Explaining short text classification with diverse synthetic exemplars and counter-exemplars
نویسندگان
چکیده
Abstract We present xspells , a model-agnostic local approach for explaining the decisions of black box models in classification short texts. The explanations provided consist set exemplar sentences and counter-exemplar sentences. former are examples classified by with same label as text to explain. latter different (a form counter-factuals). Both close meaning explain, both meaningful – albeit they synthetically generated. generates neighbors explain latent space using Variational Autoencoders encoding decoding instances. A decision tree is learned from randomly generated neighbors, used drive selection exemplars counter-exemplars. Moreover, diversity counter-exemplars modeled an optimization problem, solved greedy algorithm theoretical guarantee. report experiments on three datasets showing that outperforms well-known lime method terms quality explanations, fidelity, diversity, usefulness, comparable it stability.
منابع مشابه
Training Exemplars
Analysis of a biolgically motivated neural network for character recognition. learning for multi-layer feed-forward neural networks using the conjugate gradient method. of japanese kanji using principal component analysis as a preprocessor to an articial neural network.
متن کاملRequirements and Specification Exemplars
Page 12 meeting initiator. For instance, the meeting chair and secretary must be present; they have the highest participant importance. In a project meeting where specific tasks are discussed, task leaders would have a higher importance than normal project members and a lower importance than the meeting chair, the task speakers or the project reviewers. Meeting Significance. The meeting signifi...
متن کاملExtracting exemplars and prototypes
Next, we transform using the multinomial2logical function the multinomial predictors into dummy predictor variables , which are simply true or false for each categorical value of the selected multinomial variables. The think data frame has three extralinguistic variables that we have decided to exclude from the following statistical modeling, namely Register, Section, and Author (which are colu...
متن کاملDeblurring Face Images with Exemplars
The human face is one of the most interesting subjects involved in numerous applications. Significant progress has been made towards the image deblurring problem, however, existing generic deblurring methods are not able to achieve satisfying results on blurry face images. The success of the state-of-the-art image deblurring methods stems mainly from implicit or explicit restoration of salient ...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Machine Learning
سال: 2022
ISSN: ['0885-6125', '1573-0565']
DOI: https://doi.org/10.1007/s10994-022-06150-7