My published work and preprints. For citation metrics, see Google Scholar.
Optimizing Sequence Models for Dynamical Systems
Ablation study deconstructing sequence models. Attention-augmented Recurrent Highway Networks outperform Transformers on …...
Research
Publications and preprints on natural language processing, document understanding, and computational social science.
LLMs for Page Stream Segmentation
Enhanced TabMe benchmark for page stream segmentation, creating TabMe++, showing fine-tuned decoder-based LLMs …...
LLMs for Insurance Document Automation
LLM applications for insurance document automation using parameter-efficient fine-tuning and analysis of calibration …...
Analytical Solution to Word2Vec Softmax & Corpus Bias Probing
Analytical derivation of Word2Vec's softmax objective factorization and a new framework for detecting semantic bias in …...
EigenNoise: Data-Free Word Vector Initialization
Investigation into EigenNoise, a data-free initialization scheme for word vectors that approaches pre-trained model …...
Look, Don't Tweet: Unified Data Models for Social NLP
PyConversations library and unified data schema for normalizing 300M+ posts across Twitter, Reddit, Facebook, and 4chan....
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 …...
NewsTweet Dataset: Social Media in Digital Journalism
NewsTweet dataset for studying embedded tweets in online journalism. Analysis shows 13% of Google News stories contain …...
Coordinated Social Targeting on Twitter
Investigation into follower dynamics on high-profile Twitter accounts, documenting sub-second spikes, saw-tooth …...
Data-Driven WordNet Construction from Wiktionary
We introduce an unsupervised algorithm for inducing semantic networks from noisy, crowd-sourced data, producing a …...