Source author record

Taejoon Byun

Taejoon Byun appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Machine Learning Software Engineering

Catalog footprint

What is connected

3works

2topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2020arXiv

Manifold for Machine Learning Assurance

The increasing use of machine-learning (ML) enabled systems in critical tasks fuels the quest for novel verification and validation techniques yet grounded in accepted system assurance principles. In traditional system development, model-based techniques have been widely adopted, where the central premise is that abstract models of the required system provide a sound basis for judging its implementation. We posit an analogous approach for ML systems using an ML technique that extracts from the high-dimensional training data implicitly describing the required system, a low-dimensional underlying structure--a manifold. It is then harnessed for a range of quality assurance tasks such as test adequacy measurement, test input generation, and runtime monitoring of the target ML system. The approach is built on variational autoencoder, an unsupervised method for learning a pair of mutually near-inverse functions between a given high-dimensional dataset and a low-dimensional representation. Preliminary experiments establish that the proposed manifold-based approach, for test adequacy drives diversity in test data, for test generation yields fault-revealing yet realistic test cases, and for runtime monitoring provides an independent means to assess trustability of the target system's output.

preprint2020arXiv

Manifold-based Test Generation for Image Classifiers

Neural networks used for image classification tasks in critical applications must be tested with sufficient realistic data to assure their correctness. To effectively test an image classification neural network, one must obtain realistic test data adequate enough to inspire confidence that differences between the implicit requirements and the learned model would be exposed. This raises two challenges: first, an adequate subset of the data points must be carefully chosen to inspire confidence, and second, the implicit requirements must be meaningfully extrapolated to data points beyond those in the explicit training set. This paper proposes a novel framework to address these challenges. Our approach is based on the premise that patterns in a large input data space can be effectively captured in a smaller manifold space, from which similar yet novel test cases---both the input and the label---can be sampled and generated. A variant of Conditional Variational Autoencoder (CVAE) is used for capturing this manifold with a generative function, and a search technique is applied on this manifold space to efficiently find fault-revealing inputs. Experiments show that this approach enables generation of thousands of realistic yet fault-revealing test cases efficiently even for well-trained models.

preprint2013arXiv

Property-based Code Slicing for Efficient Verification of OSEK/VDX Operating Systems

Testing is a de-facto verification technique in industry, but insufficient for identifying subtle issues due to its optimistic incompleteness. On the other hand, model checking is a powerful technique that supports comprehensiveness, and is thus suitable for the verification of safety-critical systems. However, it generally requires more knowledge and cost more than testing. This work attempts to take advantage of both techniques to achieve integrated and efficient verification of OSEK/VDX-based automotive operating systems. We propose property-based environment generation and model extraction techniques using static code analysis, which can be applied to both model checking and testing. The technique is automated and applied to an OSEK/VDX-based automotive operating system, Trampoline. Comparative experiments using random testing and model checking for the verification of assertions in the Trampoline kernel code show how our environment generation and abstraction approach can be utilized for efficient fault-detection.