Teaching
Teaching
CS 8803: Large Scale & Real-Time Visual Analysis
Monday & Wednesday, 12:30 pm–1:45 pm
Howey Physics | Room S107
The rapid growth of visual data—from video streaming, augmented reality, autonomous systems, surveillance, and wearable devices—has created unprecedented challenges in large-scale and real-time visual analysis. Effectively processing this data requires a combination of cloud-based infrastructure and edge computing, balancing scalability with low-latency decision-making. You will explore industry and academic advancements in cloud and edge-based visual processing, covering topics such as real-time inference, distributed computing, compression techniques, and AI-driven analytics. You will read and lead discussions on key papers and collaborate on a semester-long project in groups of 2-3, tackling real-world challenges in scalable and real-time visual analysis.
Familiarity with computer systems and databases at an undergraduate level
Knowledge of AI at the application level, and systems for AI recommended but not required
Understand recent developments at the intersection of AI & visual analysis—both in the cloud and on the edge
Read, analyze, and discuss recent developments from academic papers in visual analysis
Work on a term-long research project: from problem formulation to final presentation
Prior to the start of each class, students will submit responses to 1-2 quiz questions about the assigned readings. Each class, 1-2 students will give an overview of the assigned paper, then leading a discussion around open questions, future work, and related topics. One student will be assigned as a note-taker for the course on a rotating basis.
60%: Term project
5%: Proposal
10%: Mid-term presentation
10%: Mid-term report
15%: Final presentation
20%: Report
20%: Paper presentations
15%: Quizzes (1 quiz will be dropped)
5%: Participation (note-taking, class discussion)
Georgia Tech aims to cultivate a community based on trust, academic integrity, and honor. Students are expected to act according to the highest ethical standards. Any student suspected of plagiarizing an assignment or presentation, or cheating on a quiz will be reported to the Office of Student Integrity.
To avoid plagiarism:
Attribute any words or ideas from a public source
Attribute any AI system you used and how you used it (e.g. brainstorming or rephrasing)
Don’t directly copy-paste sentences from a classmate or AI system
Contact the Office of Disability Services as soon as possible to discuss your learning needs and to obtain an accommodations letter. Please also e-mail the instructor to set up a time to discuss special needs.
For a Digital Learning Day, all class activities will be held remotely via video conferencing.
Monday: Course Overview
Wednesday: How AI is Transforming Video Surveillance Analytics
Monday: BlazeIt: Optimizing Declarative Aggregation and Limit Queries for Neural Network-Based Video Analytics
Wednesday: FiGO: Fine-Grained Query Optimization in Video Analytics
Monday: Spatialyze: A Geospatial Video Analytics System with Spatial-Aware Optimizations
Wednesday: Optimizing Video Analytics with Declarative Model Relationships
Monday: ServerlessLLM: Low-Latency Serverless Inference for Large Language Models
Wednesday: Nexus: A GPU Cluster Engine for Accelerating DNN-Based Video Analysis
Monday: Orion: Interference-aware, Fine-grained GPU Sharing for ML Applications
Wednesday: Towards Efficient Large Multimodal Model Serving
Monday: Project Proposal Presentations
Wednesday: Project Proposal Presentations
Friday: Project Proposal Report Due
Monday: Warehouse-scale video acceleration: co-design and deployment in the wild
Wednesday: vbench: Benchmarking Video Transcoding in the Cloud
Monday: Mixtera: A Data Plane for Foundation Model Training
Wednesday: Guest lecture
Monday: Project Midpoint Presentations
Wednesday: Project Midpoint Presentations
Monday: Moonshine: Speech Recognition for Live Transcription and Voice Commands
Wednesday: TensorFlow Lite Micro: Embedded Machine Learning for TinyML Systems
Monday: CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
Wednesday: Pocket: ML Serving from the Edge
Monday: CLIP: Learning Transferable Visual Models From Natural Language Supervision
Wednesday: NVILA: Efficient Frontier Visual Language Models
Monday: The Shift from Models to Compound AI Systems Towards Resource-Efficient
Compound AI Systems
Wednesday: Guest lecture
Monday: Final Presentations
Wednesday: Final Presentations
Monday: Final Presentations
Wednesday: no class
Monday: Course Summary and Feedback
Wednesday: no class
Friday: Final Reports Due