Introduction

I’m Minghao Li. I’m a 3rd-year Computer Science PhD student at Harvard University working with Prof. Minlan Yu. Previously, I got my BS degree in Computer Science at Cornell University.

My CV is available here.

Publication

• Towards Automated Safety Vetting of Smart Contracts in Decentralized Applications
  Duan, Y., Zhao, X., Pan Y., Li, S., Li, M., Xu, F. & Zhang, M.
  Accepted at CCS’22

• PECAM: Privacy-Enhanced Video Streaming & Analytics via Securely-Recoverable Transformation
  Wu, H., Tian, X., Li, M., Liu, Y., Ananthanarayanan, G., Xu, F. & Zhong, S.
  Accepted at MobiCom’21

• DAPter: Preventing User Data Abuse in Deep Learning Inference Services
  Wu, H., Tian, X., Gong, Y., Su X., Li, M. & Xu, F.
  Accepted at TheWebConf(WWW)’21

Report

• Analyzing the Security of Smart Contracts Using Neural Networks
  Available here.
  Minghao Li

Projects

• DAPter
  • A user-side DLIS(deep learning inference service)-input converter. DAPter removes unnecessary information with respect to the targeted DLIS. The converted input data by DAPter retains good inference accuracy and is difficult to label manually or automatically for new model training.
  • DAPter’s conversion is empowered by a lightweight generative model trained with a novel loss function to minimize abusable information in the input data.
  • DAPter requires no change in the existing DLIS backend and models.
• PECAM
  • A mobile privacy-enhanced VSA (Video Streaming & Analytics) system that runs on the front end of a VSA system and performs the reversible privacy-enhanced whole-frame transformation over real-time streaming.
• DAPPSCOPE
  • Automatically discovers the discrepancy between a DApp’s UI and its contract code.
  • DAPPSCOPE analyzes a DApp’s web code to discover all the smart contract functions that are invoked. Builds a business model graph for each of these functions. Locates the corresponding UI component of each function, and applies NLP techniques to the UI texts to infer the category of the business logic and the function’s particular task. Discovers underlying issues by checking the business model graph against the pre-defined high-level specifications of the inferred business logic and task.
  • Applied to 22 real-world DApps. Discovered 17 novel safety issues.
• BookHub
  • Generates book recommendations based on user preferences. Allows users to input a list of books (up to 5) and a list of genres (up to 10) they like. Then recommends a list of books of the preferred genres with Goodreads reviews that are similar to the reviews of users’ favorite books.
  • Uses information retrieval and machine learning techniques including TF-IDF, cosine similarity, Jaccard similarity, Boolean search, and topic modeling.