Explainable software analytics

Dam, Hoa Khanh; Tran, Truyen; Ghose, Aditya

doi:10.1145/3183399.3183424

Cited by 93 publications

(46 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is known that quantitative evaluation of the transparency of algorithms is challenging [30], and we cannot say our work solved the problem completely. However, deep neural inference process was black-box until now, and the experiments and discussion in this paper shows that our work moved it forward to transparency.…”

Section: Resultsmentioning

confidence: 93%

Improving transparency of deep neural inference process

2019

View full text Add to dashboard Cite

Deep learning techniques are rapidly advanced recently, and becoming a necessity component for widespread systems. However, the inference process of deep learning is black-box, and not very suitable to safety-critical systems which must exhibit high transparency. In this paper, to address this black-box limitation, we develop a simple analysis method which consists of 1) structural feature analysis: lists of the features contributing to inference process, 2) linguistic feature analysis: lists of the natural language labels describing the visual attributes for each feature contributing to inference process, and 3) consistency analysis: measuring consistency among input data, inference (label), and the result of our structural and linguistic feature analysis. Our analysis is simplified to reflect the actual inference process for high transparency, whereas it does not include any additional black-box mechanisms such as LSTM for highly human readable results. We conduct experiments and discuss the results of our analysis qualitatively and quantitatively, and come to believe that our work improves the transparency of neural networks. Evaluated through 12,800 human tasks, 75% workers answer that input data and result of our feature analysis are consistent, and 70% workers answer that inference (label) and result of our feature analysis are consistent. In addition to the evaluation of the pro-

show abstract

Section: Resultsmentioning

confidence: 93%

Improving transparency of deep neural inference process

2019

View full text Add to dashboard Cite

show abstract

“…Determination of which results to consider as the best to use, since there were different results given different algorithms, was also considered. While choosing the best algorithm to perform feature selection, the importance of the predictability power of the algorithm was conceded with the ease of algorithm explain-ability [20]. While both are important, the researcher had to choose which would take precedence.…”

Section: Resultsmentioning

confidence: 99%

Using Feature Selection Methods to Discover Common Users’ Preferences for Online Recommender Systems

Ndung'u

Kamau

Mariga

2021

IJCIT

View full text Add to dashboard Cite

Recommender systems have taken over user’s choice to choose the items/services they want from online markets, where lots of merchandise is traded. Collaborative filtering-based recommender systems uses user opinions and preferences. Determination of commonly used attributes that influence preferences used for prediction and subsequent recommendation of unknown or new items to users is a significant objective while developing recommender engines. In conventional systems, study of user behavior to know their dis/like over items would be carried-out. In this paper, presents feature selection methods to mine such preferences through selection of high influencing attributes of the items. In machine learning, feature selection is used as a data pre-processing method but extended its use on this work to achieve two objectives; removal of redundant, uninformative features and for selecting formative, relevant features based on the response variable. The latter objective, was suggested to identify and determine the frequent and shared features that would be preferred mostly by marketplace online users as they express their preferences. The dataset used for experimentation and determination was synthetic dataset. The Jupyter Notebook™ using python was used to run the experiments. Results showed that given a number of formative features, there were those selected, with high influence to the response variable. Evidence showed that different feature selection methods resulted with different feature scores, and intrinsic method had the best overall results with 85% model accuracy. Selected features were used as frequently preferred attributes that influence users’ preferences.

show abstract

“…Interpretability is important for software mining and analysis in general [13]. In other domains, various techniques have been proposed to interpret machine learning results, such as by projecting outputs of CNN models through hidden neurons to input image pixels [14], by quantifying the effects of different compositions of English sentences on NLP models [15], and by perturbing inputs for black-box neural networks [16].…”

Section: Related Workmentioning

confidence: 99%

AutoFocus: Interpreting Attention-Based Neural Networks by Code Perturbation

Bui

Jiang

2019

2019 34th IEEE/ACM International Conference on Automated Software Engineering (ASE)

View full text Add to dashboard Cite

Despite being adopted in software engineering tasks, deep neural networks are treated mostly as a black box due to the difficulty in interpreting how the networks infer the outputs from the inputs. To address this problem, we propose AutoFocus, an automated approach for rating and visualizing the importance of input elements based on their effects on the outputs of the networks. The approach is built on our hypotheses that (1) attention mechanisms incorporated into neural networks can generate discriminative scores for various input elements and (2) the discriminative scores reflect the effects of input elements on the outputs of the networks. This paper verifies the hypotheses by applying AutoFocus on the task of algorithm classification (i.e., given a program source code as input, determine the algorithm implemented by the program). AutoFocus identifies and perturbs code elements in a program systematically, and quantifies the effects of the perturbed elements on the network's classification results. Based on evaluation on more than 1000 programs for 10 different sorting algorithms, we observe that the attention scores are highly correlated to the effects of the perturbed code elements. Such a correlation provides a strong basis for the uses of attention scores to interpret the relations between code elements and the algorithm classification results of a neural network, and we believe that visualizing code elements in an input program ranked according to their attention scores can facilitate faster program comprehension with reduced code.

show abstract

Explainable software analytics

Cited by 93 publications

References 23 publications

Improving transparency of deep neural inference process

Improving transparency of deep neural inference process

Using Feature Selection Methods to Discover Common Users’ Preferences for Online Recommender Systems

AutoFocus: Interpreting Attention-Based Neural Networks by Code Perturbation

Contact Info

Product

Resources

About