2023Activity reportProject-TeamLINKS

3.2.1 AI: Circuits for Data Analysis

Circuits are concise representations of data sets that recently found a unifying interest in various areas of artificial intelligence. A circuit may for instance represent the answer set of a database query as a dag whose operators are disjoint unions (for disjunction) and Cartesian products (for conjunction). Similarly, it may also represent the set of all matches of a pattern in a graph. The structure of the circuit may give rise to efficient algorithms to process large data sets based on representation that are often much smaller. Among others, such applications range from knowledge representation/compilation, counting the number of solutions of queries, efficient query answering, factorized databases.

In a first line of research, we want to study novel problems on circuits, in which database queries are relevant to data analysis tasks from artificial intelligence, in machine learning or data mining in particular. In particular we propose to study optimization problems on answer sets of database queries based on circuits, i.e. how to find optimal solutions in the answer set for a given set of conditions. Decompressing small circuits into large answer sets would make the optimization problem unfeasible in many cases. We believe that circuits can structure certain optimization problems in such a way that it can be phrased concisely and then solved efficiently.

Second, we propose to develop a tighter integration between circuits and databases. Indeed query-related circuits are generally produced from a database. This requires that the data is copied within the circuits. This memory cost is accompanied with the loss of the environment of the DBMS which allows many optimizations and uses many low level optimizations that are hard to implement. We propose then to encode circuits directly within the database using materialized views and index structures. We shall also develop the required computational tools for maintaining and exploiting these embedded circuits.

3.2.2 Path Query Optimization

Graph databases are easily queried using path descriptions. Most often these paths are described by means of regular expressions. This makes path queries difficult as the use of Kleene star makes them recursive. In relational DBMS, recursion is almost never used and it is not advised to use it. The natural theoretical tool that pertains to recursion in the context of relational data Datalog. There has been a wealth of optimization algorithms that have been proposed for queries written in Datalog. We propose to use Datalog as a low level language to which we will compile path queries of various kinds. The idea is that the compiler will try to obtain Datalog programs that will have low execution complexity taking advantages of optimization techniques such as magic supplementary set rewriting, pre-computed indexes and also statistics computed from the graph. Our goal is to develop a compiler that will be able to efficiently evaluate path queries on large graphs which in particular will explore only a part of it.

3.3.1 Functional Programming Languges for Data Graphs

The first question is which kind of programming language to use to enable monitoring processes for data graphs based on query answering. While languages of path queries found quite some interest on data graphs, less attention has been given to the programming language tasks, that needed to be solved to produce structured output and to compose various queries with structured output into a pipeline. We believe that transferring the generalization of ideas developed for data trees in the context of XML to data graphs will allow to solve such problems in a systematic manner.

Our approach will consist in developing a functional programming language based on first principles (the lambda calculus, graph navigation, logical connective) that generalizes full XPath 3.0 to the context of graphs. Here we can rely on own previous work for data trees, such as the language X-Fun and $\lambda$-XP. After the language for data graphs is designed we shall study its behavior empirically by means of an implementation. This implementation will help us to design optimization methods so as to evaluate the queries in that language. This will allow us to use a wealth of techniques so as to optimize the computation. Indeed, we can try to compile data structures to imperative ones when possible and also exploit possibilities of parallel executions in certain cases. Functional programming comes with nice verification techniques that we are going to use in several contexts: (i) in optimizing queries (e.g. stop the evaluation when it is possible to know that no more data can contribute to the output) and (ii) to verify that the query behaves correctly. The verification methods we shall focus on will be mainly related to automata and transducers.

Finally we shall also develop a programming language that allows to describe services that use datagraphs as a backend for storing data. Here again, functional programming seems a good candidate, we would need however to orchestrate the concurrent executions of queries so as to ensure the correct behavior of services. This means that we should have concurrent constructs that are built in the language. The high level of concurrence enabled by the notion of futures seems an interesting candidate to adapt to the context of service orchestration.

3.3.2 Hyperstreaming Program Evaluation

Complex-event processing requires to monitor data graphs that are produced on input streams and to write data graphs to some output stream, which can then be used as inputs again. A major problem here is to reduce the high risk of blocking, which arises when the writing of some of the output stream suspends on a data value that will become available only in the future on some input stream. In such cases, all monitoring processes reading the output stream may have to suspend as well. In order to reduce the risk of blocking, we propose to develop the hyperstreaming approach further, of which we layed the foundations in the evaluation period based on automata techniques. The idea is to generalize streams to hyperstreams, i.e. to add holes to streams that can be filled by some other stream in the future. In order to avoid suspension as possible, a monitoring process on hyperstream must then be able to jump over the holes, and to perform some speculative computation. The objective for the next period are to develop tools for hyperstreaming query answering and to lift these to hyperstreaming program evaluation. Furthermore, on the conceptual side, the notion of certain query answers on hyperstreams needs to be lifted to certain program outputs on hyperstreams.

3.4.1 Data Quality with Schemas and Repairing with Inference

The data quality of RDF may suffer due to a number of reasons. Impurities may arise due to data value errors (misspellings, errors during data entry etc.). Such data quality problems have been thoroughly investigated in literature for relational databases and solutions include dictionary methods etc. However, it remains to be seen if the challenges of adapting the existing solutions for relational databases can be easily addressed.

One particular challenge comes from the fact that RDF allows a higher degree of structural freedom in how information is represented as opposed to relation databases, where the choice is strongly limited to flat tables. We plan to investigate suitability of existing data cleaning methods to tackle the problems of data value impurities in RDF. The structural freedom of RDF is a source of data quality issues on its own. With the recent emergence of schema formalisms for RDF, it becomes evident that significant parts of existing RDF repositories do not necessarily satisfy schemas prepared by domain experts.

In the first place, we intend to investigate defining suitable measures of quality for RDF documents. Our approaches will be based on a schema language, such as ShEx and SHACL, and we shall explore suitable variants of graph alignment and graph edit distance to capture similarity between the existing RDF document and its possible repaired versions that satisfy the schema.

The central issue here is repairing an RDF document w.r.t. schema by identifying essential fragments of the RDF that fail to satisfy the schema. Once such fragments are identified, repairing actions can be applied however there might be a significant number of alternatives. We intend to explore enumeration approaches where the space of repairing alternatives is intelligently browsed by the user and the most suitable one is chosen. Furthermore, we intend to propose a rule language for choosing the most suitable repairing action and will investigate inference methods to derive from interactions with user the optimal order in which various repairing actions are presented to the user and derive the rules for the choice of the preferred repairing action for repeating types of fragments that do not satisfy the schema.

3.4.2 Integration and Graph Mappings with Schemas and Inference

The second problem pertaining to integration of RDF data sources is their heterogeneity. We intend to continue to identify and study suitable classes of mappings between RDF documents conforming to potentially different and complementary schemas. We intend to assist the user in constructing such mappings by developing rich and expressive graphical languages for mappings. Also, we wish to investigate inference of RDF mappings with the active help of an expert user. We will need to define interactive protocols that allows the input to be sufficiently informative to guide the inference process while avoiding the pitfalls of user input being too ambiguous and causing combinatorial explosion. We intend to identify

RDF Data Quality. Approach based on a schema language (ShEx or SHACL) used to identify errors and giving a notion of a measure of quality of an RDF database. Impurities in RDF may come from data value errors (misspellings etc.) but also from the fact that RDF imposes fewer constraints on how data is structured which is a consequence of a significantly different use philosophy (just publish your data anyway you want). Repairing of RDF errors would be modeled with a localized rules (transformations that operate within a small radius of an affected node) and if several rules apply, preferences are used to identify the most desirable one. Both the repairing rules and preferences can be inferred with the help of inference algorithms in an interactive setting. Smart tools for LOD integration. Assuming that the LOD sources are of good quality, we want to build tools that assist the user in constructing mappings that integrate data in the user database. For this, we want to define inference algorithms which are guided by schemas, and which are based on comprehensible interactions with the user. For this, we need to define interactions that are rich enough to inform the algorithm, while simple enough to be understandable by a non-expert user. In particular, that means that we need to present data (nodes in a graph for instance) in a readable way. Also, we want to investigate how the - possibly inferred - schema can be used to guide the inference.

7.1.1 NetworkDisk

• Name:
  NetworkDisk
• Keywords:
  Large graphs, Python, Databases
• Functional Description:
  NetworkDisk provides a way to manipulate graphs on disk. The goal is to be as much as possible compatible with (Di)Graph objects of the NetworkX Python package but lifting memory requirement and providing persistence of the Graph.
• URL:
  https://­networkdisk.­inria.­fr/
• Contact:
  Charles Paperman

7.1.2 Bibendum

• Name:
  Bibendum
• Keyword:
  Bibliography
• Functional Description:

  Small app to fetch bibtex from a short label with the format: LastName.Year.PublicationTerm where the . denote the concatenation. For instance Codd1970Relational. LastName is the last name of one of the authors. Year is the publication year. PublicatonTerm is one meaningful word in the title of the publication.

  In case of ambiguity, an extra integer is used. Ambiguous entries are resolved by sorting dois under lexicographical order. The api is idempotent, every decision taken is recorded and replayed.

• URL:
  https://­gitlab.­inria.­fr/­cpaperma/­bibendum
• Contact:
  Charles Paperman

7.1.3 XPath AutoBench

• Name:
  A Benchmark Collection of Deterministic Automata for XPath Queries
• Keywords:
  XML, Querying, Tree Automata
• Functional Description:
  We provide a benchmark collection of deterministic automata for regular XPath queries. For this, we select the subcollection of forward navigational XPath queries from a corpus that Lick and Schmitz extracted from real-world XSLT and XQuery programs, compile them to stepwise hedge automata (SHAs), and determinize them. Large blowups by automata determinization are avoided by using schema-based determinization. The schema captures the XML data model and the fact that any answer of a path query must return a single node. Our collection also provides deterministic nested word automata that we obtain by compilation from deterministic SHAs.
• URL:
  https://­archive.­softwareheritage.­org/­browse/­origin/­directory/­?origin_url=https://­gitlab.­inria.­fr/­aalserha/­xpath-benchmark
• Contact:
  Joachim Niehren

7.1.4 rsonpath

• Keywords:
  JSon, Streaming, SIMD, Rust
• Functional Description:
  The rsonpath crate provides a JSONPath parser and a query execution engine, which utilizes SIMD instructions to provide massive throughput improvements over conventional engines.
• URL:
  https://­github.­com/­V0ldek/­rsonpath
• Contact:
  Charles Paperman
• Partner:
  Warsaw University

7.1.5 Coussinet

• Name:
  Coussinet
• Keywords:
  Enumeration, Complexity
• Functional Description:
  Coussinet is a demo illustrating a technique called geometric amortization for enumeration algorithms introduced in the paper Geometric Amortization for Enumeration Algorithms, Florent Capelli, Yann Strozecki. The result presented in this paper is about making the delay of enumeration algorithms more regular.
• URL:
  http://­florent.­capelli.­me/­coussinet/­coussinet.­html
• Contact:
  Florent Capelli
• Participants:
  Florent Capelli, Yann Strozecki

7.1.6 ShEx validator

• Name:
  Validation of Shape Expression schemas
• Keywords:
  Data management, RDF
• Functional Description:
  Shape Expression schemas is a formalism for defining constraints on RDF graphs. This software allows to check whether a graph satisfies a Shape Expressions schema.
• Release Contributions:

  ShExJava now uses the Commons RDF API and so support RDF4J, Jena, JSON-LD-Java, OWL API and Apache Clerezza. It can parse ShEx schema in the ShEcC, ShEJ, ShExR formats and can serialize a schema in ShExJ.

  To validate data against a ShExSchema using ShExJava, you have two different algorithms: - the refine algorithm: compute once and for all the typing for the whole graph - the recursive algorithm: compute only the typing required to answer a validate(node,ShapeLabel) call and forget the results.

• URL:
  https://­github.­com/­iovka/­shex-java
• Contact:
  Iovka Boneva

7.1.7 gMark

• Name:
  gMark: schema-driven graph and query generation
• Keywords:
  Semantic Web, Data base
• Functional Description:
  gMark allow the generation of graph databases and an associated set of query from a schema of the graph.gMark is based on the following principles: - great flexibility in the schema definition - ability to generate big size graphs - ability to generate recursive queries - ability to generate queries with a desired selectivity
• URL:
  https://­github.­com/­graphMark/­gmark
• Contact:
  Aurélien Lemay

8.1.1 Circuits for Data Analysis in Artificial Intelligence

In a STACS'2023 paper, Charles Paperman, Sylvain Salvati, and their PhD student Claire Soyez-Martin develop theoretical aspects of vectorial programming, the combination of SIMD instructions with usual processor instructions. This new technology is known to speed-up many standard algorithms, such as for simple regular languages. Their idea is to take advantage of the inner algebraic structure of regular languages and produce high level representations of efficient vectorial programs that recognize certain classes of regular languages. As a technical ingredient, they establish equivalences between classes of vectorial circuits and logical formalisms, namely unary temporal logic and first order logic. Their main result is the construction of compilation procedures that turns syntactic semigroups into vectorial circuits. The circuits they obtain are small in that they improve known upper-bounds on representations of automata within the logical formalisms. The gain is mostly due to a careful sharing of sub-formulas based on algebraic tools.

In an LMCS journal article 14, Capelli, Crosetti and Niehren study the problem of optimizing a linear program whose variables are the answers to a conjunctive query. For this they propose the language LP(CQ) for specifying linear programs whose constraints and objective functions depend on the answer sets of conjunctive queries. They contribute an efficient algorithm for solving programs in a fragment of LP(CQ). The natural approach constructs a linear program having as many variables as there are elements in the answer set of the queries. Their approach constructs a linear program having the same optimal value but fewer variables. This is done by exploiting the structure of the conjunctive queries using generalized hypertree decompositions of small width to factorize elements of the answer set together. They then illustrate the various applications of LP(CQ) programs on three examples: optimizing deliveries of resources, minimizing noise for differential privacy, and computing the s-measure of patterns in graphs as needed for data mining.

In an article recently accepted at ICDT'24 24, Capelli, Monet and co-authors study the problem of enumerating the satisfying assignments for circuit classes from knowledge compilation, where assignments are ranked in a specific order. In particular, they show how this problem can be used to efficiently perform ranked enumeration of the answers to MSO queries over trees, with the order being given by a ranking function satisfying a subset-monotonicity property. Assuming that the number of variables is constant, they show how to can enumerate the satisfying assignments in ranked order for so-called multivalued circuits that are smooth, decomposable, and in negation normal form (smooth multivalued DNNF). There is no preprocessing and the enumeration delay is linear in the size of the circuit times the number of values, plus a logarithmic term in the number of assignments produced so far. If one further assumes that the circuit is deterministic (smooth multivalued d-DNNF), then they show that it is possible to achieve linear-time preprocessing in the circuit, and the delay only features the logarithmic term.

In a preprint 25, Capelli and Irwin study the following problem. Given a conjunctive query $Q$ and a database $D$, a direct access to the answers of $Q$ over $D$ is the operation of returning, given an index $j$, the $j$-th answer for some order on its answers. While this problem is #P-hard in general with respect to combined complexity, many conjunctive queries have an underlying structure that allows for a direct access to their answers for some lexicographical ordering that takes polylogarithmic time in the size of the database after a polynomial time precomputation. Previous work has precisely characterised the tractable classes and given fine-grained lower bounds on the precomputation time needed depending on the structure of the query. In this paper, they generalise these tractability results to the case of signed conjunctive queries, that is, conjunctive queries that may contain negative atoms. Their technique is based on a class of circuits that can represent relational data. They first show that this class supports tractable direct access after a polynomial time preprocessing. They then give bounds on the size of the circuit needed to represent the answer set of signed conjunctive queries depending on their structure. Both results combined together allow them to prove the tractability of direct access for a large class of conjunctive queries. On the one hand, they recover the known tractable classes from the literature in the case of positive conjunctive queries. On the other hand, they generalise and unify known tractability results about negative conjunctive queries – that is, queries having only negated atoms. In particular, they show that the class of $\beta$-acyclic negative conjunctive queries and the class of bounded nest set width negative conjunctive queries admit tractable direct access.

8.1.2 Uncertainty and Explanations

In a JMLR (Journal of Machine Learning Research) journal article 12, Monet and co-authors study the computation of the SHAP-score. In Machine Learning, the SHAP-score is a version of the Shapley value that is used to explain the result of a learned model on a specific entity by assigning a score to every feature. While in general computing Shapley values is an intractable problem, they prove a strong positive result stating that the SHAP-score can be computed in polynomial time over deterministic and decomposable Boolean circuits. Such circuits are studied in the field of Knowledge Compilation and generalize a wide range of Boolean circuits and binary decision diagrams classes, including binary decision trees and Ordered Binary Decision Diagrams (OBDDs). They also establish the computational limits of the SHAP-score by observing that computing it over a class of Boolean models is always polynomially as hard as the model counting problem for that class. This implies that both determinism and decomposability are essential properties for the circuits that they consider. It also implies that computing SHAP-scores is intractable as well over the class of propositional formulas in DNF. Based on this negative result, they then look for the existence of fully-polynomial randomized approximation schemes (FPRAS) for computing SHAP-scores over such a class. In contrast to the model counting problem for DNF formulas, which admits an FPRAS, they prove that no such FPRAS exists for the computation of SHAP-scores. Surprisingly, this negative result holds even for the class of monotone formulas in DNF. These techniques can be further extended to prove another strong negative result: Under widely believed complexity assumptions, there is no polynomial-time algorithm that checks, given a monotone DNF formula $\phi$ and features $x,y$, whether the SHAP-score of $x$ in $\phi$ is smaller than the SHAP-score of $y$ in $\phi$.

In a SIGMOD record survey article 13, Monet and co-authors review the use of the Shapley Value framework in Database Management. Attribution scores can be applied in data management to quantify the contribution of individual items to conclusions from the data, as part of the explanation of what led to these conclusions. Since its invention in the 1950s, the Shapley value has been used for contribution measurement in many fields, from economics to law, with its latest researched applications in modern machine learning. Recent studies investigated the application of the Shapley value to database management. Their article gives an overview of recent results on the computational complexity of the Shapley value for measuring the contribution of tuples to query answers and to the extent of inconsistency with respect to integrity constraints. More specifically, the article highlights lower and upper bounds on the complexity of calculating the Shapley value, either exactly or approximately, as well as solutions for realizing the calculation in practice.

In a preprint 26, Monet and co-authors study Shapley value computation for probabilistic databases. Shapley values, originating in game theory and increasingly prominent in explainable AI, have been proposed to assess the contribution of facts in query answering over databases, along with other similar power indices such as Banzhaf values. In this work they adapt these Shapley-like scores to probabilistic settings, the objective being to compute their expected value. They show that the computations of expected Shapley values and of the expected values of Boolean functions are interreducible in polynomial time, thus obtaining the same tractability landscape. They investigate the specific tractable case where Boolean functions are represented as deterministic decomposable circuits, designing a polynomial-time algorithm for this setting. They present applications to probabilistic databases through database provenance, and an effective implementation of this algorithm within the ProvSQL system, which experimentally validates its feasibility over a standard benchmark.

8.2.1 Query Answering on Streams

In a FCT'2023 (Fundamentals of Computation Theory) conference article 16, Al Serhali and Niehren show how to evaluate stepwise hedge automata (Shas) with subhedge projection. This requires passing finite state information top-down, so they introduce the notion of downward stepwise hedge automata. Then use them to define an in-memory and a streaming evaluator with subhedge projection for SHAs. They then tune the streaming evaluator so that it can decide membership at the earliest time point. They apply our algorithms to the problem of answering regular XPath queries on XML streams. Their experiments show that subhedge projection of SHAs can indeed speed up earliest query answering on XML streams.

In an article in the Conference on Implementation and Application of Automata (CIAA'23) 21, Al Serhali and Niehren study the earliest query answering problem (EQA), which is the problem to enumerate certain query answers on streams at the earliest events. They consider EQA for regular monadic queries on hedges or nested words defined by deterministic stepwise hedge automata (dShas). They present an EQA algorithm for dShas that requires time $O\left(cm\right)$ per event, where m is the size of the automata and c the concurrency of the query. They show that our EQA algorithm runs efficiently on regular XPath queries in practice.

8.3.1 Integration and Graph Mappings with Schemas and Inference

In a PODS'2023 article 18, Boneva, Staworko and others investigate graph transformations defined using Datalog-like rules based on acyclic conjunctive two-way regular path queries (acyclic C2RPQs). They study two fundamental static analysis problems: type checking and equivalence of transformations in the presence of graph schemas. Additionally, they investigate the problem of target schema elicitation, which aims to construct a schema that closely captures all outputs of a transformation over graphs conforming to the input schema. They show that all these problems are in EXPTIME by reducing them to C2RPQ containment modulo schema; and also provide matching lower bounds.

9.1.1 Participation in other International Programs

9.2.1 Visits of international scientists

• Antoine Amarilli
  Télécom Paris. Regularly visits the team to collaborate with Paperman and with Monet. Antoine will in fact join the team in September 2024.
• Michaël Cadilhac
  Visited the team (December), from Depaul University.
• Niranjan Kumar
  Visited the team, in the contex of the Relax French-Indian "unité mixte" of CNRS.

Furthermore, the team organizes seminars that occur frequently on Fridays, and have invited numerous researchers over the year.

9.2.2 Visits to international teams

ANR JCJC KCODA

Participants: Florent Capelli [correspondent], Charles Paperman, Sylvain Salvati.

• Duration: 2021–2025
• Objectives: The aim of KCODA is to study how succinct representations can be used to efficiently solve modern optimization and AI problems that use a lot of data. We suggest using data structures from the field of compilation of knowledge that can represent large datasets succinctly by factoring certain parts while allowing efficient analysis of the represented data. The first goal of KCODA is to understand how one can efficiently solve optimization and training problems for data represented by these structures. The second goal of KCODA is to offer better integration of these techniques into the systems of database management by proposing new algorithms allowing to build factorized representations of the data responses to DB requests and by proposing encodings of these representations inside the DB.

ANR Bravas

Participants: Sylvain Salvati [correspondent].

• Duration: 2017–2023
• Coordinator: Jérôme Leroux, LaBRI, Université de Bordeaux
• Scientific Partner: LSV, ENS Cachan
• Objective: The goal of the BraVAS project is to develop a new and powerful approach to decide the reachability problems for Vector Addition Systems (VAS) extensions and to analyze their complexity. The ambition here is to crack with a single hammer (ideals over well-orders) several long-lasting open problems that have all been identified as a barrier in different areas, but that are in fact closely related when seen as reachability.

Participants: Antonio Al Serhali, Corentin Barloy, Iovka Boneva, Florent Capelli, Nicolas Crosetti, Aurélien Lemay, Mikael Monet, Joachim Niehren, Charles Paperman, Sylvain Salvati, Claire Soyez-Martin, Slawomir Staworko, Sophie Tison.

10.1.1 Scientific events: organisation

10.1.2 Scientific events: selection

10.1.3 Journal

10.1.4 Invited talks

• Monet
  Gave invited talks in the context of the Simons program he participated in.

10.1.5 Scientific expertise

• Capelli
  (Until September 2024) Member of Inria Lille CER (Commission des Emplois de Recherche).
• Tison
  Member of the scientific committee of the GDR IM.

10.1.6 Research administration

• Boneva
  Member of the steering committee of BDA (French association for research in databases).
• Boneva
  Leader of the Sustainable Development Commission at CRIStAL.
• Boneva
  Member of the Zero Carbon Network at University of Lille.
• Paperman
  Elected member of the faculty counsel of Faculté des sciences et technologies de l'université de Lille.
• Tison
  Member of the Steering Committee of Highlights of Logic, Automata, and Games.
• Tison
  Member of the Inria Lille center committee.

10.2.1 Supervision

• Al Serhali
  PhD project started 2020. On hyperstream programming. Supervised by Niehren.
• Barloy
  PhD project started 2021. On circuits and lower complexity bounds. Supervised by Paperman and Salvati.
• Soyez-Martin
  PhD project started 2020. On streaming with vectors and circuits. Supervised by Salvati and Paperman. Defended in December 2023.
• Oliver Irwin
  PhD project started 2022. On compilation and aggregation in databases. Supervised by Capelli.
• Crosetti
   PhD project started 2018. On enriching and solving linear programs with conjunctive queries. Supervised by Capelli, Niehren, and Tison. Defended in February 2023.

10.2.2 Teaching Responsibilities

• Capelli
  (Until September 2024) Responsible for Parcoursup for LEA department, Université de Lille.

10.2.3 Teaching Activities

• Boneva
  Teaches computer science in DUT Informatique of Université de Lille.
• Capelli
  (Until September 2024) Teaches computer science in the LEA department of Université de Lille for around 200h per year (Licence and Master).
• Lemay
  Teaches computer science in the LEA department of Université de Lille for around 200h per year (Licence and Master). He is also responsible for computer science and numeric correspondent for its department.
• Monet
  Monet teaches computer science as a temporary lecturer at Université de Lille and at Centrale Lille. He teaches databases courses, as well as a course on "algorithms and complexity".
• Niehren
  M2 Machine Learning, Université de Lille 2022-2023, Fondaments Theorique des Bases de Données.
• Paperman
  Teaches CS in master degrees of the computer science department, Miashs at bachelor level in the maths department and in the data science master degree.
• Salvati
  Teaches computer science for a total of around 230h per year in computer science department of Université de Lille.

10.2.4 Juries

• Capelli
  Member (Supervisor) of the PhD committee of Nicolas Crosetti, University of Lille, February 2023.
• Tison
  Member (Supervisor) of the PhD committee of Nicolas Crosetti, University of Lille, February 2023.
• Tison
  President of the HdR committee of Antoine Amrilli, Institut Polytechnique of Paris, April 2023.
• Tison
  Member of the HdR committee of Michaël Thomazo, ENS PAris, October 2023.
• Tison
  President of the PhD committee of Claire Soyez-Martin, University of Lille, December 2023.
• Tison
  President of the PhD committee of Guillaume Perution-Kihli, University of Montpellier, December 2023.

10.3.1 Interventions

• Niehren
  “École thématique CNRS”, Modélisation Formelle de Réseaux de Régulation Biologique, 4-9th June, 2023. Course on “Interprétation abstraite de réseaux de réactions chimiques pour la prédiction de knock-out de genes”.

10.3.2 Miscellaneous

• Monet
  Monet conceived and corrected one of the computer science entrance exam for X/ENS (2023).
• Paperman
  Oral examinator for X/ENS entrance exams.
• Tison
  Member of the steering committee of the mentoring circle “Femmes et sciences”, Université de Lille (since July 22).

International journals

• 12 articleM.Marcelo Arenas, P.Pablo Barceló, L.Leopoldo Bertossi and M.Mikaël Monet. On the Complexity of SHAP-Score-Based Explanations: Tractability via Knowledge Compilation and Non-Approximability Results.Journal of Machine Learning Research24632023, 1--58HALback to textback to text
• 13 articleL.Leopoldo Bertossi, B.Benny Kimelfeld, E.Ester Livshits and M.Mikaël Monet. The Shapley Value in Database Management.SIGMOD record522August 2023, 6-17HALDOIback to textback to text
• 14 articleF.Florent Capelli, N.Nicolas Crosetti, J.Joachim Niehren and J.Jan Ramon. Linear Programs with Conjunctive Database Queries.Logical Methods in Computer ScienceJanuary 2024HALback to text
• 15 articleJ.Jorge Lucero and S.Slawomir Staworko. A note on the class of languages generated by F-systems over regular languages.Information Processing Letters179January 2023, 106283HALDOIback to text

International peer-reviewed conferences

• 16 inproceedingsA.Antonio Al Serhali and J.Joachim Niehren. Subhedge Projection for Stepwise Hedge Automata.24th International Symposium on Fundamentals of Computation Theory, FCT 2023Trier, GermanySeptember 2023HALback to text
• 17 inproceedingsA.Antoine Amarilli and M.Mikaël Monet. Enumerating Regular Languages with Bounded Delay.STACSHamburg, Germany2023HALDOIback to text
• 18 inproceedingsI.Iovka Boneva, B.Benoit Groz, J.Jan Hidders, F.Filip Murlak and S.Slawomir Staworko. Static Analysis of Graph Database Transformations.Symposium on Principles of Database SystemsSeattle, United StatesJune 2023HALback to text
• 19 inproceedingsF.Florent Capelli and Y.Yann Strozecki. Geometric Amortization of Enumeration Algorithms.40th International Symposium on Theoretical Aspects of Computer Science (STACS 2023)Hamburg, GermanySchloss Dagstuhl – Leibniz-Zentrum für Informatik2023HALDOIback to text
• 20 inproceedingsJ.Joachim Niehren, C.Cédric Lhoussaine and A.Athénaïs Vaginay. Core SBML and its Formal Semantics.CMSB 2023 - 21th International Conference on Formal Methods in Systems BiologyLuxembourg, LuxembourgSeptember 2023HALback to text

Conferences without proceedings

• 21 inproceedingsA.Antonio Al Serhali and J.Joachim Niehren. Earliest Query Answering for Deterministic Stepwise Hedge Automata.27th International Conference on Implementation and Application of Automata (CIAA)famagusta, CyprusSeptember 2023HALback to text
• 22 inproceedingsJ.-S.Jean-Sébastien Guez, F.Françoise Coucheney, J.Joany Guy, M.Max Béchet, P.Pierre Fontanille, N.-E.Nour-Eddine Chihib, J.Joachim Niehren, F.François Coutte and P.Philippe Jacques. Bioinformatics Modelling and Metabolic Engineering of the Branched Chain Amino Acid Pathway for Specific Production of Microbial Biosurfactants and Biopesticides.BIOKET122Montreal, CanadaAugust 2023, 107HALDOIback to text

Edition (books, proceedings, special issue of a journal)

• 23 proceedingsJ.Jun PangJ.Joachim NiehrenComputational Methods in Systems Biology - 21st International Conference,CMSB 2023.14137Lecture Notes in Computer ScienceSpringer Nature Switzerland; SpringerSeptember 2023HALDOIback to text

Reports & preprints

• 24 miscA.Antoine Amarilli, P.Pierre Bourhis, F.Florent Capelli and M.Mikaël Monet. Ranked Enumeration for MSO on Trees via Knowledge Compilation.October 2023HALDOIback to text
• 25 miscF.Florent Capelli and O.Oliver Irwin. Direct Access for Conjunctive Queries with Negation.October 2023HALback to textback to text
• 26 miscP.Pratik Karmakar, M.Mikaël Monet, P.Pierre Senellart and S.Stéphane Bressan. Expected Shapley-Like Scores of Boolean Functions: Complexity and Applications to Probabilistic Databases.January 2024HALback to textback to text