Authors: Ravi Patel, Angus Brayne, Rogier Hintzen, Daniel Jaroslawicz, Georgiana Neculae, Dane Corneil

Published on: February 06, 2024

Impact Score: 8.2

Arxiv code: Arxiv:2402.04068

Summary

• What is new: Introduces Retrieve to Explain (R2E), a novel retrieval-based language model using Shapley values for evidence weighting.
• Why this is important: Difficulty in introspecting machine learning models, particularly language models, which can hide training issues and biases.
• What the research proposes: R2E model that uses evidence in documents to prioritize possible answers and uses Shapley values for evidence importance, adapting to new evidence without retraining.
• Results: Outperforms industry-standard genetics-based approaches in drug target identification from scientific literature on predicting clinical trial outcomes.

Technical Details

Technological frameworks used: Retrieve to Explain (R2E)

Models used: Retrieval-based language models

Data used: Published scientific literature

Potential Impact

Healthcare, pharmaceuticals, and biotech companies, especially in drug discovery and clinical research

Want to implement this idea in a business?

We have generated a startup concept here: Evidence-Based Insights.