Vijendran G. Venkoparao

Shubham Sangle, Sandeep Muvva, Sridhar Chimalakonda et al.

Context: Given a bug report and source code of the project, bug localization can help developers to focus on fixing probable buggy files rather than searching the entire source code repository. While existing research uses information retrieval (IR) and/or combination of machine learning (ML) or deep learning (DL) approaches, they focus primarily on benchmark Java projects, and also motivate the need for multi-language bug localization approach. Objective: To create a novel bug localization approach that leverages the syntactic structure of source code, bug report information and which can support multi-language projects along with a new dataset of C projects. Method: The proposed DRAST approach represents source code as code vectors by using its high-level AST and combines rVSM, an IR technique with ML/DL models such as Random Forest and Deep Neural Network regressor to rank the list of buggy files. We also use features such as textual similarity using IR techniques, lexical mismatch using DNNs, and history of the project using the metadata of BugC dataset. Results: We tested DRAST on seven projects from the BugC dataset, which consists of 2462 bug reports from 21 open-source C projects. The results show that DRAST can locate correct buggy files 90% of the time from top 1, 5, and 10 suggested files with MAP and MRR scores of above 90% for the randomly selected seven projects. We also tested DRAST on Tomcat and AspectJ, projects from benchmark dataset with better results at accuracy@1, MAP and MRR when compared with state-of-the-art. Conclusions: This paper presents a novel bug localization approach that works on C and Java projects and a bug localization C dataset along with a novel source code representation. The results for C projects using DRAST are promising and could motivate researchers/practitioners to focus on developing and creating multi-language bug localization approaches.

Saurav Musunuru, Jay N. Paranjape, Rahul Kumar Dubey et al.

Deep Learning has become one of the primary research areas in developing intelligent machines. Most of the well-known applications (such as Speech Recognition, Image Processing and NLP) of AI are driven by Deep Learning. Deep Learning algorithms mimic human brain using artificial neural networks and progressively learn to accurately solve a given problem. But there are significant challenges in Deep Learning systems. There have been many attempts to make deep learning models imitate the biological neural network. However, many deep learning models have performed poorly in the presence of adversarial examples. Poor performance in adversarial examples leads to adversarial attacks and in turn leads to safety and security in most of the applications. In this paper we make an attempt to characterize the solution space of a deep neural network in terms of three different subsets viz. weights belonging to exact trained patterns, weights belonging to generalized pattern set and weights belonging to adversarial pattern sets. We attempt to characterize the solution space with two seemingly different learning paradigms viz. the Deep Neural Networks and the Dense Associative Memory Model, which try to achieve learning via quite different mechanisms. We also show that adversarial attacks are generally less successful against Associative Memory Models than Deep Neural Networks.