Explore alternative energy-function architectures

Explore alternative architectures for the learnable energy function in Energy-Mamba and determine whether they improve the compatibility constraint between evolving hidden states and static local image features.

Background

Energy-Mamba currently uses a lightweight multilayer perceptron as its learnable potential-energy function. The paper explicitly leaves the exploration of alternative energy-function architectures open, indicating that the influence of the energy-function design on representational stability and classification performance has not been systematically resolved.

References

Several directions remain open for future investigation, including the extension of Energy-Mamba to 3D volumetric medical data, the extension to other tasks such as segmentation, the exploration of alternative energy function architectures, and a systematic ablation of the interplay between the kinetic and potential branches across different imaging modalities and dataset scales.

— Energy-Mamba: A Physics-Constrained State-Space Model for Medical Image Classification  (2608.19813 - Mabrok et al., 20 Aug 2026) in Conclusion