Scientific Computing
Velocity Autocorrelation Function showing the signature negative region characteristic of liquid dynamics and the cage effect discovered by Rahman

Rahman's 1964 Liquid Argon Simulation Replication

Replication of Rahman's 1964 molecular dynamics simulation using modern LAMMPS and Python, achieving quantitative …...

Scientific Computing
Velocity Autocorrelation Function showing the signature negative region characteristic of liquid dynamics and the cage effect discovered by Rahman

Replicating Rahman's 1964 Liquid Argon Simulation

Recreating the foundational 1964 molecular dynamics simulation with modern Python and LAMMPS, testing whether Rahman's …...

Time Series Forecasting

Deconstructing Neural Networks for Time Series Forecasting

Ablation study of neural network components for forecasting, finding gating and attention improve RNNs while recurrence …...

Computational Chemistry
Comparison chart showing k-NN significantly outperforming logistic regression for molecular classification across different alkane sizes

Can You Hear the Shape of a Molecule? (Part Three)

Supervised learning reveals hidden eigenvalue patterns that clustering missed, testing k-NN and logistic regression on …

Computational Chemistry
Charts showing Dunn Index, distance metrics, and computation time analysis revealing clustering performance degradation with molecular size

Can You Hear the Shape of a Molecule? (Part Two)

Clustering analysis reveals why Coulomb matrix eigenvalues struggle with larger alkanes, using Dunn Index and silhouette …

Natural Language Processing
Word vector illustration showing text classification and NLP concepts

Sarcasm Detection with Transformers: A Cautionary Tale

Learn how dataset bias can lead to misleading results in NLP: a sarcasm detection model that actually learned to …

Natural Language Processing

Analytical Model of Word2Vec and GloVe Statistics

Analytical model of Word2Vec and GloVe statistics. First analytical solution to Word2Vec's softmax skip-gram with bias …...

AI Safety

GPT-2 Susceptibility to Universal Adversarial Triggers

Investigation into whether universal adversarial triggers can control both topic and stance of GPT-2's generated text …...

Natural Language Processing
Types and distribution of coreferences in QuAC dataset showing dialogue complexity

QuAC: Question Answering in Context Dataset

Analysis of QuAC's conversational QA through student-teacher interactions, featuring 100K+ context-dependent questions …