![[2505.19193] SuperMAN: Interpretable and Expressive Networks over Temporally Sparse Heterogeneous Data](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2505.19193] SuperMAN: Interpretable and Expressive Networks over Temporally Sparse Heterogeneous Data
Summary
The paper presents SuperMAN, a framework designed for learning from temporally sparse and heterogeneous data, enhancing interpretability and expressivity in machine learning applications.
Why It Matters
SuperMAN addresses challenges in analyzing irregular temporal data, which is common in fields like healthcare. Its ability to provide interpretable insights can significantly impact decision-making in high-stakes environments, such as predicting disease onset and understanding complex data relationships.
Key Takeaways
- SuperMAN offers a novel approach to learning from temporally sparse heterogeneous data.
- The framework enhances interpretability at multiple levels, aiding in practical applications.
- SuperMAN achieves state-of-the-art performance in critical tasks like disease prediction.
- The model reveals insights into disease development phases, benefiting healthcare practitioners.
- Flexibility in interpretability allows users to balance expressivity and understanding based on domain needs.
Computer Science > Machine Learning arXiv:2505.19193 (cs) [Submitted on 25 May 2025 (v1), last revised 23 Feb 2026 (this version, v3)] Title:SuperMAN: Interpretable and Expressive Networks over Temporally Sparse Heterogeneous Data Authors:Andrea Zerio, Maya Bechler-Speicher, Maor Huri, Marie Vibeke Vestergaard, Ran Gilad-Bachrach, Tine Jess, Samir Bhatt, Aleksejs Sazonovs View a PDF of the paper titled SuperMAN: Interpretable and Expressive Networks over Temporally Sparse Heterogeneous Data, by Andrea Zerio and 7 other authors View PDF HTML (experimental) Abstract:Real-world temporal data often consists of multiple signal types recorded at irregular, asynchronous intervals. For instance, in the medical domain, different types of blood tests can be measured at different times and frequencies, resulting in fragmented and unevenly scattered temporal data. Similar issues of irregular sampling occur in other domains, such as the monitoring of large systems using event log files. Effectively learning from such data requires handling sets of temporal sparse and heterogeneous signals. In this work, we propose Super Mixing Additive Networks (SuperMAN), a novel and interpretable-by-design framework for learning directly from such heterogeneous signals, by modeling them as sets of implicit graphs. SuperMAN provides diverse interpretability capabilities, including node-level, graph-level, and subset-level importance, and enables practitioners to trade finer-grained interpretability fo...
