Fall 2019

This project will result in a visual analytics toolkit that will enable social scientists to understand the cultural groups and biases at play in a social dataset. News, books, and social media all contain biases that stem from the cultural background of the author(s). We have developed algorithms to identify the cultural groups at play in an arbitrary dataset, as well as natural language processing approaches that can discover the biases of each group. This project would help bring put these tools into the hands of social scientists by displaying the output of these algorithms in novel visualizations.

This project will result in an open-source software tool that will have general applicability for scientific publications. Papers with hundreds of authors are not uncommon in science, and it often takes many weeks to compile an author list in the desired order with proper affiliations and acknowledgments. We have implemented an algorithm that generates the author information for a paper based on the type of contribution of each author within the ENIGMA neuroscience consortium. This project would extend this software to read in compiled spreadsheets or forms and extract information about universities and other institutions from structured web sources, to interoperate with widely-used frameworks such as Wikidata.

This project will result in a Python package for automated time series analysis. Based on the characteristics of the data, you will design functions that (1) perform essential tasks in data cleaning and select appropriate methodologies, (2) implement various algorithms currently not supported through pandas and scikit-learn, and (3) create appropriate visualizations.

While many open data efforts have managed to successfully expose public data in the web, it is often complicated to determine how these records can be integrated with each other (due to heterogeneous ids, not clear how to place them into a map, etc.). In this project, the student will leverage the novel techniques for integrating, registering and connecting datasets with overlapping elements. The results will be visualized by the student using interactive maps.

Public sports data is often spread across many differing sources, creating issues of entity resolution and record linkage. Knowledge graphs are a popular conceptual technology for storing, fusing and querying information from such disparate sources. This project will focus on building a sports data knowledge graph, from various open data and asset (e.g. video) sources/API, based on a Wikidata infrastructure.

Documenting how a result was obtained from data analysis involves documenting the software, software settings, and datasets used to obtain that result so it can be explained properly. This project will design and develop a user interface for specifying provenance records using W3C standards. The interface will enable users to document the provenance of data analysis no matter what infrastructure they used (R scripts, sk-learn, etc).

A lot of data is available in the web in a tabular manner, but it’s difficult to manipulate and visualize without a significant effort. In this project, we aim to test a novel framework created at ISI to build and visualize knowledge bases. The objective is to create a knowledge base that extends the other resources in the Web such as Wikidata or Wikipedia, and visualize the results using interactive maps and plots.

Developing software and data resources for USC students. Software resources include tools to process and analyze specific types of data (eg social networks, images, text, etc), data preparation tools, or machine learning libraries. Data resources include thematic data repositories, such as urban LA data, environmental LA data, entertainment LA data, etc.

Estimates of climate variations over the past 1,000 years play an increasing role in climate assessments. A key quantity to derive from them is the transient climate response (TCR), which quantifies the warming at expected from slowly-rising CO2 concentrations. TCR helps constrain the climate models used to predict the future evolution of Earth’s climate. In this project, you will help design an efficient workflow to estimate TCR from existing paleoclimate datasets and emerging statistical methods.

Spread of misinformation has become a significant problem, raising the importance of relevant detection methods. While there are different manifestations of misinformation, in this work we focus on detecting face manipulations in videos. This project will focus on detecting face manipulations in videos. We exploit the temporal dynamics of videos with recurrent networks.