Boxiang Dong, Ph.D
Department of Computer Science
Montclair State University
Office: CCIS 227E
Email: dongb AT montclair DOT edu
Prior to that, I received my Ph.D in Computer Science in 2016 from Department of Computer Science of Stevens Institute of Technology, which is located in NJ, USA. My Ph.D advisor is Prof. Wendy Wang. The title of my Ph.D dissertation is "Privacy-preserving and Authenticated Data Cleaning on Outsourced Databases" [PDF] [Slides].
This is my CV (last updated: 09/2018).
|2020-04||Our paper "DeepICU: Imbalanced classification by using Deep Neural Networks for Network Intrusion Detection" has been accepted by IJBDI!|
|2019-11||Our paper "AuthPDB: Authentication of Probabilistic Queries onOutsourced Uncertain Data" has been accepted by CODASPY 2020!|
|2019-10||Our paper "Integrity Authentication for SQL Query Evaluation on Outsourced Databases: A Survey" has been accepted by TDKE!|
We propose a deep neural network infrastructure to detect network intrusion attacks.
To cope with the imbalanced distribution of various types of attacks, we propose a novel loss function that eliminates the bias towards the majority class.
We protect the privacy in the setting of crowdsourcing in two ways.
First, we protect task privacy by designing a task assignment strategy that minimizes the risk of information disclosure in the existence colluding workers.
Second, we protect worker privacy by designing an answer perturbation mechanism to provide local differential privacy.
We design a new authentication and indexing data structure--Merkle Bed Tree (MB-Tree in short), which obtains Merkle tree's authentication functionality and preserves Bed tree's indexing property.
The server constructs verification objects (VOs) by traversing the MB-Tree and visiting the Euclidean space to prove the result's soundness and completeness.
The client's verification cost is very small because of MB-tree's pruning power and the cheap Euclidean distance computation cost.
We define two security models: &alpha-security against frequency-analysis attack, and indistinguishability against functional dependency preserving chosen plaintext attack (IND-FCPA).
We design a frequency-hiding FD-preserving probabilistic encryption scheme that provides strong security guarantee, while preserves the utility in the encrypted dataset.
We model the system events in every host as a multipartitie graph of interactions between processes, sockets, and files.
We apply random walk on the graph to learn the routine behavior of each system entity based on a transition probability model.
We label an event sequence as abnormal if the behavior of any involved entity largely deviates from its routine role.
In order to defend against the frequency-analysis attack in the outsourced data-deduplication services, we propose two approaches.
|LSHB approach||EHS approach|
We aim at verifying if the untrusted server returns the correct and complete frequent itemset mining result.
To accomplish the goal, we require the server to return cryptographic proofs to show the support of the itemsets. The proof is based on the Secure Set Intersection Protocol.
CSIT212 Data Structures and Algorithms Fall 2017, 2018, Spring 2018
CSIT111 Fundamentals of Programming I Fall 2017
CSIT100 Introduction to Computer Science Spring 2017
CSIT345 Operating Systems Spring 2017
Introduction to Cybersecurity (part of Stevens Pre-College Programs) Summer 2016
Jan.2017 - Now
Sep.2014 - Dec.2014
May.2015 - Aug.2014