Self-explaining neural network with concept-based explanations for ICU mortality prediction

Kumar, Sayantan; Yu, Sean; Kannampallil, Thomas; Abrams, Zachary B.; Michelson, Andrew P.; Payne, Philip R. O.

doi:10.1145/3535508.3545547

Cited by 6 publications

(3 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This problem is only related to machine learning solutions, but in the field of NLP most of the solutions are based on ML. Unfortunately, current solutions, not only in the field of NLP, which additionally create explanations for taken actions, get worse results than those that do not create explanations [60]. Nevertheless, it is a very interesting direction of research, in terms of development of artificial intelligence itself and its understanding by humans.…”

Section: Hybrid Solutionsmentioning

confidence: 99%

Trends in Natural Language Understanding with sentence representation and sentiment analysis

Guminska,

Poniszewska-Maranda,

Ochelska-Mierzejewska

2023

Preprint

View full text Add to dashboard Cite

Natural Language Processing (NLP) is a crucial branch of artificial intelligence that examines how people and computers communicate through natural language. The majority of NLP research focuses on the fundamental methods that enable humans to understand and generate words, phrases, sentences, and the whole documents. The primary focus of the studies in this area concentrate on English as the common language used by scientists. Unfortunately, this language has a very straightforward structure. It indicates that most of the techniques that are very successful for documents in English do not work equally well for texts in other niche languages. To investigate the existing studies on Natural Language Understanding (NLU), a Systematic Literature Review (SLR) was performed. Through our deep research on NLU and its extensions, 46 studies were selected mentioning 38 machine learning, 3 hybrid and 4 knowledge-based approaches to NLU. The paper presents the systematized knowledge along with a critical analysis of existing research, both for text processing in English and for other native languages. It shows the solutions and reasoning of researchers, and compiles the advantages and disadvantages of the presented solutions.

show abstract

Section: Hybrid Solutionsmentioning

confidence: 99%

Trends in Natural Language Understanding with sentence representation and sentiment analysis

Guminska,

Poniszewska-Maranda,

Ochelska-Mierzejewska

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Real-time feature attribution methods, like those described in [8], [20], [23], [13], [14], [31], [32], [33], require just one iteration to generate an explanation. This is typically accomplished by training a feature selector model.…”

Section: Other Related Workmentioning

confidence: 99%

MAUSPAD: Mouse-Based Authentication Using Segmentation-Based, Progress-Adjusted DTW

Qin

Shen

Amariucai

et al. 2020

2020 IEEE 19th International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom)

View full text Add to dashboard Cite

The limited transparency of the inner decisionmaking mechanism in deep neural networks (DNN) and other machine learning (ML) models has hindered their application in several domains. In order to tackle this issue, feature attribution methods have been developed to identify the crucial features that heavily influence decisions made by these black box models. However, many feature attribution methods have inherent downsides. For example, one category of feature attribution methods suffers from the artifacts problem, which feeds out-of-distribution masked inputs directly through the classifier that was originally trained on natural data points. Another category of feature attribution method finds explanations by using jointly trained feature selectors and predictors. While avoiding the artifacts problem, this new category suffers from the Encoding Prediction in the Explanation (EPITE) problem, in which the predictor's decisions rely not on the features, but on the masks that selects those features. As a result, the credibility of attribution results is undermined by these downsides. In this research, we introduce the Double-sided Remove and Reconstruct (DoRaR) feature attribution method based on several improvement methods that addresses these issues. By conducting thorough testing on MNIST, CIFAR10 and our own synthetic dataset, we demonstrate that the DoRaR feature attribution method can effectively bypass the above issues and can aid in training a feature selector that outperforms other state-of-the-art feature attribution methods. Our code is available at https://github.com/dxq21/DoRaR.

show abstract

“…In this paper, we prefer not to try to explain the decision system as a whole but to apply an appropriate explainability method to the extracted features and possibly apply another method to the inference subsystem. We can observe two basic approaches to the explainability of recognition systems in the literature: the first approach seeks to preserve explainability by defeature (self-explaining systems, explainability by defeature) [12]- [14]. In it, we replace a part of the network (e.g., a classifier) with an explainable system (e.g., a linear system, a logic function).…”

Section: Introductionmentioning

confidence: 99%

Explainability as a Method for Learning From Computers

Klimo¹,

Kopcan²,

Králik³

2023

IEEE Access

View full text Add to dashboard Cite

Humans have rich experience applying linear models and logical thinking, but only experts understand the behaviour of non-linear systems. However, the deep neural network (DNN) implementation of non-linear systems outperforms optimal linear models. Therefore, the forward DNN (the pattern recognition system in this paper) attracts attention to the necessity of interpreting the results obtained by DNN. To preserve the high performance of DNN, we focus on a post hoc explanation; this approach means building an explainable model for the decision obtained by the black box. To avoid the interpretation of a set of millions of non-linear functions, we divide DNN into two parts: the feature extractor and the classifier. Following that, we argue for a specific interpretation of each of them. While for classifiers, we have several suitable explainable models (and we decided on the fuzzy logical function), we believe that feature interpretation is a creative scientific activity corresponding to the usual research. The paper presents a tool to help researchers and users understand extracted features not necessarily known in the specific application domain. Explanation of the new features is the way of learning from computers.

show abstract

Self-explaining neural network with concept-based explanations for ICU mortality prediction

Cited by 6 publications

References 16 publications

Trends in Natural Language Understanding with sentence representation and sentiment analysis

Trends in Natural Language Understanding with sentence representation and sentiment analysis

MAUSPAD: Mouse-Based Authentication Using Segmentation-Based, Progress-Adjusted DTW

Explainability as a Method for Learning From Computers

Contact Info

Product

Resources

About