, architectural flow, weights and biases that are fundamentals to any AI model. Also for language models the tokanizer reversing through model will be discussed. In short mathematical structure of deep learning model will be reversed
Yashodhan is a Security Researcher with over 13 years of cutting-edge experience at the intersection of IoT and AI innovation. A tech visionary currently pursuing a doctorate in Satellite and Security, Yashodhan’s academic journey spans M.Tech in Satellite Communication, M.Tech in Signal Processing, and a B.E. in Electronics & Telecommunication.
He has spearheaded the development of industrial IoT products infused with AI, blending innovation with functionality. His research portfolio encompasses Signal Processing, RF and Electromagnetics, and CubeSat technologies. Notably, he has designed his own CubeSat and an astrophotonics chip, showcasing his passion for pushing the boundaries of technology.
With experience in security assessments, threat modeling, and architecting secure IoT solutions, Yashodhan tries to bridge the gap between robust security and futuristic innovation.