Thibault Dardinier
PhD student, Programming Methodology Group, ETH Zurichthibault.dardinier[at]inf.ethz.ch
About Publications Formalizations Education Experience Awards Teaching Other Activities
About
I am a PhD student in the Programming Methodology Group at ETH Zurich, supervised by Prof. Peter Müller. My goal is to develop formal foundations for automated deductive verification based on separation logic, with a particular focus on intermediate verification languages (such as Viper).
Before this, I graduated from ETH Zurich (as part of the Direct Doctorate program) and from École Polytechnique in France. During my studies, I completed internships at SRI International, Siemens, and the Ministry of Defense (France).
Publications
A Formally Verified, Optimized Monitor for Metric First-Order Dynamic Logic
David Basin, Thibault Dardinier, Lukas Heimes, Srđan Krstić, Martin Raszyk, Joshua Schneider, Dmitriy TraytelInternational Joint Conference on Automated Reasoning (IJCAR)
Abstract Link
Runtime monitors for rich specification languages are sophisticated algorithms, especially when they are heavily optimized. To gain trust in them and safely explore the space of possible optimizations, it is important to verify the monitors themselves. We describe the development and correctness proof in Isabelle/HOL of a monitor for metric first-order dynamic logic. This monitor significantly extends previous work on formally verified monitors by supporting aggregations, regular expressions (the dynamic part), and optimizations including multi-way joins adopted from databases and a new sliding window algorithm.
Beyond the Frame Rule: Static Inlining in Separation Logic
Thibault DardinierMaster's Thesis. ETH Zurich, Switzerland.
Abstract Link
Various formal verification techniques can be used to automatically verify the absence of errors in programs. This provides an advantage over testing approaches, namely the guarantee that a program is correct for any possible execution. However, such approaches often require a user to provide additional specifications to guide the verification, in the form of loop invariants and method preconditions and postconditions, which places a burden on the user. When no specifications are provided, verifiers usually report potential errors which are not actual errors, hence lowering confidence in error reporting. Users might want to learn quickly (without providing too many specifications) and with high confidence whether a program is incorrect. This would speed up the development and verification process by only having to provide specifications when one is fairly certain that the program is correct. Static inlining, that is inlining of method calls and unrolling of loop iterations, is an interesting approach to tackle this issue. Using approaches based on static inlining, verifiers could inform a user of the existence of fundamental errors, errors for which no annotation can make the program verify. The existence of fundamental errors indicates an error in the program itself, informing the user this program cannot be verified, without the user needing to waste time and energy in the search of the right annotation. One would expect that errors reported in an inlined program always correspond to fundamental errors in the original program, since annotations only serve as approximations of method calls and loops. Surprisingly, this is not always the case. Indeed, Viper, a verification infrastructure for permissionbased reasoning, partly based on separation logic and implicit dynamic frames with fractional permissions, has special features (such as permission introspection) which give rise to examples where this is not the case. These examples have all in common that they do not satisfy the frame rule. In this thesis, we find (and prove correct) a soundness condition which characterizes the set of Viper programs for which static inlining is sound, that is errors in the inlined program correspond to fundamental errors in the original program. We define a parametric language which generalizes Viper, where program states are elements of a separation algebra, define the soundness condition in terms of this language, and prove the soundness property of inlining under this soundness condition, using the proof assistant Isabelle/HOL. We then show how one can instantiate this parametric language to transfer the results to Viper. We also explore a completeness property of static inlining, and consider extensions to different loop semantics and different ways of inlining.
A new analysis method for evolutionary optimization of dynamic and noisy objective functions
Raphaël Dang-Nhu, Thibault Dardinier, Benjamin Doerr, Gautier Izacard, Dorian NognengProceedings of the Genetic and Evolutionary Computation Conference (GECCO)
Abstract BibTex Link
Evolutionary algorithms, being problem-independent and randomized heuristics, are generally believed to be robust to dynamic changes and noisy access to the problem instance. We propose a new method to obtain rigorous runtime results for such settings. In contrast to many previous works, our new approach mostly relies on general parameters of the dynamics or the noise models, such as the expected change of the dynamic optimum or the probability to have a dynamic change in one iteration. Consequently, we obtain bounds which are valid for large varieties of such models. Despite this generality, for almost all particular models regarded in the past our bounds are stronger than those given in previous works. As one particular result, we prove that the (1 + λ) EA can optimize the OneMax benchmark function efficiently despite a constant rate of 1-bit flip noise. For this, a logarithmic size offspring population suffices (the previous-best result required a super-linear value of λ). Our results suggest that the typical way to find the optimum in such adverse settings is not via a steady approach of the optimum, but rather via an exceptionally fast approach after waiting for a rare phase of low dynamic changes or noise.
Isabelle/HOL Formalizations
Formalization of an Optimized Monitoring Algorithm for Metric First-Order Dynamic Logic with Aggregations
Thibault Dardinier, Lukas Heimes, Martin Raszyk, Joshua Schneider and Dmitriy TraytelAbstract BibTex Link
A monitor is a runtime verification tool that solves the following problem: Given a stream of time-stamped events and a policy formulated in a specification language, decide whether the policy is satisfied at every point in the stream. We verify the correctness of an executable monitor for specifications given as formulas in metric first-order dynamic logic (MFODL), which combines the features of metric first-order temporal logic (MFOTL) and metric dynamic logic. Thus, MFODL supports real-time constraints, first-order parameters, and regular expressions. Additionally, the monitor supports aggregation operations such as count and sum. This formalization, which is described in a forthcoming paper at IJCAR 2020, significantly extends previous work on a verified monitor for MFOTL. Apart from the addition of regular expressions and aggregations, we implemented multi-way joins and a specialized sliding window algorithm to further optimize the monitor.
Formalization of Multiway-Join Algorithms
Thibault DardinierAbstract BibTex Link
Worst-case optimal multiway-join algorithms are recent seminal achievement of the database community. These algorithms compute the natural join of multiple relational databases and improve in the worst case over traditional query plan optimizations of nested binary joins. In 2014, Ngo, Ré, and Rudra gave a unified presentation of different multi-way join algorithms. We formalized and proved correct their "Generic Join" algorithm and extended it to support negative joins.
Education
Direct Doctorate, ETH Zurich
Zurich, Switzerland The Direct Doctorate program combines a Master and a PhD in 5-6 years.Master of Computer Science with distinction, ETH Zurich, 5.8/6
Zurich, SwitzerlandDiplôme d'Ingénieur (MSc), École Polytechnique, 3.96/4 (top 5%)
Palaiseau, FranceClasses Préparatoires (MPSI-MP), Lycée du Parc
Lyon, FranceExperience
Static Inlining in Separation Logic, ETH Zurich (Programming Methodology Group)
Master's Thesis Supervision: Gaurav Parthasarathy, Prof. Peter Müller- Exploration of static inlining in non-monotonic separation logic to detect program errors early (without annotations), applied to the intermediate verification language Viper.
- Received the ETH Medal.
- See the thesis.
Formalizing Multiway-Join Algorithms in Isabelle/HOL, ETH Zurich (Information Security Group)
Direct doctorate research project Supervision: Dr. Dmitriy Traytel, Martin RaszykResearch Fellow, SRI International (CSL)
Menlo Park, California, USASupervision: Dr. Susmit Jha
Theoretical Analysis of Evolutionary Algorithms, École Polytechnique (LIX)
Supervision: Prof. Benjamin DoerrR&D Intern, Siemens (Corporate Technology)
Munich, GermanyResearch Intern, Ministry of Defense (France)
Paris, FranceAwards
ETH Medal, ETH Zurich
Awarded by ETH Zürich for my Master's thesis on inlining in separation logic, along with a financial sum. The ETH Medal is awarded to less than 2.5% of all Master's graduates of an ETH department.Research Internship Prize, École Polytechnique
Awarded by École Polytechnique for my work on Gaussian Processes at SRI International.National Defense Medal (France)
Awarded for my internship at the Ministry of Defense.3rd Place, Prologin (National Programming Contest)
Prologin is a contest for French speaking students under 21, with qualifiers. The best 100 candidates meet and discover a multiplayer game created for the contest. They have 36 hours in a row to create an AI for the game. A ranking is established with the results of a contest between all AIs. Ranked 3rd.2nd Place, Prologin (National Programming Contest)
Ranked 2nd.Teaching
Teaching Assistant, ETH Zurich
Zurich, SwitzerlandOral Examiner in Mathematics, Lycée Blaise Pascal
Orsay, FranceTwo hours weekly for undergraduates preparing competitive entrance exams to France’s top-ranking scientific schools.
Other Activities
Development of an Online Video Platform for the Video Association (JTX)
Development of an Optimization Algorithm (Bin Packing) for the Industry
Volunteer, Mission Potosi
Potosi, BoliviaMission Potosi is an organisation with about 50 students from 4 schools (business, medical and engineering), which collaborates with local ONGs (Cepromin, Pasocap) and a german ONG (Kindernothilfe) in Potosi (Bolivia) to provide childcare for mining families.
Military Training as an Officer
Coëtquidan, FranceThis military training, part of the cursus at École polytechnique, teaches us a lot on various subjects such as strategy, communication, leadership, etc.