Fisher Subspace Tree Classifier Based on Neural Networks

Chen, Dongyue; Lu, Xiaodan; Zhang, Liming

doi:10.1007/11427445_3

Cited by 6 publications

(24 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since BLEU-4 and EM Acc metrics may not directly measure the semantic correctness of the results, we use computational accuracy [21] to evaluate whether the programs produced by SDA-Trans can generate the same outputs as the ground truth given the same input data. Except for previous baselines, we also compare the results with previously introduced baseline approaches and two rulebased approaches: j2py 7 for Java → Python translation, and a commercial tool 8 , which can perform translation from C++ to Java. In this experiment, all the translations are generated using beam size 5.…”

Section: B Rq2 How Do Different Components In Sda-trans Contribute To...mentioning

confidence: 99%

“…With the development of NMT techniques, researchers began to explore the possibility of adapting neural machine translation to program translation task. Chen et al [8] proposed a tree-to-tree neural architecture. They parsed programs into ASTs and converted them into binary trees, then fed the trees into a Tree-LSTM based encoder-decoder neural model.…”

Section: Translation Of Programming Languagesmentioning

confidence: 99%

“…To meet new business needs and environments, one software initially developed in a language needs to migrate to other languages [40,28]. Code migration is a crucial activity during software development and maintenance [8]. There is a tremendous need for translating programs between different programming languages.…”

Section: Introductionmentioning

confidence: 99%

“…For example, it takes about 5 years and * Corresponding author $750 million for Australia Commonwealth Bank to convert the platform from COBOL to Java [21]. To improve the quality and efficiency, researchers began to explore the automatic approaches to build a program translator [27,28,8,21]. Based on the naturalness theory of programming languages, i.e., source code is repetitive and predictable, statistical machine learning methods have been used for modeling programs and code migration.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Syntax and Domain Aware Model for Unsupervised Program Translation

Liu¹,

Li²,

Zhang³

2023

Preprint

View full text Add to dashboard Cite

There is growing interest in software migration as the development of software and society. Manually migrating projects between languages is error-prone and expensive. In recent years, researchers have begun to explore automatic program translation using supervised deep learning techniques by learning from large-scale parallel code corpus. However, parallel resources are scarce in the programming language domain, and it is costly to collect bilingual data manually. To address this issue, several unsupervised programming translation systems are proposed. However, these systems still rely on huge monolingual source code to train, which is very expensive. Besides, these models cannot perform well for translating the languages that are not seen during the pre-training procedure. In this paper, we propose SDA-Trans, a syntax and domain-aware model for program translation, which leverages the syntax structure and domain knowledge to enhance the cross-lingual transfer ability. SDA-Trans adopts unsupervised training on a smallerscale corpus, including Python and Java monolingual programs. The experimental results on function translation tasks between Python, Java, and C++ show that SDA-Trans outperforms many large-scale pre-trained models, especially for unseen language translation.

show abstract

Section: B Rq2 How Do Different Components In Sda-trans Contribute To...mentioning

confidence: 99%

Section: Translation Of Programming Languagesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Syntax and Domain Aware Model for Unsupervised Program Translation

Liu¹,

Li²,

Zhang³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Abstract syntax tree is completed in the syntax analysis phase, used to represent the abstract syntax structure of source code in a programming language. Chen et al [23] proposed a Tree-to-Tree model with Tree-LSTM [9] to encode the input AST of source program into vectors. When the decoder expands a non-terminal, it employs an attention mechanism to identify the corresponding subtree in the source tree.…”

Section: Abstract Syntax Tree-based Program Translationmentioning

confidence: 99%

Human-Machine Cooperative Program Translation Based on Abstract Syntax Tree

Zhang,

Liu,

et al. 2023

Preprint

View full text Add to dashboard Cite

Program translation involves converting one programming language into another, which can modernize and migrate applications across platforms. Most of the current program translation methods are sequence-based, which cannot capture the semantic information of the code. To overcome this, abstract syntax tree-based program translation rises up. However, even the state-of-the-art machine translation methods can not guarantee error-free translation results. Consequently, software engineers need to make corrections on the machine's results before using, which is labor-intensive. To solve this problem, we propose the Human-Machine Cooperative Program Translation Based on Abstract Syntax Tree (HMPT-AST), which feeds the human's modification into the model in each iteration and the model updates the result based on human's feedback. It enables two patterns of human participation: First-error-based and Layer-based HMPT-AST. In the former one, the software engineer corrects one error node in each iteration. In the latter one, all error nodes within the same layer are corrected. In addition, the proposed method can also reduce the interactive response time from two aspects: avoiding generating the correct AST subtree repeatedly with parameter cache module, as well as reducing the generation of unnecessary AST subtree with early termination module. We conducted experiments on multiple dataset. The results show that compared with baselines, our method reduces the human effort up to 94.44\% and reduces the response time up to 94.32\%. We have made the code publicly available.\footnote{https://drive.google.com/drive/folders/1878Or6cEsW21fu_cpdpnv_dGevfFr9mB?usp=sharing}

show abstract

Enhancing Bug Report Summaries Through Knowledge-Specific and Contrastive Learning Pre-Training

Shao,

Xiang

2024

IEEE Access

View full text Add to dashboard Cite

Bug reports are crucial in software maintenance, with concise summaries significantly enhancing the efficiency of bug triagers and ultimately contributing to the development of high-quality software products. Contemporary methods for automatic bug report summarization primarily utilize neural networks' robust learning capabilities. However, these approaches often produce suboptimal summaries due to two primary limitations: 1) the difficulty in assimilating the domain-specific knowledge inherent in bug reports, and 2) the limitations of purely supervised learning in comprehending the comprehensive context of bug reports. To address the above two problems, in this paper, we propose a new approach for bug report summarization, namely KSCLP, which leverages large language models and domain-specific pre-training strategies, i.e., Knowledge-Specific and Contrastive Learning Pre-training. Specifically, the Knowledge-Specific strategy allows to pre-train KSCLP on project-specific bug reports corpus, by which the model can fully learn internal knowledge of bug reports, learning bug report-aware representation. As for the Contrastive Learning strategy, it performs a sequence-level pre-training for KSCLP, helping it capture the semantic information of bug reports on a global level. Upon completion of the pre-training phase, KSCLP undergoes further refinement through a Sequence-to-Sequence framework specifically tailored for bug report summarization. The efficacy of KSCLP is rigorously evaluated against five baseline models using a publicly available dataset. The empirical results demonstrate that KSCLP outperforms all baselines, achieving remarkable improvements by up to 23.73, 13.97, and 20.89 points in ROUGE-1, ROUGE-2, and ROUGE-L metrics, thereby setting new benchmarks in the field of bug report summarization.

show abstract

Fisher Subspace Tree Classifier Based on Neural Networks

Cited by 6 publications

References 3 publications

Syntax and Domain Aware Model for Unsupervised Program Translation

Syntax and Domain Aware Model for Unsupervised Program Translation

Human-Machine Cooperative Program Translation Based on Abstract Syntax Tree

Enhancing Bug Report Summaries Through Knowledge-Specific and Contrastive Learning Pre-Training

Contact Info

Product

Resources

About