ICML 2025: One Main Track Paper and Four Workshop Papers Accepted

🎉🎉🎉🎉🎉 AIML@K members are presenting one main track paper and four workshop papers in ICML 2025! Congratulations!

Broadband Ground Motion Synthesis by Diffusion Model with Minimal Condition presents a professional-level generative model for earthquake shockwave synthesis. This work is going to be presented as a main-track poster, between 11AM and 1:30PM PDT on July 16, 2025.

Catalyst: Structured Pruning with Robust Bifurcation Dynamics proposes a mathematically principled structured pruning regularizer with provable bifurcation dynamics. This work is going to be presented in 3rd Workshop on High-dimensional Learning Dynamics (HiLD) at ICML 2025.

Emergent Linear Separability of Unseen Data Points in High-dimensional Last-Layer Feature Space reports an emergent linearity in last layer feature space of neural networks. This work is also going to be presented in 3rd Workshop on High-dimensional Learning Dynamics (HiLD) at ICML 2025.

Enhancing Generative Seismic Modeling via Proposed Paired Dataset Construction Method presents how public seismological waveform data can be turned into a novel set of earth observation data. This work is going to be presented in TerraBytes Workshop at ICML 2025.

Broadband Ground Motion Synthesis by Diffusion Model with Minimal Condition emphasizes the waveform generative model element of the same titled ICML 2025 main track paper, which bested state-of-the-art time series generative models. This work is going to be presented in ICML 2025 Workshop on Machine Learning for Audio.