ICML 2025 paper proposing energy conservation metrics as critical diagnostics for machine learning interatomic potentials and introducing eSEN, a novel architecture designed to bridge the gap between test-set accuracy and real simulation performance on materials property prediction.
A NeurIPS 2021 method paper introducing Noise Contrastive Priors to address the VAE ‘prior hole’ problem, where standard Gaussian priors assign high density to regions of latent space that don’t correspond to realistic data, using energy-based models trained with contrastive learning to match the aggregate posterior.
Get a practical overview of the GEOM dataset and learn how it’s advancing 3D molecular machine learning by bridging static graphs and dynamic reality.
A provocative 2024 Nature Machine Intelligence paper challenging the assumption that invalid SMILES are failures, showing empirically that the ability to generate invalid outputs actually improves chemical language model performance by enabling quality filtering and providing richer training signals.
An end-to-end data factory for molecular machine learning that transforms raw chemical formulas (e.g., C6H14) into labeled 3D conformer datasets, using MAYGEN for structural isomer enumeration, RDKit for 3D embedding, and physics-based featurization to address data scarcity in computational drug discovery.
A complete guide to implementing modern Variational Autoencoders in PyTorch. Includes a copy-pasteable implementation, explanation of KL annealing to fix posterior collapse, and a deep dive into stable standard deviation parameterizations.
What happens when you achieve 99.8% accuracy on sarcasm detection? You might have accidentally built a domain classifier. A cautionary ML tale about dataset bias.
Can mathematical signatures capture molecular shape? We test whether Coulomb matrix eigenvalues can distinguish alkane constitutional isomers, from unsupervised clustering failures to supervised learning successes.
We test three ML models on 48K congressional bills to see how well they can predict policy areas from bill text. Results show logistic regression achieves 89% F1 score.
A practical introduction to Coulomb matrices: how they transform molecular 3D structures into ML features, complete with Python examples and honest assessment of their limitations.
Only 2% of congressional bills become law. We analyze 15K bills from 2021-2023 to understand what drives legislative success and failure.
A computational social science project that engineered a custom extraction engine to build a 47,000+ bill knowledge graph from Congress.gov (115th-117th Congresses), creating a novel legislative graph with co-sponsorship networks and establishing an 87% accuracy benchmark for policy area classification now available on Hugging Face.