Adapting kernel-level scheduling, memory management, and I/O to support the resource demands of modern machine learning workloads, particularly on minimal or constrained operating systems.
Distributing model inference across clusters of resource-constrained devices, including sharding strategy, fault tolerance, and communication-efficient coordination between nodes.
Mapping neural network computation onto open instruction set architectures, with an emphasis on vectorized execution and hardware-software co-design for edge and embedded targets.
Investigating how compact a model architecture can be made without sacrificing performance, through architecture search, knowledge distillation, and quantization aimed at constrained hardware.
A two-part tool where students write Prisoner's Dilemma strategies, run every simulation on their own laptop, and submit only the code to a server that trusts none of it.
How to fork, write, and submit a post, and a full reference for the Markdown this blog renders, including math.
Who we are, and the kind of hands-on systems work you'll find on this blog.
Systems Research @ IBA's website. Next.js, TypeScript, Tailwind CSS v4. Blog content is pulled in from a separate repo via git submodule; project listings are fetched live from GitHub.
Markdown source for the Systems Research @ IBA blog. No CMS: fork, write a post, open a pull request. Rendered into the main site via git submodule.
AlphaZero-style Othello engine: MCTS guided by a self-play-trained CNN, distilled for deployment. FastAPI backend on Cloud Run. Play it: https://site-zeta-rust-98.vercel.app/playground/othello