Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

Learning to Route: Per-Sample Adaptive Routing for Multimodal Multitask Prediction

Authors: Marzieh Ajirak, Oded Bein, Ellen Bowen, Dora Kanellopoulos, Avital Falk, FAITH GUNNING, Nili Solomonov, Logan Grosenick

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the model on both synthetic data and real-world psychotherapy notes, predicting depression and anxiety outcomes. Our experiments show that our method consistently outperforms fixed multitask or single-task baselines, and that the learned routing policy provides interpretable insights into modality relevance and task structure.
Researcher Affiliation Academia 1Weill Cornell Medicine, Cornell University, NY, USA 2Feil Family Brain & Mind Research Institute, Cornell University, NY, USA
Pseudocode Yes Algorithm 1 Synthetic Data Generation 1: Sample x(num), x(text) N(0, I16) 2: Compute ϕ(x(text)), ψ(x(num)) via RFFs 3: Sample αk, βk, ωk, γk as above 4: Compute y1, y2 as above, add noise ϵk
Open Source Code Yes Open source code is available at: https://github.com/Grosenick-Lab-Cornell/learning-to-route.
Open Datasets No The clinical data included as one experiment is HIPAA-protected data and cannot be released. We will provide open access to our code and to our synthetic data in compliance with Neur IPS guidelines. We will provide scripts to reproduce our synthetic data results.
Dataset Splits Yes Number of samples: 1000 (train), 1000 (test)
Hardware Specification No The paper discusses computational efficiency and scaling in Appendix B and C, but does not provide specific hardware details such as GPU/CPU models or memory used for running the experiments.
Software Dependencies No The paper mentions several software components and models like MPNet, Sentence-BERT, BERT, Flan-T5, and Phi-2, but does not provide specific version numbers for any of these or other libraries used in the implementation.
Experiment Setup No The paper details the parameters for synthetic data generation in Section E, but it does not specify concrete experimental setup details for model training, such as hyperparameters (e.g., learning rate, batch size, number of epochs) or optimizer settings.