NeurIPS 2025: Four Workshop Papers

🎉🎉🎉🎉 AIML@K contributes four workshop papers to NeurIPS 2025!

Kudos to Taehun, Jeung-un, Suhyun, and our external collaborators!

Best of Both Worlds: Bridging Laplace and Fourier for Generalizable and Efficient Operator Learning presents Laplace–Fourier Neural Operator (LFNO) that combines Laplace and Fourier operator ideas to model both transient and steady-state system dynamics, demonstrating superior generalization and efficiency on ODE/PDE tasks like the Duffing equation and Euler–Bernoulli beam.

Feature Learning as a Virtual Covariance Learning connects feature learning and SGD by interpreting “virtual updates” on hidden states, enabling direct weight construction aligned with covariance structure—achieving effective learning in 1–2 epochs vs. the typical 10–20.

Bridging data-rich and data-poor domains on Lithium-Ion Battery via Scanning Electron Microscopic data through Convolutional Neural Network Transfer Learning presents a two-stage transfer learning approach that enables accurate lithium-ion battery SEM image classification across domain shifts, achieving over 0.98 accuracy with limited data.

Numbers Already Carry Their Own Embeddings constructs Adelic Operation-preserved Embeddings (AOE), a training-free representation for numbers such that the embeddings preserve more operational properties than widely used tokenization methods, resulting in much improved AI4Math prediction task performances.