CENTRA 8 - Taiwan

• Name badge and swag pick-up.
• Laptop charging.
• General questions.

COVID-19 Protocol
While we hope that everyone arrives healthy and without incident, we know many have traversed long distances in confined spaces and life can be unpredictable. If you find yourself feeling ill (or having tested positive for COVID-19), please opt to self-quarantine and alert a member of the organizing committee at pti.iu.edu so we may assist with any necessary arrangements.

Please be respectful of those who choose to social distancing and wear masks.

Susumu Date, Chair | ~15-20 minutes per presenter (including Q/A)

• Patra Model Cards for Federated Learning – Beth Plale, Professor, IU, United States
• A Case Study of Federated Learning System for Pill Prediction Application – Truong Thao Nguyen, AIST
  • YouTube Video embedded in presentation
• Federated Contrastive Domain Adaptation for Category-Inconsistent Object Detection – Wei-Yu Chen, Researcher, NCHC, Taiwan

The rapid development of large language models (LLMs) has introduced unprecedented possibilities in the field of humanities research. In Taiwan, the TAIwan HUmanities Conversational AI Knowledge Discovery System (TAIHUCAIS) integrates unique Taiwanese data with advanced AI technologies to address challenges related to fragmented data management and complex query requirements in humanities research. Grounded in user experience research, the system offers an intuitive conversational platform that streamlines the knowledge discovery process to meet the needs of humanities scholars. This presentation will share the development journey of TAIHUCAIS, including real-world application outcomes, and explore challenges encountered while integrating Taiwanese humanities data, such as difficulties in obtaining data authorization and issues of LLMs bias. Additionally, it will explore how these experiences can inspire research endeavors in other cultural and linguistic contexts, paving the way for future collaborations between LLMs and IT, and further promoting innovation and international connectivity.

~15 minutes per presenter (including Q/A):

• Investigating the Impact of Data Storage Mechanisms on Distributed Software-Defined Networking Performance - Kang Xingyuan, NIST, Japan
• Reducing Overhead in Time-Sliced GPU Sharing with NVIDIA MPS in Kubernetes - Papon Choonhaklai, NIST, Japan
• A Proposal of an Efficient Path Selection Method Using INT-Based Delay Measurement - Kota Nakagawa, NIST
• LLM Assisted Online Root-Causing of Misconfiguration Diagnosis in Enterprise Computing Systems - Krishna Kant, Temple University, United States
• Rule-base Multitask Learning Model Transfer Learning on Multispectral and Multitemporal UAV Image for Disaster Identification - Hsin-Hung Tseng, NCHU, Taiwan

Please feel free to use this Google Doc to take notes of your group's discussion.

For new projects developed during CENTRA 8’s Project Group Discussions, project leads will be asked to complete and submit a Project Description Form.
For updates to CENTRA Projects listed on CENTRA’s website, project leads are welcome to use the form.

Renato Figueiredo, Chair | ~15 minutes per presenter (including Q/A):

• A Greener Path: Improving AI training energy efficiency through Workload-Aware GPU Tuning - Ricardo Macedo, INESC TEC, Portugal
• P4 Testbed and Cross-Site IN-band Telemetry - Te-Lung Liu, NCHC, Taiwan
• Infrastructure Management at HLRS, Germany - Uwe Woessner, HLRS, Germany

• Discussion groups report back
• Steering committee report back
• Announcing next CENTRA