TRADER: Trace Divergence Analysis and Embedding Regulation for Debugging Recurrent Neural Networks
Technical
Recurrent Neural Networks (RNN) can deal with (textual) input with various length and hence have a lot of applications in software systems and software engineering applications. RNNs depend on word embeddings that are usually pre-trained by third parties to encode textual inputs to numerical values. It is well known that problematic word embeddings can lead to low model accuracy. In this paper, we propose a new technique to automatically diag- nose how problematic embeddings impact model performance, by comparing model execution traces from correctly and incorrectly executed samples. We then leverage the diagnosis results as guid- ance to harden/repair the embeddings. Our experiments show that TRADER can consistently and effectively improve accuracy for real world models and datasets by 5.37% on average, which represents substantial improvement in the literature of RNN models.
Fri 10 Jul Times are displayed in time zone: (UTC) Coordinated Universal Time change
16:05 - 17:05: A24-Testing and Debugging 4Paper Presentations / Technical Papers / New Ideas and Emerging Results / Journal First / Demonstrations at Silla Chair(s): Yijun YuThe Open University, UK | |||
16:05 - 16:11 Talk | Manifold for Machine Learning AssuranceNIER New Ideas and Emerging Results | ||
16:11 - 16:23 Talk | On Learning Meaningful Assert Statements for Unit Test CasesTechnical Technical Papers Cody WatsonWashington and Lee University, Michele TufanoMicrosoft, Kevin MoranWilliam & Mary/George Mason University, Gabriele BavotaUniversità della Svizzera italiana, Denys PoshyvanykWilliam and Mary Pre-print Media Attached | ||
16:23 - 16:35 Talk | TRADER: Trace Divergence Analysis and Embedding Regulation for Debugging Recurrent Neural Networks Technical Papers Guanhong TaoPurdue University, Shiqing MaRutgers University, Yingqi LiuPurdue University, USA, Qiuling XuPurdue University, Xiangyu ZhangPurdue University Pre-print | ||
16:35 - 16:38 Talk | DeepMutation: A Neural Mutation ToolDemo Demonstrations Michele TufanoMicrosoft, Jason KimkoWilliam & Mary, Shiya WangWilliam & Mary, Cody WatsonWashington and Lee University, Gabriele BavotaUniversità della Svizzera italiana, Massimiliano Di PentaUniversity of Sannio, Denys PoshyvanykWilliam and Mary Pre-print | ||
16:38 - 16:46 Talk | Specification Patterns for Robotic MissionsJ1 Journal First Claudio MenghiUniversity of Luxembourg, Christos TsigkanosTU Vienna, Patrizio PelliccioneUniversity of L'Aquila and Chalmers | University of Gothenburg, Carlo GhezziPolitecnico di Milano, Thorsten BergerChalmers | University of Gothenburg | ||
16:46 - 16:54 Talk | ProXray: Protocol Model Learning and Guided Firmware AnalysisJ1 Journal First Farhaan FowzeUniversity of Florida, Dave (Jing) TianPurdue University, Grant HernandezUniversity of Florida, Kevin ButlerUniv. Florida, Tuba YavuzUniversity of Florida | ||
16:54 - 17:00 Talk | Visual Sketching: From Image Sketches to CodeNIER New Ideas and Emerging Results Marcelo d'AmorimFederal University of Pernambuco, Rui AbreuInstituto Superior Técnico, U. Lisboa & INESC-ID, Carlos MelloFederal University of Pernambuco Pre-print Media Attached |