Roberto Giacobazzi

It is widely known that the precision of a program analyzer is closely related to intensional program properties, namely, properties concerning how the program is written. This explains, for instance, the interest in code obfuscation techniques, namely, tools explicitly designed to degrade the results of program analysis by operating syntactic program transformations. Less is known about a possible relation between what the program extensionally computes, namely, its input-output relation, and the precision of a program analyzer. In this paper we explore this potential connection in an effort to isolate program fragments that can be precisely analyzed by abstract interpretation, namely, programs for which there exists a complete abstract interpretation. In the field of static inference of numeric invariants, this happens for programs, or parts of programs, that manifest a monotone (either non-decreasing or non-increasing) behavior. We first formalize the notion of program monotonicity with respect to a given input and a set of numerical variables of interest. A sound proof system is then introduced with judgments specifying whether a program is monotone relatively to a set of variables and a set of inputs. The interest in monotonicity is justified because we prove that the family of monotone programs admits a complete abstract interpretation over a specific class of non-trivial numerical abstractions and inputs. This class includes all non-relational abstract domains that refine interval analysis (i.e., at least as precise as the intervals abstraction) and that satisfy a topological convexity hypothesis.

Robustness is a key and desirable property of any classifying system, in particular, to avoid the ever-rising threat of adversarial attacks. Informally, a classification system is robust when the result is not affected by the perturbation of the input. This notion has been extensively studied, but little attention has been dedicated to how the perturbation affects the classification. The interference between perturbation and classification can manifest in many different ways, and its understanding is the main contribution of the present article. Starting from a rigorous definition of a standard notion of robustness, we build a formal method for accommodating the required degree of robustness—depending on the amount of error the analyst may accept on the classification result. Our idea is to precisely model this error as an abstraction . This leads us to define weakened forms of robustness also in the context of programming languages, particularly in language-based security, e.g., information-flow policies, and in program verification. The latter is possible by moving from a quantitative (standard) model of perturbation to a novel qualitative model, given by means of the notion of abstraction. As in language-based security, we show that it is possible to confine adversities, which means to characterize the degree of perturbation (and/or the degree of class generalization) for which the classifier may be deemed adequately robust. We conclude with an experimental evaluation of our ideas, showing how weakened forms of robustness apply to state-of-the-art image classifiers.

interpretation is a well-known and extensively used method to extract over-approximate program invariants by a sound program analysis algorithm. Soundness means that no program errors are lost and it is, in principle, guaranteed by construction. Completeness means that the abstract interpreter reports no false alarms for all possible inputs, but this is extremely rare because it needs a very precise analysis. We introduce a weaker notion of completeness, called local completeness , which requires that no false alarms are produced only relatively to some fixed program inputs. Based on this idea, we introduce a program logic, called Local Completeness Logic for an abstract domain A , for proving both the correctness and incorrectness of program specifications. Our proof system, which is parameterized by an abstract domain A , combines over- and under-approximating reasoning. In a provable triple ⊦ A [ p ] 𝖼 [ q ], 𝖼 is a program, q is an under-approximation of the strongest post-condition of 𝖼 on input p such that their abstractions in A coincide. This means that q is never too coarse, namely, under some mild assumptions, the abstract interpretation of 𝖼 does not yield false alarms for the input p iff q has no alarm . Therefore, proving ⊦ A [ p ] 𝖼 [ q ] not only ensures that all the alarms raised in q are true ones, but also that if q does not raise alarms, then 𝖼 is correct. We also prove that if A is the straightforward abstraction making all program properties equivalent, then our program logic coincides with O’Hearn’s incorrectness logic, while for any other abstraction, contrary to the case of incorrectness logic, our logic can also establish program correctness.

Imprecision is inherent in any decidable (sound) approximation of undecidable program properties. In abstract interpretation this corresponds to the release of false alarms, e.g., when it is used for program analysis and program verification. As all alarming systems, a program analysis tool is credible when few false alarms are reported. As a consequence, we have to live together with false alarms, but also we need methods to control them. As for all approximation methods, also for abstract interpretation we need to estimate the accumulated imprecision during program analysis. In this paper we introduce a theory for estimating the error propagation in abstract interpretation, and hence in program analysis. We enrich abstract domains with a weakening of a metric distance. This enriched structure keeps coherence between the standard partial order relating approximated objects by their relative precision and the effective error made in this approximation. An abstract interpretation is precise when it is complete. We introduce the notion of partial completeness as a weakening of precision. In partial completeness the abstract interpreter may produce a bounded number of false alarms. We prove the key recursive properties of the class of programs for which an abstract interpreter is partially complete with a given bound of imprecision. Then, we introduce a proof system for estimating an upper bound of the error accumulated by the abstract interpreter during program analysis. Our framework is general enough to be instantiated to most known metrics for abstract domains.