BrainMAP: Learning Multiple Activation Pathways in Brain Networks
Authors: Song Wang, Zhenyu Lei, Zhen Tan, Jiaqi Ding, Xinyu Zhao, Yushun Dong, Guorong Wu, Tianlong Chen, Chen Chen, Aiying Zhang, Jundong Li
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our comprehensive experiments highlight Brain MAP s superior performance. Furthermore, our framework enables explanatory analyses of crucial brain regions involved in tasks. 5 Experiments In this section, we aim to answer the following research questions (RQs). RQ1. How well can Brain MAP perform on brain-related tasks compared to other alternatives? RQ2. How does each component contribute to the overall predictive performance? RQ3. How effectively can Brain MAP elucidate the rationale behind its predictive outcomes? RQ4. What impact does the design of Mo E have on performance? 5.2 Main Results To answer RQ1, we first evaluate the performance of Brain MAP in comparison to all baselines on the HCP datasets. Table 2: Performance comparison of different models across various datasets. 5.3 Ablation Studies To address RQ2, we conduct ablation studies on Brain MAP by removing different components. |
| Researcher Affiliation | Academia | 1University of Virginia 2Arizona State University 3University of North Carolina at Chapel Hill 4Florida State University EMAIL, EMAIL, EMAIL, EMAIL, guorong EMAIL |
| Pseudocode | No | The paper describes methodologies and mathematical formulations but does not contain a clearly labeled pseudocode or algorithm block. |
| Open Source Code | Yes | Code https://github.com/Lzy Fischer/Brain MAP |
| Open Datasets | Yes | In our experiments, we consider the Human Connectome Project (HCP) dataset (Van Essen et al. 2013), which is a comprehensive publicly available neuroimaging dataset that includes both imaging data and a wide range of behavioral and cognitive data. We process the HCP-Task dataset by parcellating it into 360 distinct brain regions. With respect to other datasets, we use the processed ones from the Neuro Graph benchmark (Said et al. 2023). |
| Dataset Splits | No | The paper mentions using datasets from the Human Connectome Project (HCP) and Neuro Graph benchmark but does not explicitly provide information regarding training, validation, and test splits (e.g., percentages, sample counts, or specific split files). Table 1 provides dataset statistics but no split information. The caption for Table 2 states "All experiments are repeated with 3 different random seeds" but this does not specify the dataset splits. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments. |
| Software Dependencies | No | For the sequential model in our framework, we utilize the Mamba (Gu and Dao 2023), which is particularly effective in capturing long-range dependencies. The implementation details are explained in Appendix C (in the Arxiv version of this work). |
| Experiment Setup | No | The main text lacks specific hyperparameters (e.g., learning rate, batch size, number of epochs) or system-level training settings. It states, "For the training of gating functions and experts in Brain MAP, we adopt the cross-entropy (CE) loss for classification and the mean absolute error (MAE) for regression tasks," and "The implementation details are explained in Appendix C (in the Arxiv version of this work)," which defers the full setup details. |