Visual computing tasks such as computational imaging, 3D graphics, image/video understanding, generative content creation (images, videos, 3D models, interactive worlds), and AI-driven problem solving (agents) are key responsibilities of modern computer systems ranging from sensor-rich smart phones, autonomous robots, and large datacenters. These workloads demand exceptional system efficiency and this course examines the key ideas, techniques, and challenges associated with the design of parallel, heterogeneous systems that accelerate visual computing applications. This course is intended for systems students interested in architecting efficient graphics, image processing, and computer vision platforms (both new hardware architectures and domain-optimized programming frameworks for these platforms) and for graphics, vision, and AI students that wish to understand throughput computing principles to design new algorithms that map efficiently to these machines.
| Apr 01 |
|
|
Discussion of modern visual computing applications, a design exercise
|
| Apr 03 |
|
|
Algorithms for taking raw sensor pixels to an RGB image: demosaicing, sharpening, correcting lens aberrations, multi-shot alignment/merging, multi-scale processing with Gaussian and Laplacian pyramids, HDR (local tone mapping)
|
| Apr 08 |
|
|
The Frankencamera, modern camera APIs, advanced image understanding responsibilities of digital cameras (portrait mode, autofocus, etc)
|
| Apr 10 |
|
|
Key optimization ideas, a detailed look at Halide's scheduling algebra
|
| Apr 15 |
|
|
Hardware Acceleration: What's in a Modern GPU + AI Accelerators
GPUs, AI accelerators, special instructions for DNN evaluation (and their efficiency vs custom ASIC), choice of precision in arithmetic, flexibility vs efficiency trade-offs
|
| Apr 17 |
|
|
The importance of predictable control in content creation. Techniques for inserting new forms of control into generative image synthesis, role of human-interpretable abstractions.
|
| Apr 22 |
|
|
Aligning controls with user expectations, forms of "loose" control, how modern systems increasingly combine traditional symbolic structures with learned structures. The role of code as a representation for content. What should be learned and what should kept interpretable?
|
| Apr 24 |
|
|
A look at the data curation and data cleaning pipelines that select data that goes into modern generative AI models.
|
| Apr 29 |
|
|
Potential impact on creating functional content, role of agents as verifiers, implications to real-world robotics, etc.
|
| May 01 |
|
|
Motivation for batch simulators (training RL agents), the design of the Madrona simulation engine.
|
| May 06 |
|
|
More detail on fast simulation platforms for training agents in virtual worlds.
|
| May 08 |
|
|
Generative AI models that yield interactive worlds, their relationship to traditional simulation engines, when are models to desirable? What is a future simulation engine?
|
| May 13 |
|
|
LLM-based techniques for creating agents can learn skills in virtual worlds.
|
| May 15 |
|
|
Techniques for using AI agents to simulate humans and human populations.
|
| May 20 |
|
|
H.264 video representation/encoding, parallel encoding, motivations for ASIC acceleration, ML-based compression methods, emerging opportunities for compression when machines, not humans, will observe most images
|
| May 22 |
|
|
Systems for Developing AI agents
The DSPy system for efficiently developing and executing agentic designs.
|
| May 27 |
|
|
The role of differentiable rendering and (differentiable programming in general) to recover 3D shapes/scenes, textures, materials, etc. from images.
|
| May 29 |
|
|
How the Slang language provides flexible support for auto-differentiation. Understanding the difference between mechanism and policy in system design.
|
| Jun 03 |
|
|
Project Presentations
Students present projects.
|
| Jun 4 | Term Project Information |