Reproduce “MEMOR-E: In-Context and Fine-Tuned LLM Personalization for Alzheimer's Assistive Robotics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Maissa Abir Smaili, Eren Sadikoglu, Ransalu Senanayake
MEMOR-E is a privacy-preserving, mobile, and explainable robotic system designed to provide personalized, stage-adaptive cognitive assistance to individuals with Alzheimer's disease. The system combines local large language models (LLMs) with explainable AI (XAI) to generate non-diagnostic, interpretable cognitive summaries from speech, aiming to support daily care and enhance caregiver oversight.
Key insights
- Combining transformer-based classifiers with attribution-based XAI can effectively extract clinically relevant linguistic signals and provide transparent, interpretable cognitive insights from speech without exposing raw patient data.
- Large language models can be leveraged for behavioral simulation and to generate personalized, stage-adaptive assistive recommendations from structured cognitive summaries, even without medical-specific fine-tuning.
- Linguistic features such as filler disfluencies, pause markers, and pronoun usage are discriminative indicators of cognitive impairment, providing plausible explanations for the model's decision processes in Alzheimer's discourse.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.