Addressable Memory for Video World Models
Paper • 2608.07408 • Published • 15
Jonathan Lorraine
lorraine2
AI & ML interests
Senior Research Scientist at NVIDIA Research (LPR). Agentic AI for science and engineering, world models, generative models, and scalable nested optimization.
Recent Activity
authored a paper 4 days ago
Addressable Memory for Video World Models posted an update 4 days ago
🧠 New NVIDIA paper: WorldTrace 🧠
Long-horizon video world models can store prior scenes in their KV cache and still fail to retrieve them once RoPE offsets leave the training range. We show this is an addressability problem, not just a compression problem.
WorldTrace assigns compressed memory slots fixed, in-distribution positions. WT-Field improves temporal consistency by +15.5%, while WT-Landmark improves episodic recall by +19.5% on LoopBench, without retraining.
🏆 Best Paper + Oral, ICML 2026 From Frames to Stories Workshop
🔎 Project page: https://research.nvidia.com/labs/sil/projects/WorldTrace/
📄 Paper: https://huggingface.co/papers/2608.07408
Led by Xindi Wu, with Sven Elflein, James Lucas, Olga Russakovsky, Laura Leal-Taixé, Despoina Paschalidou, Aljoša Ošep, and me. updated a dataset 7 days ago
lorraine2/fusion-eq-pred-dev1Organizations
authored a
paper 4 days ago
posted an update 4 days ago
Post
1724
🧠 New NVIDIA paper: WorldTrace 🧠
Long-horizon video world models can store prior scenes in their KV cache and still fail to retrieve them once RoPE offsets leave the training range. We show this is an addressability problem, not just a compression problem.
WorldTrace assigns compressed memory slots fixed, in-distribution positions. WT-Field improves temporal consistency by +15.5%, while WT-Landmark improves episodic recall by +19.5% on LoopBench, without retraining.
🏆 Best Paper + Oral, ICML 2026 From Frames to Stories Workshop
🔎 Project page: https://research.nvidia.com/labs/sil/projects/WorldTrace/
📄 Paper: Addressable Memory for Video World Models (2608.07408)
Led by Xindi Wu, with Sven Elflein, James Lucas, Olga Russakovsky, Laura Leal-Taixé, Despoina Paschalidou, Aljoša Ošep, and me.
Long-horizon video world models can store prior scenes in their KV cache and still fail to retrieve them once RoPE offsets leave the training range. We show this is an addressability problem, not just a compression problem.
WorldTrace assigns compressed memory slots fixed, in-distribution positions. WT-Field improves temporal consistency by +15.5%, while WT-Landmark improves episodic recall by +19.5% on LoopBench, without retraining.
🏆 Best Paper + Oral, ICML 2026 From Frames to Stories Workshop
🔎 Project page: https://research.nvidia.com/labs/sil/projects/WorldTrace/
📄 Paper: Addressable Memory for Video World Models (2608.07408)
Led by Xindi Wu, with Sven Elflein, James Lucas, Olga Russakovsky, Laura Leal-Taixé, Despoina Paschalidou, Aljoša Ošep, and me.
updated a
dataset 7 days ago
published a
dataset 7 days ago
updated a
collection 8 days ago
upvoted a paper 8 days ago
authored 3
papers 3 months ago
JacNet: Learning Functions with Structured Jacobians
Paper • 2408.13237 • Published • 1
Variance Reduction for Expectations with Diffusion Teachers
Paper • 2605.21489 • Published • 1
NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle Simulation
Paper • 2606.03159 • Published • 23
upvoted a paper 3 months ago
updated a
collection 3 months ago
upvoted a paper 3 months ago
liked a
model 3 months ago
upvoted a collection 3 months ago
updated a
collection 3 months ago
upvoted a paper 3 months ago
updated a
collection 3 months ago