Exploiting structure for scalable software verification
Bibliographic record
Abstract
Software bugs are expensive. Recent estimates by the US National Institute of Standards and Technology claim that the cost of software bugs to the US economy alone is approximately 60 billion USD annually. As society becomes increasingly software-dependent, bugs also reduce our productivity and threaten our safety and security. Decreasing these direct and indirect costs represents a significant research challenge as well as an opportunity for businesses. Automatic software bug-finding and verification tools have a potential to completely revolutionize the software engineering industry by improving reliability and decreasing development costs. Since software analysis is in general undecidable, automatic tools have to use various abstractions to make the analysis computationally tractable. Abstraction is a double-edged sword: coarse abstractions, in general, yield easier verification, but also less precise results. This thesis focuses on exploiting the structure of software for abstracting away irrelevant behavior. Programmers tend to organize code into objects and functions, which effectively represent natural abstraction boundaries. Humans use such structural abstractions to simplify their mental models of software and for constructing informal explanations of why a piece of code should work. A natural question to ask is: How can automatic bug-finding tools exploit the same natural abstractions? This thesis offers possible answers. More specifically, I present three novel ways to exploit structure at three different steps of the software analysis process. First, I show how symbolic execution can preserve the data-flow dependencies of the original code while constructing compact symbolic representations of programs. Second, I propose structural abstraction, which exploits the structure preserved by the symbolic execution. Structural abstraction solves a long-standing open problem --- scalable interprocedural path- and context-sensitive program analysis. Finally, I present an automatic tuning approach that exploits the fine-grained structural properties of software (namely, data- and control-dependency) for faster property checking. This novel approach resulted in a 500-fold speedup over the best previous techniques. Automatic tuning not only redefined the limits of automatic software analysis tools, but also has already found its way into other domains (like model checking), demonstrating the generality and applicability of this idea.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.017 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.005 |
| Scholarly communication | 0.003 | 0.009 |
| Open science | 0.002 | 0.006 |
| Research integrity | 0.001 | 0.004 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".