DenoiseVAE: Learning Molecule-Adaptive Noise Distributions for Denoising-based 3D Molecular Pre-training
Liu et al.'s ICLR 2025 paper introducing DenoiseVAE, which learns adaptive, atom-specific noise for better molecular …...
Methods
Personal notes and analysis on methods in machine learning, generative modeling, and scientific computing.
Beyond Atoms: Enhancing Molecular Pretrained Representations with 3D Space Modeling
Lu et al. introduce SpaceFormer, a Transformer that models entire 3D molecular space—not just atoms—for superior …...
Efficient and Scalable Density Functional Theory Hamiltonian Prediction through Adaptive Sparsity
Luo et al. introduce SPHNet, using adaptive sparsity to dramatically improve SE(3)-equivariant Hamiltonian prediction …...
Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction
Fu et al. propose energy conservation as a key MLIP diagnostic and introduce eSEN, bridging test accuracy and real …...
The Dark Side of the Forces: Assessing Non-Conservative Force Models for Atomistic Machine Learning
Bigi et al. critique non-conservative force models in ML potentials, showing their simulation failures and proposing …...
Embedded-atom method: Derivation and application to impurities, surfaces, and other defects in metals
Daw and Baskes's foundational 1984 paper introducing the Embedded-Atom Method (EAM), a many-body potential for metal …...
Nonphysical Sampling Distributions in Monte Carlo Free-Energy Estimation: Umbrella Sampling
Torrie and Valleau's 1977 paper introducing Umbrella Sampling, an importance sampling technique for Monte Carlo …...
A Contrastive Learning Approach for Training Variational Autoencoder Priors
Dai et al.'s NeurIPS 2021 paper introducing Noise Contrastive Priors (NCPs) to address VAE's 'prior hole' problem with …...
Processes of Adsorption and Diffusion on Solid Surfaces
Lennard-Jones's 1932 foundational paper introducing potential energy surface models to unify physical and chemical …...
3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data
Weiler et al.'s NeurIPS 2018 paper introducing SE(3)-equivariant CNNs for volumetric data using group theory and …...