High Performance Computing Cluster for NLP and Deep Learning Approaches to Understanding Social, Psy
Abstract
Dr. Rebecca Goolsby | Social Networks and Computational Social Science | ONR Code 341Publicly Releasable Project Summary/Abstract In, the 21st century security environment, the information domain has become a primary arena of both nation-state and non-state actor c,onflict. Populations are influenced, democracies disrupted, and human behaviors shaped through manipulations of information and expl,oitation of information channels. Military services and their information operations and intelligence directorates need faster and m,ore accurate insights into how the range of disinformation and influence techniques an adversary brings to bear can influence a popu,lation, from deep-fakes to adversarial framing to botnets and troll farms. A common characteristic of investigations into disinforma,tion techniques and technologies is a need for high performance computing to process massive amounts of data, run generative network,s, and train deep learning models.In order to support multidisciplinary, DOD-funded research into disinformation, information confli,ct, and influence currently underway at Arizona State University (ASU), we propose to acquire a five-node, high performance computin,g system. Each node will include the necessary memory, CPU power, and GPU power to efficiently run Natural Language Processing tasks,, such as training of language models or generation of synthetic data for development of detection and attribution algorithms. Each, node will include a CPU module with AMD processors with 64 cores, 512GB RAM, and four 40GB A100 NVIDIA GPUs. The proposed equipment, will expand research capabilities for several existing DOD-funded research projects at ASU, and will enable new opportunities for D,OD research in a variety of disciplines. The five-node array will allow for efficient computation of especially large tasks, and mor,e importantly enable the simultaneous execution of multiple research tasks engaged in: Development of new capabilities for informat,ion environment assessment in dynamic information contestsNew technologies for detecting social hysteria propagation, crowd manipul,ation and group polarization processesCyber conflict resolution and understanding the problem of influence, socially, psychological,ly and cross-culturallyTo robustly answer questions about how to rapidly attribute synthetically-generated deceptive media artifacts, to nefarious actors, how to accurately identify and characterize and counter cross-cultural vectors of influence, and how to detect, propaganda and disinformation using unsupervised machine learning, multidisciplinary collaboration is required between psychologist,s, anthropologists, communication scholars, media scholars, as well as computer scientists, data mining experts, and data visualizat,ion specialists. At ASU, these collaborations are ongoing and facilitated by the Center on Narrative, Disinformation, and Strategic, Influence (NDSI, directed by PI Ruston) in collaboration with the Center for Strategic Communication (CSC, directed by Co-PI Corman,) and the Data Mining and Machine Learning Lab (DMML, directed by Co-PI Liu). These three research organizations have five projects, currently underway engaged in detection, attribution, and characterization of disinformation and/or covert online influence in diff,erent cultural contexts. Additionally, NDSI sponsors ASUs Disinformation Working Group, assembling approximately 20 scholars from a,cross the university to engage with mission-relevant research supporting the research areas articulated above.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 01, 2022
- Source ID
- N000142212291
Entities
People
- Scott Ruston
Organizations
- Arizona State University
- Office of Naval Research
- United States Navy