Software components provide essential means to structure and organize software effectively. However, frequently, required component abstractions are not available in a programming language or system, or are not adequately combinable with each other. Invasive software composition (ISC) is a general approach to software composition that unifies component-like abstractions such as templates, aspects and macros. ISC is based on fragment composition, and composes programs and other software artifacts at the level of syntax trees. Therefore, a unifying fragment component model is related to the context-free grammar of a language to identify extension and variation points in syntax trees as well as valid component types. By doing so, fragment components can be composed by transformations at respective extension and variation points so that always valid composition results regarding the underlying context-free grammar are yielded. However, given a language’s context-free grammar, the composition result may still be incorrect. Context-sensitive constraints such as type constraints may be violated so that the program cannot be compiled and/or interpreted correctly. While a compiler can detect such errors after composition, it is difficult to relate them back to the original transformation step in the composition system, especially in the case of complex compositions with several hundreds of such steps. To tackle this problem, this thesis proposes well-formed ISC—an extension to ISC that uses reference attribute grammars (RAGs) to specify fragment component models and fragment contracts to guard compositions with context-sensitive constraints. Additionally, well-formed ISC provides composition strategies as a means to configure composition algorithms and handle interferences between composition steps. Developing ISC systems for complex languages such as programming languages is a complex undertaking. Composition-system developers need to supply or develop adequate language and parser specifications that can be processed by an ISC composition engine. Moreover, the specifications may need to be extended with rules for the intended composition abstractions. Current approaches to ISC require complete grammars to be able to compose fragments in the respective languages. Hence, the specifications need to be developed exhaustively before any component model can be supplied. To tackle this problem, this thesis introduces scalable ISC—a variant of ISC that uses island component models as a means to define component models for partially specified languages while still the whole language is supported. Additionally, a scalable workflow for agile composition-system development is proposed which supports a development of ISC systems in small increments using modular extensions. All theoretical concepts introduced in this thesis are implemented in the Skeletons and Application Templates framework SkAT. It supports “classic”, well-formed and scalable ISC by leveraging RAGs as its main specification and implementation language. Moreover, several composition systems based on SkAT are discussed, e.g., a well-formed composition system for Java and a C preprocessor-like macro language. In turn, those composition systems are used as composers in several example applications such as a library of parallel algorithmic skeletons.

<p>Synthesized speech may be used for nefarious purposes, such as fraud, spoofing, and misinformation campaigns. We present several speech forensics methods based on deep learning to protect against such attacks. First, we use a convolutional neural network (CNN) and transformers to detect synthesized speech. Then, we investigate closed set and open set speech synthesizer attribution. We use a transformer to attribute a speech signal to its source (i.e., to identify the speech synthesizer that created it). Additionally, we show that our approach separates different known and unknown speech synthesizers in its latent space, even though it has not seen any of the unknown speech synthesizers during training. Next, we explore machine learning for an objective in the aerospace domain.</p> <p><br></p> <p>Compared to conventional ballistic vehicles and cruise vehicles, hypersonic glide vehicles (HGVs) exhibit unprecedented abilities. They travel faster than Mach 5 and maneuver to evade defense systems and hinder prediction of their final destinations. We investigate machine learning for identifying different HGVs and a conic reentry vehicle (CRV) based on their aerodynamic state estimates. We also propose a HGV flight phase prediction method. Inspired by natural language processing (NLP), we model flight phases as “words” and HGV trajectories as “sentences.” Next, we learn a “grammar” from the HGV trajectories that describes their flight phase transition patterns. Given “words” from the initial part of a HGV trajectory and the “grammar”, we predict future “words” in the “sentence” (i.e., future HGV flight phases in the trajectory). We demonstrate that this approach successfully predicts future flight phases for HGV trajectories, especially in scenarios with limited training data. We also show that it can be used in a transfer learning scenario to predict flight phases of HGV trajectories that exhibit new maneuvers and behaviors never seen before during training.</p>