When: Monday, October 8, 2018
Where: PGH 563
Time: 11:00 AM

Modeling Noisiness to Recognize Named Entities using Multitask Neural Networks on Social Media

Speaker: Gustavo Aguilar

Recognizing named entities in a document is a key task in many NLP applications. Although current state-of-the-art approaches to this task reach a high performance on clean text (e.g. newswire genres), those algorithms dramatically degrade when they are moved to noisy environments such as social media domains. We present two systems that address the challenges of processing social media data using character-level phonetics and phonology, word embeddings, and Part-of-Speech tags as features. The first model is a multitask end-to-end Bidirectional Long Short-Term Memory (BLSTM)-Conditional Random Field (CRF) network whose output layer contains two CRF classifiers. The second model uses a multitask BLSTM network as feature extractor that transfers the learning to a CRF classifier for the final prediction. Our systems outperform the current F1 scores of the state of the art on the Workshop on Noisy User-generated Text 2017 dataset by 2.45% and 3.69%, establishing a more suitable approach for social media environments.

Bio:

Gustavo Aguilar received his Bachelor's degree in Computer Science from Universidad Centroamericana José Simeón Cañas in 2015 (El Salvador). He is currently pursuing a Ph.D. degree in Computer Science at the University of Houston. Before joining the Ph.D. program, Gustavo worked for three years as a game developer at Beyond Games. Last summer, he interned at Amazon Alexa as an applied machine learning scientist. His research interests focus on Natural Language Processing and Machine Learning.

Hexahedral Mesh Structure Visualization and Evaluation

Speaker: Cotrik (Kaoji) Xu

Understanding hexahedral (hex-) mesh structures is important for a number of hex-mesh generation and optimization tasks. However, due to various configurations of the singularities in a valid pure hex-mesh, the structure (or base complex) of the mesh can be arbitrarily complex. In this work, we present a first and effective method to help meshing practitioners understand the possible configurations in a valid 3D base complex for the characterization of their complexity. In particular, we propose a strategy to decompose the complex hex-mesh structure into multi-level sub-structures so that they can be studied separately, from which we identify a small set of the sub-structures that can most efficiently represent the whole mesh structure. Furthermore, from this set of sub-structures, we attempt to define the first metric for the quantification of the complexity of hex-mesh structure. To aid the exploration of the extracted multi-level structure information, we devise a visual exploration system coupled with a matrix view to help alleviate the common challenge of 3D data exploration (e.g., clutter and occlusion). We have applied our tool and metric to a large number of hex-meshes generated with different approaches to reveal different characteristics of these methods in terms of the mesh structures they can produce. We also use our metric to assess the existing structure simplification techniques in terms of their effectiveness.

Bio:

Kaoji Xu is a fifth year PhD student who is interested Geometry Modeling, processing, and Visualization. Specifically, he is doing research on hexahedral mesh generation, optimization, and visualization. He has solid skills in Software Engineering, Performance Analysis and Optimization.

UWB Physical Layer Adaptation for Best Ranging Performance within Application Constraints

Speaker: Hessam Mohammadmordi

Indoor localization has been a hot area of research for many years. There are many research proposals and commercial products which can accurately locate moving objects inside the buildings. Ultra-wideband signals can locate objects with less than 5 cm error in the reasonably inexpensive price. Despite very accurate localization achievable by UWB based systems, building a robust and reliable indoor localization system based on UWB signals remains as a challenge. In this paper, we investigated the impact of different channel configuration parameters on the robustness of UWB-based indoor localization. Based on conducted experiments, we proposed an efficient algorithm to find the best setting for UWB communication channel to meet accuracy, power, and air utilization requirements. We evaluated the performance of our framework in real world environment scenarios, and our results show an average 20% reduction in rang errors achieved by our proposed method through proper setting of UWB physical layer parameters.

Bio:

Hessam, is PhD student at department of computer science, at University of Houston, working under supervision of Prof. Gnawali. His research interests are mostly about wireless sensor networks and IoT application design and evaluation. In his PhD thesis, he is trying to build a robust, accurate and reliable indoor positioning system utilizing ultra-wideband (UWB) signals.