Date of Award

Summer 2020

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

School

Computing Sciences and Computer Engineering

Committee Chair

Dr. Andrew H. Sung

Committee Chair School

Computing Sciences and Computer Engineering

Committee Member 2

Dr. Bikramjit Banerjee

Committee Member 2 School

Computing Sciences and Computer Engineering

Committee Member 3

Dr. Ramakalavathi Marapareddy

Committee Member 3 School

Computing Sciences and Computer Engineering

Committee Member 4

Dr. Parthapratim Biswas

Committee Member 4 School

Mathematics and Natural Sciences

Committee Member 5

Dr. Sungwook Lee

Committee Member 5 School

Mathematics and Natural Sciences

Committee Member 6

N/A

Abstract

Researchers in multiple disciplines have recently adopted deep learning because of its ability of high accuracy representation learning from big and complex data. My research goal in this thesis is developing deep learning models for information diffusion analysis on social networks and collective tasks learning in swarm robotics. Firstly, the information diffusion on social networks is modeled as a multivariate time series in three dimensions with ten features. Then, we applied time-series clustering algorithms with Dynamic Time Warping to discover different patterns of our models. Then, we build a prediction model based on LSTM, which outperforms traditional time-series prediction methods. Extending the first study, we conduct the second study to measure the users' influence on social networks. Our deep learning model outperforms the baseline Linear Influence Model in modeling the global influences of nodes over time and predicting the temporal volume of information diffusion processes. Our third study proposes the Multi-Feed Weighted Topic Embeddings (MFWTE) model to analyze user network interactions and topic diffusion patterns on Twitter. Our model is evaluated on the friendship recommendations and retweet link prediction tasks. The performance shows that our MFWTE model outperforms various single feed methods and can help to study topic diffusion patterns. In swarm robotics, we employ deep reinforcement learning methods to solve collective task learning problems. The fourth study proposes a cumulative training method using transfer learning to develop a multi-robot collision avoidance navigation system. Our approach improves the shared policy between multiple agents through transfer learning, reward shaping, and multi-stage training. The performance shows that our method helps to iii produce a robust navigation plan that generalizes well to complex indoor scenarios. Finally, we propose a new multi-robot foraging approach based on Reinforcement learning as a Rehearsal framework. This approach takes both the visible and hidden features, trains through Deep Q-Learning, and generates a robust policy to decide the role choice between walker and beacon. While the hidden features are available during training time, we make them available during execution time by learning a generation model based on Mixture Density Networks through deep learning.

ORCID ID

0000-0002-6695-2034

Share

COinS