FIMAP: Feature Importance by Minimal Adversarial Perturbation
Authors: Matt Chapman-Rounds, Umang Bhatt, Erik Pazos, Marc-Andre Schulz, Konstantinos Georgatzis11433-11441
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that this approach combines the two paradigms, recovering the output of feature-weighting methods in continuous feature spaces, whilst indicating the direction in which the nearest counterfactuals can be found. Our method provides an implicit confidence estimate in its own explanations, something existing methods lack. Additionally, FIMAP improves upon the speed of sampling-based methods, such as LIME, by an order of magnitude, allowing for explanation deployment in time-critical applications. We extend our approach to categorical features using a partitioned Gumbel layer and demonstrate its efficacy on standard datasets. Before concluding, we show empirical results on synthetic and real-world experiments. For all experiments below, expect otherwise stated, the neural network parameterising g(x; θg) consists of four fully connected layers of size 100 with Re LU nonlinearities... Results We first demonstrate that in simple continuous input spaces, FIMAP closely approximates LIME on standard a synthetic dataset, and succeeds in highlighting regions of interest in a manner unavailable to LIME. We trained a Random Forest with 200 trees to classify the half-moons dataset (with a train/test split of 8000/2000)... The classifier had an f-score of 0.97 on the test set. |
| Researcher Affiliation | Collaboration | Matt Chapman-Rounds1, Umang Bhatt2, Erik Pazos3, Marc-Andre Schulz4, Konstantinos Georgatzis3 1Department of Informatics, University of Edinburgh, 2Department of Engineering, University of Cambridge, 3Quantum Black, 4Department of Psychiatry and Psychotherapy, Charit e Universit atsmedizin Berlin |
| Pseudocode | No | The paper describes the methods and steps using mathematical formulations and textual descriptions but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any statement about releasing source code, nor does it include a link to a code repository for the methodology described. |
| Open Datasets | Yes | For simple synthetic data we used 10000 samples from the half moons dataset, available on scikit-learn (Pedregosa et al. 2011), with Gaussian noise with standard deviation 0.2 added to the data. A more realistic continuous dataset was MNIST (Le Cun et al. 1998), which we converted to a binary classification task by using only the digits 8 and 3... Finally, to test the performance of our method on a mix of categorical data and continuous data, we used a dataset available from the UCI machine learning repository (Dua and Graff 2017). |
| Dataset Splits | No | The paper mentions 'train/test split' for the datasets (e.g., '8000/2000' for half-moons, '11982/1984' for MNIST, and '17133/5711' for Adult dataset), but it does not specify a separate validation dataset split. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running the experiments, such as CPU or GPU models, memory, or cloud computing instance types. |
| Software Dependencies | No | The paper mentions 'scikit-learn' and 'Adam' but does not provide specific version numbers for these software components or any other libraries that would be necessary for reproduction. |
| Experiment Setup | Yes | For all experiments below, expect otherwise stated, the neural network parameterising g(x; θg) consists of four fully connected layers of size 100 with Re LU nonlinearities and a partial gumbel layer... We used a dropout percentage of 20 for every layer. The neural network parameterising the surrogate s(x; θs) consists of three fully connected layers of size 200, with the first two nonlinearities being Re LU, and the final Softmax. We used a cross-entropy loss, as is standard for classification, and trained both models using Adam (Kingma and Ba 2014), with a learning rate of 1e-3. We found (see discussion in Results, below), that for simple datasets setting λ to around an order of magnitude smaller than η yielded good results. (the values for the above perturbations were η = 2.0, λ = 0.1). |