Bi-MIChI

Bi-Capacity Choquet Integral for Sensor Fusion with Label Uncertainty

Hersh Vakharia and Xiaoxiao Du

Accepted to 2024 FUZZ-IEE, Presented at 2024 WCCI in Yokohama, Japan

[arXiv] [IEEEXplore]

Installation Prerequisites

This code was tested using Python 3.10.
Install requirements with pip3 install -r requirements.txt.

Demo

The demo reproduces the "UM" experiement from the paper via the jupyter notebook um_test.ipynb. Simply run the notebook to see the results.

Code Structure

Here is a brief summary of the code structure. For detailed usage, see the docustrings in the code, as well as the demo jupyter notebook.

bicap.py:

• Bicapacity (class): defines the structure and usage of a bicapacity
• BicapacityGenerator (class): random generation of new bicapacities
  • useZeroBound param determines whether bicapacity zero bound is enforced (Obj 1 vs Obj 2)
• choquet_integral (function): given data and a bicapacity, computes and returns the choquet integral.

bicap_train.py:

• BicapEvolutionaryTrain (class): Contains code for the optimization framework, given data, labels, and params
  • Params dictionary example:

    param = {
      "max_iter": 5000, # maximum optimization iterations
      "eta": 0.8, # small-scale mutation rate
      "pop_size": 8, # bicap population size
      "fitness_thresh": 0.001, # fitness threshold for stopping condition
      "use_zero_bound": True # enforce zero bound or not
    }

utils.py: Contains miscellaneous utility functions used throughout the rest of the code.

License

This source code is licensed under the license found in the LICENSE file in the root directory of this source tree.

This product is Copyright (c) 2024 H. Vakharia and X. Du. All rights reserved.

Citing Bi-MIChI

If you use the Bi-MIChI fusion framework, please cite the following reference using the following BibTeX entries.

@INPROCEEDINGS{10611865,
  author={Vakharia, Hersh and Du, Xiaoxiao},
  booktitle={2024 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE)}, 
  title={Bi-Capacity Choquet Integral for Sensor Fusion with Label Uncertainty}, 
  year={2024},
  volume={},
  number={},
  pages={1-10},
  keywords={Training;Uncertainty;Limiting;Soft sensors;Measurement uncertainty;Data integration;Object detection;bi-capacity;choquet integral;fuzzy measures;sensor fusion;label uncertainty;classification},
  doi={10.1109/FUZZ-IEEE60900.2024.10611865}}

Related Work

Multiple Instance Choquet Integral (MICI) [arXiv] [Code Repo]

Multiple Instance Multi-Resolution Fusion (MIMRF) [arXiv] [Code Repo]

MIMRF with Binary Fuzzy Measures (MIMRF-BFM) [arXiv] [Code Repo]