Selected Publications
TOP: Backdoor Detection in Neural Networks via Transferability of Perturbation
Todd Huster, and Emmanuel Ekwedike
Brief Description: Deep neural networks (DNNs) are vulnerable to “backdoor” poisoning attacks, in
which an adversary implants a secret trigger into an otherwise normally functioning model. Detection of
backdoors in trained models without access to the training data or example triggers is an important open
problem. In this paper, we identify an interesting property of these models: adversarial perturbations
transfer from image to image more readily in poisoned models than in clean models.
□PDF
Feedback-Based Tree Search for Reinforcement Learning
Daniel R. Jiang, Emmanuel Ekwedike, and Han
Liu
International Conference on Machine Learning, ICML 2018.
Brief Description:We describe a technique that iteratively applies MCTS on batches of small,
finite-horizon versions of the original infinite-horizon MDP. We show that a deep neural network
implementation of the technique can create a competitive AI agent for a popular multi-player online battle
arena (MOBA) game.
□PDF
Highly Scalable Verifiable Encrypted Search
Whitney A. Drazen, Emmanuel Ekwedike, and
Rosario Gennaro
Conference on Communications and Network Security, CNS
2015.
Brief Description: In encrypted search, a server holds an encrypted database of documents but not
the keys under which the documents are encrypted. The server answer keyword queries from a client with the
list of documents matching the query. In this paper we present two highly scalable protocols to search over
encrypted data which achieve full security against a possibly malicious server and supports conjunctive
queries where the client submits many keywords and is asking the server to identify the documents that match
all the keywords.
□PDF
Group Symmetries and Bike Sharing for M/M/1/k Queueing Transience
Emmanuel Ekwedike, Jamol J. Pender, Robert C.
Hampshire, and William A. Massey
Brief Description: Fundamental stochastic models for studying the dynamics of bike-sharing systems
can be found within the transient behavior of the M/M/1/k queue and related stopped processes. We develop
new techniques involving group symmetries and complex analysis to obtain exact solutions for their
transition probabilities. These methods are based on the underlying Markovian structure of these random
processes and do not involve any indirect analysis from using generating functions or Laplace transforms.
□PDF
Optimal Decision Making via Stochastic Modeling and Machine Learning: Applications to Resource Allocation Problems and Sequential Decision Problems
Emmanuel Ekwedike.
Brief Description:
This thesis is about optimal decision making for resource allocation problems and sequential decision problems. There are two parts to this thesis: the first part focuses on developing a new stochastic analysis of queues and algorithms necessary to assist in the management of bike-sharing systems. The second part focuses on developing a general learning framework for sequential decision problems.
□PDF
