Jierui Lin

Excited to share our new pre-print arxiv.org/pdf/2502.01600 We train a digital agent that solves diverse day-to-day tasks from the AppWorld benchmark by interacting with its stateful environment using API calls. AppWorld is hard! The previous best open-weight agent (Llama 3 70B) reached only a 7% success rate on the hardest test split. Our RL algorithm, LOOP - a PPO variant with Monte Carlo baselines - achieves a 45.7% success rate, 24% over the base Qwen 2.5 32B, and 9% higher than a much larger OpenAI o1.

We've built a simulated driving agent that we trained on 1.6 billion km of driving with no human data. It is SOTA on every planning benchmark we tried. In self-play, it goes 20 years between collisions.

I’ll demo the technology behind the Memory Key Photo on IOS devices: Learning iconic scenes with differential privacy. Come check out our Apple booth!

I'll be at #ICML2023 with Apple! Learn more about how to visit our booth, read our accepted papers, experience our demos, and attend our talks here: machinelearning.apple.com/updates/apple-….

For the past two years, we've been building the foundation of a new, transformative approach to self-driving. Research and innovation are central to this approach, and today I'm excited to unveil that research to the world. waabi.ai/why-purposeful…

Official PyTorch code is released! github.com/apple/ml-tract

Twitter recommendation source code now available to all on GitHub github.com/twitter/the-al…

New paper TRACT - Faster diffusion model sampling - Single-step diffusion SotA for CIFAR10 and ImageNet64 with L2 loss without architecture changes - Up to 2.4x FID improvement arxiv.org/abs/2303.04248

TRACT: Denoising Diffusion Models with Transitive Closure Time-Distillation For single step diffusion, TRACT improves FID by up to 2.4× on the same architecture, and achieves new single-step Denoising Diffusion Implicit Models (DDIM) state-of-the-art FID (7.4 for ImageNet64, 3.8 for CIFAR10) abs: arxiv.org/abs/2303.04248

Thrilled to announce I will join Vladlen Koltun’s research organization at Apple! Time to go back to science again (after 2 months) and do impactful work!

We had a terrific TEROS event today! Delighted to meet roboticists across Texas. Such a vibrant research community! See you next year🤠 teros-texas.github.io

In 1994, Jeff Bezos famously spotted a stat that made him leave his high-paying PE job to start Amazon: 💡 The Internet was growing 2300% per year. What are the generation-defining stats of today? I'll post a few to kick things off...

@RaquelUrtasun Congrats @RaquelUrtasun

I got promoted to Full Professor at #UofT! This is dedicated to my team, which are the ones that make my success. So proud to work with you all!