Research map / Large language models / Retrieval and knowledge
Enhancing Predictive Model Learning via Domain-Knowledge Augmented Latent Feature Mining
Bingxuan Li, Pengyi Shi, Amy R Ward
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add FLAME-inferred latent features to gradient boosted trees | GBT classifier · Incarceration diversion outcome prediction | ROC AUC: 84% -> 92% | Improved |
| add FLAME-inferred latent features to logistic regression | LR classifier · Incarceration diversion outcome prediction | ROC AUC: 70% -> 85% | Improved |
| add FLAME-inferred latent features to MLP | MLP classifier · Incarceration diversion outcome prediction | ROC AUC: 81% -> 88% | Improved |
| add FLAME-inferred latent features to random forests | RF classifier · Incarceration diversion outcome prediction | F1: 70% -> 75% | Improved |
| add FLAME-inferred social support feature to logistic regression | LR classifier · MIMIC discharge location prediction | ROC AUC: 65.22% -> 71.22% | Improved |
| add FLAME-inferred social support feature to MLP | MLP classifier · MIMIC discharge location prediction | ROC AUC: 63.19% -> 74.40% | Improved |
| add FLAME-inferred social support feature to GBT | GBT classifier · MIMIC discharge location prediction | F1: 65.09% -> 75.38% | Improved |
| use fine-tuned GPT-3.5 (FLAME g) instead of tabular ML models to infer risk level | latent feature inference · Risk level inference | accuracy: 58.24% -> 77.01% | Improved |
| use LLaMA2 instead of GPT-3.5 as FLAME base model | FLAME latent feature inference · Risk level inference | accuracy: 77.01% -> 63.08% | Hurt |
| add expert-provided socio-economic zip code information to the prompt (ablation) | FLAME latent feature extraction · Incarceration diversion outcome prediction | feature extraction quality | Improved |
| replace external socio-economic data with GPT-4 internal knowledge of zip codes (ablation) | FLAME domain knowledge · 50 zip codes | accuracy of generated context: 33 of 50 accurate, 17 hallucinated, 5 undetermined | Hurt |
Closest papers by meaning
- Why Struggle with Continuous Latents? Interpretable Discrete Latent Reasoning via Rendered Compression (NeurIPS 2026 · similarity 0.83)
- Latent Concept Disentanglement in Transformer-based Language Models (ICLR 2026 · similarity 0.83)
- Concept Bottleneck Large Language Models (ICLR 2025 · similarity 0.83)
- Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space (NeurIPS 2026 · similarity 0.83)
- Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoning (ICML 2025 · similarity 0.83)
- LLM Processes: Numerical Predictive Distributions Conditioned on Natural Language (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
