Jiwoo Kim

Selected Publications

(* indicates equal contribution)

1. 
   NNiT: Width-Agnostic Neural Network Generation with Structurally Aligned Weight Spaces

   Jiwoo Kim, Swarajh Mehta, Hao-Lun Hsu, Hyunwoo Ryu, Yudong Liu, Miroslav Pajic

   ICML, 2026

   arXiv

   We introduce NNiT, which generates MLP weights in a width-agnostic manner by tokenizing weight matrices into patches and modeling them as structured fields. NNiT generalizes to architecture topologies unseen during training.

2. 
   Diffusion-EDFs: Bi-equivariant Denoising Generative Modeling on SE(3) for Visual Robotic Manipulation

   Hyunwoo Ryu, Jiwoo Kim, Junwoo Chang, Hyun Seok Ahn, Taehan Kim, Yubin Kim, Joohwan Seo, Jongeun Choi, Roberto Horowitz

   CVPR, 2024 Highlight · top 11.9%

   project arXiv code

   We introduce SE(3)-equivariant diffusion generative modeling for visual robotic manipulation, achieving sample-efficient policy learning that generalizes across object poses and viewpoints.

3. 
   Denoising Heat-inspired Diffusion with Insulators for Collision-Free Motion Planning

   Junwoo Chang*, Hyunwoo Ryu*, Jiwoo Kim, Soochul Yoo, Joohwan Seo, Nikhil Prakash, Jongeun Choi, Roberto Horowitz

   NeurIPS Workshop on Diffusion Models, 2023

   project arXiv

   At inference time, we simultaneously generate only reachable goals and plan obstacle-free motions from a single visual input.

4. 
   Robotic Manipulation Learning with Equivariant Descriptor Fields: Generative Modeling, Bi-equivariance, Steerability, and Locality

   Jiwoo Kim*, Hyunwoo Ryu*, Jongeun Choi, Joohwan Seo, Nikhil Prakash, Ruolin Li, Roberto Horowitz

   RSS Workshop on Symmetries in Robot Learning, 2023 Oral · Best Paper

   OpenReview

   We introduce Equivariant Descriptor Fields (EDFs) and analyze four key properties: generative modeling, bi-equivariance, steerable representations, and locality.