Misuse index for explainable artificial intelligence in computing environments
A mechanism is described for facilitating misuse index for explainable artificial intelligence in computing environments, according to one embodiment. A method of embodiments, as described herein, includes mapping training data with inference uses in a machine learning environment, where the training data is used for training a machine learning model. The method may further include detecting, based on one or more policy/parameter thresholds, one or more discrepancies between the training data and the inference uses, classifying the one or more discrepancies as one or more misuses, and creating a misuse index listing the one or more misuses.
1 . A method, comprising:
obtaining, based on a policy for a neural network, a threshold value for detecting whether the neural network is misused;
adjusting the threshold value based on a change associated with the policy;
determining a misuse index value from training data and inference data of the neural network, the training data used to train the neural network, the inference data generated from inference of the neural network; and
modifying the neural network based on a comparison of the adjusted threshold value and the misuse index value.
2 . The method of claim 1 , wherein the change associated with the policy is a change from the policy to a different policy.
3 . The method of claim 1 , wherein the change associated with the policy is a change to the training data or the inference data.
4 . The method of claim 1 , wherein determining the misuse index value from the training data and inference data of the neural network comprises:
comparing the training data and inference data; and
determining the misuse index value based on a result of comparing the training data and inference data.
5 . The method of claim 4 , wherein comparing the training data and inference data comprises:
comparing a range of the training data with a range of the inference data; or
comparing a setting under which the training data is generated with a setting under which the inference data is generated.
6 . The method of claim 4 , wherein the training data includes one or more training images, the inference data includes one or more inference images, and comparing the training data and inference data comprises:
comparing a foreground or background attribute of the one or more training images with a foreground or background attribute of the one or more inference images.
7 . The method of claim 1 , wherein determining the misuse index value from the training data and inference data of the neural network comprises:
determining the misuse index value by mapping an attribution of at least one neuron associated with the training data with an attribution of at least one neuron associated with the inference data.
8 . The method of claim 1 , wherein modifying the neural network comprises:
augmenting the training data; and
further training the neural network using the augmented training data.
9 . The method of claim 1 , wherein modifying the neural network comprises:
further training the neural network by performing a biased training process using a labeled output signal.
10 . The method of claim 1 , wherein modifying the neural network comprises:
modifying one or more inference parameters of the neural network to align the one or more inference parameters with the change associated with the policy.
11 . One or more non-transitory computer-readable media storing instructions executable to perform operations, the operations comprising:
obtaining, based on a policy for a neural network, a threshold value for detecting whether the neural network is misused;
adjusting the threshold value based on a change associated with the policy;
determining a misuse index value from training data and inference data of the neural network, the training data used to train the neural network, the inference data generated from inference of the neural network; and
modifying the neural network based on a comparison of the adjusted threshold value and the misuse index value.
12 . The one or more non-transitory computer-readable media of claim 11 , wherein the change associated with the policy is a change from the policy to a different policy.
13 . The one or more non-transitory computer-readable media of claim 11 , wherein the change associated with the policy is a change to the training data or the inference data.
14 . The one or more non-transitory computer-readable media of claim 11 , wherein determining the misuse index value from the training data and inference data of the neural network comprises:
comparing the training data and inference data; and
determining the misuse index value based on a result of comparing the training data and inference data.
15 . The one or more non-transitory computer-readable media of claim 11 , wherein determining the misuse index value from the training data and inference data of the neural network comprises:
determining the misuse index value by mapping an attribution of at least one neuron associated with the training data with an attribution of at least one neuron associated with the inference data.
16 . The one or more non-transitory computer-readable media of claim 11 , wherein modifying the neural network comprises:
augmenting the training data; and
further training the neural network using the augmented training data.
17 . The one or more non-transitory computer-readable media of claim 11 , wherein modifying the neural network comprises:
further training the neural network by performing a biased training process using a labeled output signal.
18 . The one or more non-transitory computer-readable media of claim 11 , wherein modifying the neural network comprises:
modifying one or more inference parameters of the neural network to align the one or more inference parameters with the change associated with the policy.
19 . An apparatus, comprising:
a computer processor for executing computer program instructions; and
a non-transitory computer-readable memory storing computer program instructions executable by the computer processor to perform operations comprising:
obtaining, based on a policy for a neural network, a threshold value for detecting whether the neural network is misused,
adjusting the threshold value based on a change associated with the policy,
determining a misuse index value from training data and inference data of the neural network, the training data used to train the neural network, the inference data generated from inference of the neural network, and
modifying the neural network based on a comparison of the adjusted threshold value and the misuse index value.
20 . The apparatus of claim 19 , wherein modifying the neural network comprises:
further training the neural network or modifying one or more inference parameters of the neural network.