PhD student in Computer Science at the University of Warwick, working on reinforcement learning for text-to-speech.
- Reinforcement learning for speech and text-to-speech
- High-performance machine learning systems
- Model optimisation, compilation, and inference efficiency
- Distributed training and ML infrastructure
- Low-level systems, compilers, and compute tooling
- Languages: Python, Rust, Go, C++, C, TypeScript, Haskell
- ML / Research: PyTorch, JAX, Reinforcement Learning, TTS, Distributed Computing
- HPC / Systems: LLVM, MPI, OpenMP, CUDA, Vulkan
- Tooling / Web: Postgres, Svelte, Vite, Tailwind
-
JRAP: JPEG Resistant Adversarial Perturbations to Disrupt Diffusion Based Inpainting
- At time of writing SOTA method, for protecting images against diffusion inpainting models.
- Works on consumer graphics cards under 8GB of VRAM.
-
vk-rs-bindings: Auto-generated Vulkan bindings for Rust
- Feature complete Vulkan bindings for Rust, generated from the official Vulkan XML registry.
- All +400 vulkan extensions, versions, and APIs are feature gated, for extra compile time validation against the Vulkan specification.
- Provides thin lifetime safe handles for better ergonomics and memory safety.
- Vulkan loader and GPU memory allocator included.
- LinkedIn: jakub-czarlinski