Foundations of Data Management Workshop

Dates: July 27 -30, 2026

Location: KO2-F-152, main building of the University of Zurich

The field of data management has evolved rapidly in response to increasingly complex, large-scale, and heterogeneous data. Modern applications — ranging from machine learning to large-scale query evaluation — demand new paradigms that go beyond traditional approaches. These challenges have encouraged the Foundations of Data Management (FDM) community to engage with other disciplines, including algorithm and information theory for efficient processing, machine learning theory for integrating AI systems, formal language theory for query design, and knowledge representation for reasoning over diverse data.

This workshop primarily brings together FDM researchers from Switzerland and Chile, and further researchers from Europe, Asia, and North America to discuss the current state of FDM, reflect about the future of this research field, establish new collaborations, and strengthen existing ones.

08:30 – 08:45 Coffee

08:45 – 09:00 Welcome

09:00 – 10:15 Logical Foundations of AI

• Pablo Barceló (Pontifical Catholic U Chile): The role of activation functions in graph neural networks Show abstract We study the numerical and Boolean expressiveness of MPLang, a declarative language that captures the computation of graph neural networks (GNNs) through linear message passing and activation functions. We prove the first expressive separation between unbounded and bounded activations in the presence of linear layers: MPLang with ReLU is strictly more powerful for numerical queries than MPLang with eventually constant activation functions, e.g., truncated ReLU.
• Floris Geerts (U Antwerp): Relational learning Show abstract Relational data underlies databases, knowledge graphs, scientific repositories, enterprise records, and temporal networks, yet machine learning often treats it as graphs, tables, text, or flat features. This talk asks what it means to design neural networks natively for relational data, combining ideas from graph neural networks, graph transformers, database theory, and relational reasoning. I will focus on what such models can represent, how schema, joins, heterogeneity, and time affect learning, and where current architectures fall short.
• Michael Benedikt (U Oxford): Convergence and concentration laws for aggregates and graph neural networks Show abstract Logical convergence laws say that for each expression E in a given logic, when considering random input structures, the probability of the expression reaches a limit. The canonical example is the zero-one law for first-order logic, due independently to Fagin and Glebskii et al. In this talk we discuss convergence laws for real-valued logics that include aggregate operations, like average and sum. We focus in this talk on the case where the input is a graph, possibly with data annotations on nodes, and the notion of random input is a classical random graph model with (Erdos-Renyi random graphs). We discuss some applications to the analysis of graph neural networks.
  
  The talk will include joint work with:
  Sam Adam-Day, Ismail Ceylan, and Ben Finkelstein, Neurips '24
  Sam Adam-Day and Alberto Larrauri, LICS '25
  Maksim Zuhkovski, https://arxiv.org/abs/2602.17752
  It also relates to prior work by Sam Aday-Day, Ismail Ceylan, Theodor Ilian, Neurip '23 and by Vera Koponen, Theoretical Computer Science 2020.

10:45 – 12:00 AI for Data Management

• Juan Reutter (Pontifical Catholic U Chile): Foundational models for knowledge graphs Show abstract Graph Foundation Models (GFMs) aim to solve tasks over previously unseen graphs. Ideally, they would be pre-trained on massive amounts of data and then deployed on arbitrary graphs across a wide range of downstream tasks. Yet, compared with foundation models for text and images, GFMs are still in their formative stages. In this talk, we will discuss some of the key challenges that must be addressed to build better GFMs. We will then focus on one such challenge: finding (or learning) a common vocabulary across knowledge graphs from different domains, where we have recently made substantial progress. Looking ahead, the techniques used to learn these common vocabularies rely on querying and transforming graphs, creating an opportunity for database researchers to improve the performance of GFMs and make their training much more efficient.
• Omer Abramovich (Tel Aviv U): Shapley-based fact attribution in databases Show abstract Quantifying the contribution of database facts to query results has emerged as a principled approach to explain query outputs. The Shapley and Banzhaf values, originating in cooperative game theory, provide natural measures of fact attribution, yet their efficient computation for database queries is challenging.
  In this talk, I will provide an intuitive overview of how these values capture fact attribution in databases, and present algorithmic techniques that make their computation scalable in practice. Building on theoretical insights, our methods achieve over four orders of magnitude speedup over prior work, and constitute the first practical algorithms for aggregate queries.
  This talk is based on joint work with Daniel Deutch, Ahmet Kara, and Dan Olteanu.
• Daniel Deutch (Tel Aviv U): Counterfactual explanations for classifiers via denial constraints Show abstract In the realm of Explainable AI, classification results are often explained via counterfactuals (CFs for short), which are (ideally small) perturbations to an instance that lead to a change of classification label. Such CFs may serve as explanations for the prediction, pinpointing the features that were important. Existing explainability solutions typically aim at minimizing the distance of CFs from the original instance so that they are specific to it and/or maximizing the diversity of CFs to cover multiple facets of the reasons underlying the prediction. Here, we note that in pursuing these aims, state-of-the-art explainability solutions may (and often do) yield counterfactual explanations that do not correspond to realistic instances. This limits their applicability and usefulness in practice.
  
  To remedy this, we combine ideas from Explainable AI with ideas from data management. Specifically, we capture realism of CFs via logical constraints that hold with respect to a dataset of examples (e.g. training set); the class of such constraints that we focus on is that of denial constraints, extensively studied in the context of relational databases. Algorithmically, we then combine explainable AI solutions to yield CFs, with ideas from data cleaning that we adapt to this unique setting, to transform CFs into realistic ones. Extensive experiments across four datasets validate that our solutions achieve realism with relatively minor compromise in terms of distance and diversity. They further validate that dedicated optimizations that we have developed to speed up the search for CFs are indeed effective.
  
  To appear in KDD '26.

12:00 – 13:00 Lunch break

13:00 – 14:40 Query Maintenance

• Eden Chmielewski (U Zurich): Maintaining queries under updates using heavy-light partitioning of the input relations Show abstract In this talk, I will overview a recent incremental view maintenance approach for conjunctive queries under inserts and deletes to the input database. Its update time matches or improves the best update time reported in prior work, while allowing the constant-delay enumeration of the tuples in the query output.
  Whereas prior approaches are manually tailored to each of a handful of queries, our approach generalizes to arbitrary queries. It combines three techniques: delta queries, trees of materialized views, and heavy-light data partitioning. The overall update time incurred by our approach is characterized by the maintenance width of the query, a measure parameterized by the heavy-light threshold for data partitioning.
  This talk is based on joint work with Mahmoud Abo Khamis, Andrei Draghici, Ahmet Kara, and Dan Olteanu.
• Qichen Wang (NTU Singapore): A fine-grained view of the hardness of conjunctive query maintenance Show abstract Maintaining conjunctive queries under updates is a classical problem in database systems. Given a fixed query and a database undergoing tuple insertions and deletions, the goal is to maintain query results much faster than by recomputing them from scratch. This problem is central to materialized views, streaming analytics, continuous query processing, and real-time decision-making. However, even simple join queries can be hard to maintain under fully dynamic updates.
  
  In this talk, I will present a fine-grained view of conjunctive query maintenance that goes beyond worst-case lower bounds. The first perspective is update-dependent: real-world update sequences are often much more structured than adversarial sequences. I will discuss how insertion-only, FIFO, and mixed update patterns can make otherwise hard queries easier to maintain, and how notions such as enclosureness and update-dependent structure lead to optimal maintenance algorithms. The second perspective is query-dependent: the structure of the query itself can be analyzed through finer parameters, such as height and dimension, leading to tighter lower bounds for dynamic query evaluation. Together, I will also discuss how this theory can guide the design and optimization of query maintenance pipelines.
• Ahmet Kara (OTH Regensburg): Query maintenance over static and dynamic relations Show abstract In this talk, I will discuss how to evaluate conjunctive queries efficiently when some input relations are static, while others are dynamic and subject to inserts and deletes. The starting point is the observation that many real-world workloads naturally distinguish between data that changes frequently and data that remains fixed for long periods of time. Exploiting this distinction can lead to substantially more efficient query evaluation.
  I will present a syntactic characterisation of conjunctive queries that admit constant update time and constant enumeration delay after polynomial-time preprocessing in the mixed static-dynamic setting. I will also discuss the complete sub-class of such queries that can be handled after only linear-time preprocessing together with conditional lower bounds for queries outside this class based on the Boolean Matrix-Matrix Multiplication and Online Matrix-Vector Multiplication conjectures.
  The algorithmic approach rewrites a query into a hierarchy of views, which are then maintained incrementally under updates. I will explain how this framework can be refined to take into account more practical information, such as how frequently different relations are updated and how selective the join and selection conditions are.
  Finally, I will present experimental results using the open-source F-IVM engine. These results show that exploiting static-dynamic information and update frequencies can improve the throughput of maintaining TPC-H-like queries by up to two orders of magnitude compared with state-of-the-art approaches that do not use this information.
  This talk is based on joint work with Milos Nikolic, Dan Olteanu, and Haozhe Zhang
• Andrei Draghici (U Zurich): The role of semirings in incremental view maintenance Show abstract We study the problem of incremental view maintenance (IVM) under inserts to K-databases, where K is a commutative semiring without additive inverse. The key observation put forward is that the complexity of the IVM problem depends fundamentally on the underlying semiring.
  For Boolean-databases the tractable query classes are well-understood. On the one hand, tractable maintenance under inserts-only can only be achieved for the free-connex alpha-acyclic queries. On the other hand, tractable maintenance under both inserts and deletes can only be achieved for the q-hierarchical queries.
  In case of inserts-only over K-databases we show that a query can be maintained with amortized constant update time and constant enumeration delay if and only if it is alpha-acyclic p-hierarchical, for a large class of semirings K. All lower bounds assume queries with no self-joins.
  This is joint work with Eden Chmielewski, Dan Olteanu, and Haozhe Zhang.

15:10 – 16:00 Data and AI for Social Good

• Ioana Manolescu (INRIA and Ecole Polytechnique): Knowledge is power: Fighting disinformation with data and AI Show abstract The harmful impact of disinformation on society and individuals is large and growing. Reliable knowledge about the world, and about our information space, can be leveraged to diminish this negative impact. I will present research carried in my team to leverage high-quality reference information. In the StatCheck project, transferred to RadioFrance, we focus on making high-quality statistics available and easy to search. We show that naive use of LLMs is not only expensive, it also fails to handle information searches in very large statistic corpora. Instead, we present a frugal and higher-quality approach, that follows the specific nature of statistic datasets to better answer user queries over such corpora. Then, I will present an efficient approach, based on reinforcement learning, which identifies within a given Web site all the files having certain user-specified types. This allows in particular to find all statistic files within varied-structure web sites, to be used as a basis for fact-checking.
  Joint work with Oana Balalau, Helena Galhardas (INESC-ID Portugal), Antoine Gauquier and Pierre Senellart (ENS and Inria), and other colleagues.
• Michael Theologitis (U Washington): The Promise of AI agents for computational journalism and fact-checking Show abstract On a daily basis, modern investigative journalists must make sense of data at scale, including messy relational databases, document collections, and public reports. They are required to understand this data well-enough to make sharp connections and uncover hidden truths — for example, a policymaker who has acquired immense wealth but registered it under the name of a relative, or multiple companies receiving government contracts despite all being linked to the same beneficial owner. Making these connections across different data sources traditionally requires a high level of technical expertise (e.g., SQL, Python, etc.) which is not part of the skill set of many journalists. At the same time, modern generative AI approaches promise to “revolutionize journalism in a way analogous to the emergence of the internet” (Diakopoulos et al., 2024), offering journalists the possibility of delegating substantial parts of complex data investigations to computational AI systems. In this talk, we briefly review the history of computational methods in journalism up to the current era of AI agents. We then discuss the foundational importance of fact-checking in investigative journalism through the perspective of an investigative journalist at The Seattle Times. Next, we introduce our fact-verification benchmark, ClaimDB, where the evidence for judging claims is scattered across large structured data mirroring the challenges journalists face in real-life investigations. Finally, we present an agentic approach based on SQL agents and evaluate whether more than twenty state-of-the-art LLMs are capable of investigating the veracity of factual claims over these large databases.

17:00 – 19:00 City Tour starting at Central (Polybahn station)

19:00 - 23:00 Dinner at Zunfthaus zur Schmiden

08:45 – 09:00 Coffee

09:00 – 09:50 Vectors and Tensors

• Yannis Chronis (ETH Zurich): Semantic search in databases: Opportunities and challenges Show abstract Semantic search has been popularized in databases over the past couple of years due to the advancements of embedding models. Representing data as embeddings captures meaning and allows information to be retrieved by similarity rather than exact matches. As a result, databases can support queries that jointly process structured and unstructured data and understand user intent rather than literal predicates. Realizing this opportunity also raises challenges: integrating vector indices into query engines, developing optimization techniques that reason about both semantic and relational operations, and achieving high, stable, and declarative recall to ensure consistency and ease of use. This talk discusses the state of the art in semantic search for databases, along with the open challenges and opportunities the field presents.
• Kyle Deeds (U Boston): Fraternal twins: The long history of tables and tensors Show abstract Going all the way back to 1870, researchers have recognized the uncanny similarity between relations and tensors. This similarity has resulted in an intermittent stream of work that attempts to bridge the two fields in both systems and theory. The database field, both theory and systems, has embarked on an inexorable march to capture array workloads within database frameworks and systems, e.g. in-database ML and array databases. At the same time, there have been sporadic attempts by the HPC community to show that tensor algorithms and hardware could revolutionize graph and relational workloads, e.g. GraphBLAS and the TQP system. In addition to importing problems, practitioners in both communities have slowly imported techniques from the other, e.g. COO formats and GPU databases. In this talk, I will first attempt to provide a history of the interaction between these fields, then I will establish a basic correspondence between their data models with an eye towards the subtle complexity that arises in the case of sparse tensors.

10:00 – 10:50 Worst-case Optimal Query Evaluation

• Gonzalo Navarro (U Chile): Compact data structures for worst-case-optimal joins and beyond Show abstract I'll revisit our recent results on implementing worst-case-optimal (wco) joins in graph databases within reduced space. Space usage has been a barrier to a wider adoption of in-memory WCO algorithms. I'll show how the use of compact data structures allows implementing them with no extra space over the plain representation of the data, and even within compressed overall space, allowing in addition to outperform classic implementations in time. I will also show how compact data structures provide further functionality that is hard to implement on classic representations, which yields from improved WCO join strategies to the incorporation of multimodal data into the semantics-agnostic graph join algorithms, retaining worst-case optimality. We show applications of these ideas to temporal graphs, graphs with similarity metrics, graphs with topology information, and others.
• Moritz Hoffmann (Materialize Zurich): Datatoad: a worst-case optimal Datalog Show abstract Datatoad is an interpreted, columnar Datalog engine built on worst-case optimal joins, but in a way that bounds the end-to-end time of the entire semi-naive process. The total time matches the worst-case optimal bound of evaluating each of the rules on their final relation sizes.
  This approach follows Ammar et al.'s columnar implementation of GenericJoin for streaming relations.
  Having leaned into columnar organization of both data and logic, the API extends to general relations, including antijoins, logic-backed relations, and unmaterialized views. The result is a DatalogZ (with integers) in which omni-directional logic atoms participate in WCO routines and can meaningfully improve efficiency by doing so.
  Datatoad is publicly available and you can try it online.

11:00 – 12:15 Graph Databases

• Domagoj Vrgoc (Pontifical Catholic U Chile): Theory in practice (a personal experience) Show abstract In this talk, I will discuss the painful process of converting theoretically optimal algorithms into production-ready code, grounding the discussion in concrete systems we built at IMFD Chile. In particular, I will talk about the graph engine MillenniumDB and the perils of worst-case-optimal algorithms and path query processing. I will also discuss ReMatch, a regex engine implementing the document spanners framework, and finish with our latest projects. While originally built for educational purposes, these projects made me realize that I had no idea (and still don't) about transactional query processing (both in the classical on-premises setup and in the cloud). The talk aims to present new research directions that are wide open in each of these settings, particularly in terms of theory.
• Running Hou (U Zurich): Precise search for path enumeration Show abstract Enumerating all paths that satisfy a regular path query (RPQ) with concatenations within the scope of a Kleene star operator is an open problem for graph database systems (GDBS). To answer such queries a search must be performed to determine all paths, and the paths must be extracted from the search result. Doing this correctly and efficiently is challenging since redundant computations must be pruned, exactly the paths that conform to the RPQ must be reported, and cycles must be dealt with.
  In this talk, I will present an automaton-guided precise search that addresses these challenges. It only traverses edges conforming to the RPQ and constructs a path graph that compactly represents the paths. To curb the search and result space, we show how to merge states and perform cycle deduplication. I will describe the implementation of our solution in the Kùzu graph database system and present experimental results over existing methods across diverse graph workloads.
• Mahmoud Abo Khamis (RelationalAI Berkeley): Conjunctive regular path queries: Recent advances on evaluation and future directions Show abstract Graph databases are increasingly prevalent, driven by applications such as knowledge graphs, social networks, and the semantic web. Regular Path Queries (RPQs) and their conjunctive counterparts (CRPQs) are at the heart of graph query languages used to query those databases, e.g. SPARQL and Cypher. This motivates the need for the database theory community to rise up to the occasion and develop new algorithms to optimize and efficiently evaluate these queries.
  
  In this talk, we present some of the recent results on the evaluation of RPQs and acyclic CRPQs. These results give output-sensitive algorithms, i.e., whose runtimes depend not only on the input size but also on the output size. We show how to leverage recent advances on the evaluation of Conjunctive Queries (CQs) to transfer both techniques and lower bounds to the CRPQ world. This perspective opens the door to many interesting questions; perhaps the most compelling one is whether such techniques can be generalized further to any recursive Datalog program.

12:15 – 13:15 Lunch break

13:15 – 14:30 Robust, Adaptive, Factorized Query Evaluation

• Dan Suciu (U Washington): On the space-time complexity of sum-product queries Show abstract While the time complexity of query evaluation is well understood, the space complexity becomes increasingly more important, because modern hardware accelerators are typically memory bound, not compute bound. In this talk I will discuss the space-time tradeoff for query evaluation. Most of the talk will focus on Boolean CQs (or, more generally, scalar sum-product queries), where the output is small, and the challenge is to reduce the amount of intermediate memory needed for the evaluation. I will describe several classes of space-efficient query evaluation algorithms (tree decompositions, pseudo-trees, various caching mechanisms, and their combinations). These algorithms exhibit an interesting space-time tradeoff (reminiscent of the AT^2 tradeoff in VLSI design): we can reduce the space consumption at the cost of increasing the time complexity. I will end the talk with a discussion of the challenges faced by theory in capturing the complexity of query evaluation on modern hardware accelerators.
  Joint work with Kyle Deeds, Timo Camillo Merkl, Reinhard Pichler
• Altan Birler (TU Munich): Robust join optimization in SQL query processing Show abstract We are building Umbra, a query engine that aims to support the full SQL standard with consistently high performance. Joins are an essential part of SQL, and unfortunately, finding the optimal execution plan for a join query is NP-hard. Systems must optimize queries containing potentially thousands of joins, so they must either exploit the shape of the query or rely on heuristics during plan search. This is further complicated by SQL's support for join types beyond inner joins, namely left/right outer joins, full outer joins, semi joins, anti joins, mark joins, group joins, single joins, and lateral joins.
  In this talk, I will discuss some progress in handling a wider variety of join queries gracefully during optimization, as well as remaining open problems.
• Clément Rouvroy (ENS PSL): FRANTIC: Adaptive and Factorized Query Processing Show abstract In this talk, I will briefly overview FRANTIC, our query processor that combines three main components: worst-case optimal joins, factorization, and adaptivity via reinforcement learning. FRANTIC does not rely on a prior query optimization step, instead it just starts processing fragments of the input database right away and on the fly adjusts its execution strategy, called variable order, to lower the amount of processing needed to compute the query result. Yet it can quickl finds a good, even (nearly) optimal variable order.
  I will discuss the design space of FRANTIC and explain the decisions behind its current design. One technical novelty of FRANTIC at the intersection of factorized query processing and reinforcement learning is a new Upper Confidence Bound approach that exploits the structure of the space of possible factorized orders over the query variables.
  Our preliminary experiments with several workloads of acyclic and cyclic queries show that FRANTIC can outperform a variety of query engines by up to two orders of magnitude on commodity hardware.
  This talk is based on joint work with Joe Hellerstein, Christoph Mayer, and Dan Olteanu.

14:40 – 16:20 Cardinality Estimation

• Christoph Koch (EPFL): Query Result size estimation for Datalog Show abstract EDB-bounded datalog is a framework for deriving upper bounds on intermediate result sizes and the asymptotic complexity of recursive queries in datalog. I present an algorithm that, given an arbitrary datalog program, constructs an EDB-bounded datalog program in which every rule is adorned with a (non-recursive) conjunctive query that subsumes the result of the rule, thus acting as an upper bound. From such adornments, we obtain a notion of width based on (integral or fractional) edge-cover widths. As a consequence, we obtain, for every IDB predicate, worst-case upper bounds on their sizes, which are polynomial in the input data size, given a fixed program structure. We also obtain fixed-parameter tractable, output-sensitive asymptotic complexity bounds for evaluating datalog programs.
  Join work with Christian Fattebert, Zhekai Jiang, Reinhard Pichler, and Qichen Wang.
• Chris Mayer (U Zurich): CorrBound: Cardinality estimation accounting for inter- and intra-relation correlations Show abstract Cardinality estimation is a critical component of query optimization and resource allocation, yet state-of-the-art estimators often ignore correlations between join columns across and within relations, leading to poor estimates.
  Our recent work, CorrBound, captures such inter- and intra-relation correlations by extending the information-theoretic framework we developed in our previous work on LpBound. A key challenge is that the extension gives rise to the Cardinality Tensor, a high-dimensional structure that grows with the number of join columns. To keep estimation time practical, CorrBound employs a low-rank decomposition to compress the Cardinality Tensor while preserving accuracy. Experimental results show significant improvements in accuracy compared to a wide range of estimators.
  This talk is based on joint work with Mahmoud Abo Khamis, Kyle Deeds, Dan Olteanu, Dan Suciu, and Haozhe Zhang.
• Tiemo Bang (Microsoft GSL): Industrial optimism on pessimistic research Show abstract Traditional Cardinality Estimation (CE) remains a notorious Achilles' heel in database systems. While statistical approaches perform well on average, their rare but catastrophic misestimations violently cascade through the system. In cloud environments like Microsoft’s Fabric Warehouse, these errors derail not just query plan selection, but also highly sensitive resource allocation. Pessimistic Cardinality Estimation (PCE) is emerging as a powerful remedy by shifting the paradigm from best-effort guessing to hard, mathematical guarantees.
  This talk reveals how Microsoft is adopting PCE not as a replacement for legacy estimators, but as a definitive mathematical guardrail to achieve regression-free improvement. Our core message is a call to action: to fully tame catastrophic misestimation in production, the industry urgently needs more research on pessimism. We will outline the critical, unsolved research problems currently blocking wider adoption—from remaining gaps in cardinality upper bound estimation (such as string support) to the lack of rigorous lower bounds, and the necessity of expanding pessimism to other estimation problems in the DBMS.
• Mihail Stoian (U Technology Nuremberg): xBound: Join size lower bounds Show abstract Despite decades of research, cardinality estimation remains the optimizer’s Achilles heel, with industrial-strength systems exhibiting a systemic tendency toward underestimation. At cloud scale, this is a severe production vulnerability: in Microsoft’s Fabric Data Warehouse (DW), a mere 0.05% of extreme underestimates account for 95% of all CPU under-allocation, causing preventable slowdowns for thousands of queries daily. Yet recent theoretical work on provable upper bounds only corrects overestimation, leaving the more harmful problem of underestimation unaddressed.
  We introduce xBound, the first theoretical framework for computing provable join size lower bounds. By clipping the optimizer’s estimates from below, xBound offers strict safety nets demanded by production systems, using only a handful of lightweight base table statistics.
  We demonstrate xBound’s practical impact on Fabric DW: on the StackOverflow-CEB benchmark, it corrects 23.6% of Fabric DW’s underestimates, yielding end-to-end query speedups of up to 20.1x, demonstrating that even a first step toward provable lower bounds can deliver meaningful production gains and motivating the community to further pursue this critical, open direction.
  xBound's demo is available online (VLDB'26 Demo).

17:00 – 18:30 Hike (or alternatively public transportation) to Uto Kulm

18:30 - 23:00 Apéro (until 19:30) followed by dinner at Uto Kulm

Day trip: Reflect & hike

Hike at the Rigi mountain, followed by a boat trip to Luzern

Highlights of the trip:

• Rigi Kulm
• hike to Kaltbad
• cog-wheel train to Vitznau
• boat on Vierwaldstättersee to Luzern
• city of Luzern

Closer to the date, we will consider further variants if the weather does not favour the above plan.

08:45 – 09:00 Coffee

09:00 – 09:30 Where we are and where we go from here

• Wim Martens (Bayreuth U): Future query languages Show abstract Database query languages are perhaps the greatest success story of declarative programming languages. By separating intent from execution, declarative languages allow agents or programmers to focus on getting the correct results instead of writing efficient programs to obtain them. However, today’s query languages are data sublanguages: they focus on storage and retrieval of data from databases.
  
  The aim of this talk is to explore how query languages can evolve into full-fledged programming languages, which is in line with Jim Gray’s dream of automatic programming. This talk has two main goals. First, I want to draw our attention to the goal of automatic programming, i.e., expanding the scope of query languages to full-fledges programming languages. Second, I will argue how we need to revisit some basic principles of query languages if we want to achieve this goal, notably nulls and bag semantics.
  
  The Rel language, developed at RelationalAI, is an example of how database principles can be used to expand the scope of query languages beyond the sublanguage paradigm. It is based on set semantics and uses no nulls.

09:45 – 10:45 Discussion and outlook on Logical Foundations of AI

                            Discussion led by the presenters in the AI sessions

11:00 – 12:00 Discussion and outlook on Query Optimization

                            Discussion led by the presenters in the two query optimization sessions

12:00 – 13:00 Lunch break

13:00 – 14:00 Discussion and outlook on Query Evaluation and Maintenance

                            Discussion led by the presenters in the two query evaluation and the query maintennce sessions

14:00 – 14:05 Closing Remarks

15:15 – 16:30 Boat from Bürkliplatz to Au Peninsula

18:00 Dinner at the restaurant on Au Peninsula

This is the list of in-person participants. The workshop has a hybrid format and also includes online participation.

• Mahmoud Abo Khamis, RelationalAI Berkeley
• Omer Abramovich, Tel Aviv University
• Marcelo Arenas, Pontifical Catholic University of Chile
• Tiemo Bang, Microsoft GSL Barcelona
• Pablo Barceló, Pontifical Catholic University of Chile
• Michael Benedikt, University of Oxford
• Altan Birler, Technical University of Munich
• Eden Chmielewski, University of Zurich
• Yannis Chronis, ETH Zurich
• Kyle Deeds, University of Boston
• Daniel Deutch, Tel Aviv University
• Andrei Draghici, University of Zurich
• Floris Geerts, University of Antwerp
• Egor Gorbachev, ETH Zurich
• Niko Göbel, RelationalAI Zurich
• Yuchen He, University of Zurich
• Running Hou, University of Zurich
• Moritz Hoffmann, Materialize Inc Zurich
• Ahmet Kara, OTH Regensburg
• Christoph Koch, EPFL
• Ioana Manolescu, INRIA and LIX/Ecole Polytechnique
• Wim Martens, Bayreuth University
• Christoph Mayer, University of Zurich
• Gonzalo Navarro, University of Chile
• Dan Olteanu, University of Zurich
• Juan Reutter, Pontifical Catholic University of Chile
• Miguel Angel Romero Orth, Pontifical Catholic University of Chile
• Clément Rouvroy, L'École Normale Supérieure, Paris Sciences et Lettres
• Mihail Stoian, University of Technology Nuremberg
• Dan Suciu, University of Washington
• Michael Theologitis, University of Washington
• Domagoj Vrgoc, Pontifical Catholic University of Chile
• Qichen Wang, Nanyang Technological University
• Haozhe Zhang, University of Zurich

Dan Olteanu would like to thank the Swiss National Science Foundation (SNSF) Scientific Exchange program and UZH Foundation for partially supporting this workshop and Denise Gloor for assistance with the logistics.