Project Information

Objective

Human-generated passwords are particularly susceptible to guessing attacks as a consequence of the limitation of precise recall and hence are not random. In this project, we make use of the excellent expressive power of sequence modeling neural networks such as LSTMs and GRUs to effectively guess passwords as compared to other cutting-edge password guessing techniques like Markov models, JtR, and PCFGs.

Achievements

  • Trained LSTM, GRU, and GAN seq2seq models totaling 4.8M+ parameters to generate human-like passwords that matched ∼55% of the 14M+ passwords within 10^9 guesses.
  • Research paper was selected by Springer Singapore to be presented at an international conference.

    • Tech Stack

    Embedding Model, LSTM, GRU, GAN