Optimizing Sequence Models for Dynamical Systems
Ablation study deconstructing sequence models. Attention-augmented Recurrent Highway Networks outperform Transformers on …
Evaluation
String Representations for Chemical Image Recognition
Ablation study comparing SMILES, DeepSMILES, SELFIES, and InChI for OCSR. SMILES achieves highest accuracy; SELFIES …
Imago: Structure Recognition at TREC-CHEM 2011
Open-source C++ toolkit for extracting 2D chemical structures from scientific literature using heuristic image …
MolRec: Chemical Structure Recognition at CLEF 2012
MolRec achieves 95%+ accuracy on simple structures but struggles with complex diagrams, revealing rule-based OCSR limits …
MolRec: Rule-Based OCSR System
Rule-based system for optical chemical structure recognition using vectorization and geometric analysis, achieving 95% …
Modernizing Rahman''s 1964 Argon Simulation
A high-fidelity replication of foundational molecular dynamics using modern software engineering practices: caching, …
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 …
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 …
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 …
Classifying Congressional Bills with Machine Learning
Testing ML classification of congressional bills by policy area. Comparing Naive Bayes, Logistic Regression, and XGBoost …
How Does Congress Actually Work? Data from 15K Bills
What happens to bills in Congress? Analyzing 15K+ bills from the 117th Congress to understand legislative patterns, …
Analytical Solution to Word2Vec Softmax & Bias Probing
Analytical derivation of Word2Vec's softmax objective factorization and a new framework for detecting semantic bias in …