Date on Master's Thesis/Doctoral Dissertation

8-2019

Document Type

Doctoral Dissertation

Degree Name

Ph. D.

Department

Electrical and Computer Engineering

Degree Program

Electrical Engineering, PhD

Committee Chair

Li, Hongxiang

Committee Co-Chair (if applicable)

Zurada, Jacek

Committee Member

Zurada, Jacek

Committee Member

Faul, Andre J.

Committee Member

Bai, Lihui

Author's Keywords

Representation learning; real-time graph embedding; dynamic spectrum access; knowledge graph; distributed graph process

Abstract

This dissertation includes two topics. The first topic studies a promising dynamic spectrum access algorithm (DSA) that improves the throughput of satellite communication (SATCOM) under the uncertainty. The other topic investigates distributed representation learning for streaming and complex networks. DSA allows a secondary user to access the spectrum that are not occupied by primary users. However, uncertainty in SATCOM causes more spectrum sensing errors. In this dissertation, the uncertainty has been addressed by formulating a DSA decision-making process as a Partially Observable Markov Decision Process (POMDP) model to optimally determine which channels to sense and access. Large-scale networks have attracted many attentions to discover the hidden information from big data. Particularly, representation learning embeds the network into a lower vector space while maximally preserving the similarity among nodes. I propose a real-time distributed graph embedding algorithm (RTDGE) which is capable of distributively embedding the streaming graph by combining a novel edge partition approach and an incremental negative sample approach. Furthermore, a platform is prototyped based on Kafka and Storm. Real-time Twitter network data can be retrieved, partitioned and processed for state-of-art tasks. For knowledge graphs, existing works cannot capture the complex connection patterns and never consider the impacts from complicated relations, due to the unquantifiable relationships. A novel embedding algorithm is proposed to hierarchically measure the structural similarity and the impacts from relations by constructing a multi-layer graph. Then, an advanced representation learning model is designed based on an entity's context generated by random walks on the multi-layer content graph.

Download

Share

COinS