Software

Other Projects: Github

• PyMetaMap
  Python interface for MetaMap, the biomedical named entity recognition and concept mapping system developed by the National Library of Medicine.
  [code]


• LEML
  Python implementation of a matrix factorization-based method for large-scale extreme multi-label classification.
  [code]


• pyClausIE
  Python interface for ClausIE, an open information extraction system for identifying propositions and relations from text.
  [code]


• advRelCNN
  Python implementation of an adversarial neural network for unsupervised domain adaptation in biomedical relation classification.
  [code]


• Multi-Label Zero-Shot Learning
  Neural methods for few-shot and zero-shot multi-label text classification when labels have a known structured relationship.
  [code]


• Med-Match CNN
  Matching-network-based convolutional neural network for few-shot assignment of medical diagnosis codes to electronic medical records.
  [code]


• Bio-CNN
  CNN framework for biomedical text classification with support for auxiliary wide features and ordinal classification losses.
  [code]


• Relation Extraction RNN
  Bidirectional recurrent neural network implementation for biomedical relation extraction, including drug-drug interaction extraction.
  [code]


• SMM4H
  Supervised NLP framework for detecting adverse drug reaction mentions and classifying medication-intake messages in social media.
  [code]


• DeepRel
  Deep neural relation extraction framework combining convolutional and recurrent neural network approaches for chemical-protein relation extraction.
  [code]


• Geographic Performance Disparities
  Data and analysis resources for measuring geographic disparities in the performance of offensive-language classification systems.
  [code/data]


• SDOH-N2C2
  Neural information extraction system for identifying social determinants of health, their arguments, and relationships in clinical notes.
  [code]


• RadiologySumm
  Multimodal framework for generating radiology report impressions from textual findings and chest X-ray information using retrieval and neural models.
  [code]


• CivilPromptReasoningGPT4
  GPT-4 prompting, chain-of-thought, retrieval, and prompt-ensembling framework for legal argument reasoning in civil procedures.
  [code]


• LLM-Cipher
  Framework for detecting machine-generated text across different LLM generators and domains using representations from pretrained language models.
  [code]


• BikeFrames
  Code and data for detecting how cyclists are portrayed in news coverage, including the BikeFrame Chain-of-Code framework for analyzing perception, accidents, and fault.
  [code/data]


• Text-to-SQL Schema Inference
  Framework for studying zero-knowledge schema inference attacks and information leakage in LLM-based text-to-SQL systems.
  [code]


• MA-ToT
  Multi-agent Tree-of-Thought reasoning framework that uses multiple reasoning agents together with a validator agent to identify and remove faulty reasoning paths.
  [code]


• Clinical LLM Knowledge Boundaries
  Probing framework for comparing the internal representations, knowledge boundaries, and robustness of clinical and general-purpose large language models.
  [code]


• Speculative Storytelling for Healthcare AI
  Human-centered framework that generates speculative user stories and supports multi-agent discussions to help people anticipate potential benefits and harms of healthcare AI systems.
  [code]


• Vision-to-Policy (V2P)
  Multimodal access-control framework that uses vision-language models to translate access-control diagrams into entities, relationships, and structured policy representations.
  [code]


• LLM Access-Control Datasets
  Access-control benchmark generation framework and datasets built from Spider and BIRD for evaluating whether LLMs correctly permit or deny database queries under role-based policies.
  [code/data]


• TRACER
  Framework for detecting task-oriented dialogue failures before a conversation ends using dialogue trajectories and language-model representations, with tools for failure recovery and evaluation.
  [code]


• BiomedicalRAG
  Large-scale evaluation framework for biomedical retrieval-augmented generation across multiple medical QA datasets, retrieval corpora, retrieval methods, and open-source LLMs.
  [code]


• MedProb
  Framework for probing the internal representations of vision-language models for medical visual question answering, including hidden-state extraction, linear probing, prompting, and fine-tuning experiments.
  [code]


• Prompting Underestimates
  Evaluation framework for comparing prompting and representation probing of large language and vision-language models on time-series classification tasks.
  [code]


• DoppelBot
  Experimental platform, analysis tools, and anonymized data for studying how middle-school students distinguish human collaborators from LLM-controlled agents in live interactions.
  [code/data]