Tenghao Huang, Kinjal Basu, Ibrahim Abdelaziz, Pavan Kapanipathi, Jonathan May, Muhao Chen. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Kinjal Basu, Ibrahim Abdelaziz, Kiran Kate, Mayank Agarwal, Maxwell Crouse, Yara Rizk, Kelsey Bradford, Asim Munawar, Sadhana Kumaravel, Saurabh Goyal, Xin Wang, Luis A. Lastras, Pavan Kapanipathi. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Enterprises have a growing need to identify relevant tables in data lakes; e.g. tables that are unionable, joinable, or subsets of each other. Tabular neural models can be help-ful for such data discovery tasks. In this paper, we present TabSketchFM, a neural tabular model for data discovery over data lakes. First, we propose novel pre-training: a sketch-based approach to enhance the effectiveness of data discovery in neural tabular models. Second, we finetune the pretrained model for identifying unionable, joinable, and subset table pairs and show significant improvement over previous tabular neural models. Third, we present a detailed ablation study to highlight which sketches are crucial for which tasks. Fourth, we use these finetuned models to perform table search; i.e., given a query table, find other tables in a corpus that are unionable, joinable, or that are subsets of the query. Our results demonstrate significant improvements in F1 scores for search compared to state-of-the-art techniques. Finally, we show significant transfer across datasets and tasks establishing that our model can generalize across different tasks and over different data lakes.

Kinjal Basu, Ibrahim Abdelaziz, Subhajit Chaudhury, Soham Dan, Maxwell Crouse, Asim Munawar, Vernon Austel, Sadhana Kumaravel, Vinod Muthusamy, Pavan Kapanipathi, Luis Lastras. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Careful choice of feature transformations in a dataset can help predictive model performance, data understanding and data exploration. However, finding useful features is a challenge, and while recent Automated Machine Learning (AutoML) systems provide some limited automation for feature engineering or data exploration, it is still mostly done by humans. We demonstrate a system called SemFORMS (Semantic Transforms), which attempts to mine useful expressions for a dataset from access to a repository of code that may target the same dataset/similar dataset. In many enterprises, numerous data scientists often work on the same or similar datasets, but are largely unaware of each other's work. SemFORMS finds appropriate code from such a repository, and normalizes the code to be an actionable transform that can prepended into any AutoML pipeline. We demonstrate SemFORMS operating over example datasets from the OpenML benchmarks where it sometimes leads to significant improvements in AutoML performance.

Using reinforcement learning for automated theorem proving has recently received much attention. Current approaches use representations of logical statements that often rely on the names used in these statements and, as a result, the models are generally not transferable from one domain to another. The size of these representations and whether to include the whole theory or part of it are other important decisions that affect the performance of these approaches as well as their runtime efficiency. In this paper, we present NIAGRA; an ensemble Name InvAriant Graph RepresentAtion. NIAGRA addresses this problem by using 1) improved Graph Neural Networks for learning name-invariant formula representations that is tailored for their unique characteristics and 2) an efficient ensemble approach for automated theorem proving. Our experimental evaluation shows state-of-the-art performance on multiple datasets from different domains with improvements up to 10% compared to the best learning-based approaches. Furthermore, transfer learning experiments show that our approach significantly outperforms other learning-based approaches by up to 28%.

Traditional automated theorem provers have relied on manually tuned heuristics to guide how they perform proof search. Recently, however, there has been a surge of interest in the design of learning mechanisms that can be integrated into theorem provers to improve their performance automatically. In this work, we describe TRAIL (Trial Reasoner for AI that Learns), a deep learning-based approach to theorem proving that characterizes core elements of saturation-based theorem proving within a neural framework. TRAIL leverages (a) an effective graph neural network for representing logical formulas, (b) a novel neural representation of the state of a saturation-based theorem prover in terms of processed clauses and available actions, and (c) a novel representation of the inference selection process as an attention-based action policy. We show through a systematic analysis that these components allow TRAIL to significantly outperform previous reinforcement learning-based theorem provers on two standard benchmark datasets (up to 36% more theorems proved). In addition, to the best of our knowledge, TRAIL is the first reinforcement learning-based approach to exceed the performance of a state-of-the-art traditional theorem prover on a standard theorem proving benchmark (solving up to 17% more theorems).

Data preparation is a crucial first step to any data analysis problem. This task is largely manual, performed by a person familiar with the data domain. DataRinse is a system designed to extract relevant transforms from large scale static analysis of repositories of code. Our motivation is that in any large enterprise, multiple personas such as data engineers and data scientists work on similar datasets. However, sharing or re-using that code is not obvious and difficult to execute. In this paper, we demonstrate DataRinse to handle data preparation, such that the system recommends code designed to help with the preparation of a column for data analysis more generally. We show that DataRinse does not simply shard expressions observed in code but also uses analysis to group expressions applied to the same field such that related transforms appear coherently to a user. It is a human-in-the-loop system where the users select relevant code snippets produced by DataRinse to apply on their dataset.

Code understanding is an increasingly important application of Artificial Intelligence. A fundamental aspect of understanding code is understanding text about code, e.g., documentation and forum discussions. Pre-trained language models (e.g., BERT) are a popular approach for various NLP tasks, and there are now a variety of benchmarks, such as GLUE, to help improve the development of such models for natural language understanding. However, little is known about how well such models work on textual artifacts about code, and we are unaware of any systematic set of downstream tasks for such an evaluation. In this paper, we derive a set of benchmarks (BLANCA - Benchmarks for LANguage models on Coding Artifacts) that assess code understanding based on tasks such as predicting the best answer to a question in a forum post, finding related forum posts, or predicting classes related in a hierarchy from class documentation. We evaluate performance of current state-of-the-art language models on these tasks and show that there is significant improvement on each task from fine tuning. We also show that multi-task training over BLANCA tasks help build better language models for code understanding.

Prithviraj Sen, Breno William Carvalho, Ibrahim Abdelaziz, Pavan Kapanipathi, Salim Roukos, Alexander Gray. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

AutoML systems build machine learning models automatically by performing a search over valid data transformations and learners, along with hyper-parameter optimization for each learner. Many AutoML systems use meta-learning to guide search for optimal pipelines. In this work, we present a novel meta-learning system called KGpip which (1) builds a database of datasets and corresponding pipelines by mining thousands of scripts with program analysis, (2) uses dataset embeddings to find similar datasets in the database based on its content instead of metadata-based features, (3) models AutoML pipeline creation as a graph generation problem, to succinctly characterize the diverse pipelines seen for a single dataset. KGpip's meta-learning is a sub-component for AutoML systems. We demonstrate this by integrating KGpip with two AutoML systems. Our comprehensive evaluation using 121 datasets, including those used by the state-of-the-art systems, shows that KGpip significantly outperforms these systems.

Knowledge Base Question Answering (KBQA) is a task where existing techniques have faced significant challenges, such as the need for complex question understanding, reasoning, and large training datasets. In this work, we demonstrate Deep Thinking Question Answering (DTQA), a semantic parsing and reasoning-based KBQA system. DTQA (1) integrates multiple, reusable modules that are trained specifically for their individual tasks (e.g. semantic parsing, entity linking, and relationship linking), eliminating the need for end-to-end KBQA training data; (2) leverages semantic parsing and a reasoner for improved question understanding. DTQA is a system of systems that achieves state-of-the-art performance on two popular KBQA datasets.

Automated theorem provers have traditionally relied on manually tuned heuristics to guide how they perform proof search. Deep reinforcement learning has been proposed as a way to obviate the need for such heuristics, however, its deployment in automated theorem proving remains a challenge. In this paper we introduce TRAIL, a system that applies deep reinforcement learning to saturation-based theorem proving. TRAIL leverages (a) a novel neural representation of the state of a theorem prover and (b) a novel characterization of the inference selection process in terms of an attention-based action policy. We show through systematic analysis that these mechanisms allow TRAIL to significantly outperform previous reinforcement-learning-based theorem provers on two benchmark datasets for first-order logic automated theorem proving (proving around 15% more theorems).

Analogy is core to human cognition. It allows us to solve problems based on prior experience, it governs the way we conceptualize new information, and it even influences our visual perception. The importance of analogy to humans has made it an active area of research in the broader field of artificial intelligence, resulting in data-efficient models that learn and reason in human-like ways. While cognitive perspectives of analogy and deep learning have generally been studied independently of one another, the integration of the two lines of research is a promising step towards more robust and efficient learning techniques. As part of a growing body of research on such an integration, we introduce the Analogical Matching Network: a neural architecture that learns to produce analogies between structured, symbolic representations that are largely consistent with the principles of Structure-Mapping Theory.

Knowledge graphs have been proven extremely useful in powering diverse applications in semantic search and natural language understanding. In this work, we present GraphGen4Code, a toolkit to build code knowledge graphs that can similarly power various applications such as program search, code understanding, bug detection, and code automation. GraphGen4Code uses generic techniques to capture code semantics with the key nodes in the graph representing classes, functions and methods. Edges indicate function usage (e.g., how data flows through function calls, as derived from program analysis of real code), and documentation about functions (e.g., code documentation, usage documentation, or forum discussions such as StackOverflow). Our toolkit uses named graphs in RDF to model graphs per program, or can output graphs as JSON. We show the scalability of the toolkit by applying it to 1.3 million Python files drawn from GitHub, 2,300 Python modules, and 47 million forum posts. This results in an integrated code graph with over 2 billion triples. We make the toolkit to build such graphs as well as the sample extraction of the 2 billion triples graph publicly available to the community for use.

Textual entailment is a fundamental task in natural language processing. Most approaches for solving this problem use only the textual content present in training data. A few approaches have shown that information from external knowledge sources like knowledge graphs (KGs) can add value, in addition to the textual content, by providing background knowledge that may be critical for a task. However, the proposed models do not fully exploit the information in the usually large and noisy KGs, and it is not clear how it can be effectively encoded to be useful for entailment. We present an approach that complements text-based entailment models with information from KGs by (1) using Personalized PageRank to generate contextual subgraphs with reduced noise and (2) encoding these subgraphs using graph convolutional networks to capture the structural and semantic information in KGs. We evaluate our approach on multiple textual entailment datasets and show that the use of external knowledge helps the model to be robust and improves prediction accuracy. This is particularly evident in the challenging BreakingNLI dataset, where we see an absolute improvement of 5-20% over multiple text-based entailment models.

3D hand shape and pose estimation from a single depth map is a new and challenging computer vision problem with many applications. The state-of-the-art methods directly regress 3D hand meshes from 2D depth images via 2D convolutional neural networks, which leads to artefacts in the estimations due to perspective distortions in the images. In contrast, we propose a novel architecture with 3D convolutions trained in a weakly-supervised manner. The input to our method is a 3D voxelized depth map, and we rely on two hand shape representations. The first one is the 3D voxelized grid of the shape which is accurate but does not preserve the mesh topology and the number of mesh vertices. The second representation is the 3D hand surface which is less accurate but does not suffer from the limitations of the first representation. We combine the advantages of these two representations by registering the hand surface to the voxelized hand shape. In the extensive experiments, the proposed approach improves over the state of the art by47.8% on the SynHand5M dataset. Moreover, our augmentation policy for voxelized depth maps further enhances the accuracy of 3D hand pose estimation on real data. Our method produces visually more reasonable and realistic hand shapes on NYU and BigHand2.2M datasets compared to the existing approaches.

Natural Language Inference (NLI) is fundamental to many Natural Language Processing (NLP) applications including semantic search and question answering. The NLI problem has gained significant attention due to the release of large scale, challenging datasets. Present approaches to the problem largely focus on learning-based methods that use only textual information in order to classify whether a given premise entails, contradicts, or is neutral with respect to a given hypothesis. Surprisingly, the use of methods based on structured knowledge – a central topic in artificial intelligence – has not received much attention vis-a-vis the NLI problem. While there are many open knowledge bases that contain various types of reasoning information, their use for NLI has not been well explored. To address this, we present a combination of techniques that harness external knowledge to improve performance on the NLI problem in the science questions domain. We present the results of applying our techniques on text, graph, and text-and-graph based models; and discuss the implications of using external knowledge to solve the NLI problem. Our model achieves close to state-of-the-art performance for NLI on the SciTail science questions dataset.

Existing query engines for RDF graphs follow one of two design paradigms: relational or graph-based. We explore sparse matrix algebra as a third paradigm and propose MAGiQ: a framework for implementing SPARQL query engines that are portable on various hardware architectures, scalable over thousands of compute nodes, and efficient for very large RDF datasets. MAGiQ represents the RDF graph as a sparse matrix and defines a domain-specific language of algebraic operations. SPARQL queries are translated into matrix algebra programs that are oblivious to the underlying computing infrastructure. Existing matrix algebra libraries, optimized for each particular architecture, are called to execute the program and handle the performance issues. We present three case studies of matrix algebra back-end libraries: SuiteSparse, MATLAB, and CombBLAS; we demonstrate how MAGiQ can effortlessly be ported on a variety of architectures such as Intel CPUs, NVIDIA GPUs, and Cray XC40 supercomputers. Our experiments on large-scale real and synthetic datasets show that MAGiQ performs comparably to or better than existing specialized SPARQL query engines for data-intensive queries, scales to very large computing infrastructures, and handles datasets with up to 512 billion triples.

Existing RDF engines follow one of two design paradigms: relational or graph-based. Such engines are typically designed for specific hardware architectures, mainly CPUs, and are not easily portable to new architectures. Porting an existing engine to a different architecture (e.g., many-core architectures) entails almost redesign from scratch. We explore sparse matrix algebra as a third paradigm for designing a portable, scalable, and efficient RDF engine. We demonstrate MAGiQ; a matrix algebra approach for evaluating complex SPARQL queries over large RDF datasets. MAGiQ represents an RDF graph as a sparse matrix, and translates SPARQL queries to matrix algebra programs. MAGiQ takes advantage of the existing rich software infrastructure for processing sparse matrices, optimized for many architectures (e.g., CPUs, GPUs, distributed), effortlessly. This demo motivates the adoption of matrix algebra in RDF graph processing by showing MAGiQ's performance with different matrix algebra backend engines. MAGiQ, using a GPU, is orders of magnitude faster in solving complex queries on a billion edge graph than state-of-the-art RDF systems.

Many data center applications nowadays rely on distributed computation models like MapReduce and Bulk Synchronous Parallel (BSP) for data-intensive computation at scale [4]. These models scale by leveraging the partition/aggregate pattern where data and computations are distributed across many worker servers, each performing part of the computation. A communication phase is needed each time workers need to synchronize the computation and, at last, to produce the final output. In these applications, the network communication costs can be one of the dominant scalability bottlenecks especially in case of multi-stage or iterative computations [1].

Programmable data plane hardware creates new opportunities for infusing intelligence into the network. This raises a fundamental question: what kinds of computation should be delegated to the network?

Applications in life sciences, decentralized social networks, Internet of Things, and statistical linked dataspaces integrate data from multiple decentralized RDF graphs via SPARQL queries. Several approaches have been proposed to optimize query processing over a small number of heterogeneous data sources by utilizing schema information. In the case of schema similarity and interlinks among sources, these approaches cause unnecessary data retrieval and communication, leading to poor scalability and response time. This paper addresses these limitations and presents Lusail, a system for scalable and efficient SPARQL query processing over decentralized graphs. Lusail achieves scalability and low query response time through various optimizations at compile and run times. At compile time, we use a novel locality-aware query decomposition technique that maximizes the number of query triple patterns sent together to a source based on the actual location of the instances satisfying these triple patterns. At run time, we use selectivity-awareness and parallel query execution to reduce network latency and to increase parallelism by delaying the execution of subqueries expected to return large results. We evaluate Lusail using real and synthetic benchmarks, with data sizes up to billions of triples on an in-house cluster and a public cloud. We show that Lusail outperforms state-of-the-art systems by orders of magnitude in terms of scalability and response time.

There has been a proliferation of datasets available as interlinked RDF data accessible through SPARQL endpoints. This has led to the emergence of various applications in life science, distributed social networks, and Internet of Things that need to integrate data from multiple endpoints.

Distributed SPARQL engines promise to support very large RDF datasets by utilizing shared-nothing computer clusters. Some are based on distributed frameworks such as MapReduce; others implement proprietary distributed processing; and some rely on expensive preprocessing for data partitioning. These systems exhibit a variety of trade-offs that are not well-understood, due to the lack of any comprehensive quantitative and qualitative evaluation. In this paper, we present a survey of 22 state-of-the-art systems that cover the entire spectrum of distributed RDF data processing and categorize them by several characteristics. Then, we select 12 representative systems and perform extensive experimental evaluation with respect to preprocessing cost, query performance, scalability and workload adaptability, using a variety of synthetic and real large datasets with up to 4.3 billion triples. Our results provide valuable insights for practitioners to understand the trade-offs for their usage scenarios. Finally, we publish online our evaluation framework, including all datasets and workloads, for researchers to compare their novel systems against the existing ones.

The RDF data model allows publishing interlinked RDF datasets, where each dataset is independently maintained and is queryable via a SPARQL endpoint. Many applications would benefit from querying the resulting large, decentralized, geo-distributed graph through a federated SPARQL query processor. A crucial factor for good performance in federated query processing is pushing as much computation as possible to the local endpoints. Surprisingly, existing federated SPARQL engines are not effective at this task since they rely only on schema information. Consequently, they cause unnecessary data retrieval and communication, leading to poor scalability and response time. This paper addresses these limitations and presents Lusail , a scalable and efficient federated SPARQL system for querying large RDF graphs that are geo-distributed on different endpoints. Lusail uses a novel query rewriting algorithm to push computation to the local endpoints by relying on information about the RDF instances and not only the schema. The query rewriting algorithm has the additional advantage of exposing parallelism in query processing, which Lusail exploits through advanced scheduling at query run time. Our experiments on billions of triples of real and synthetic data show that Lusail outperforms state-of-the-art systems by orders of magnitude in terms of scalability and response time.

Modern applications require sophisticated analytics on RDF graphs that combine structural queries with generic graph computations. Existing systems support either declarative SPARQL queries, or generic graph processing, but not both. We bridge the gap by introducing Spartex, a versatile framework for complex RDF analytics. Spartex extends SPARQL to combine seamlessly generic graph algorithms (e.g., PageRank, Shortest Paths, etc.) with SPARQL queries. Spartex builds on existing vertex-centric graph processing frameworks, such as Graphlab or Pregel. It implements a generic SPARQL operator as a vertex-centric program that interprets SPARQL queries and executes them efficiently using a built-in optimizer. In addition, any graph algorithm implemented in the underlying vertex-centric framework, can be executed in Spartex. We present various scenarios where our framework simplifies significantly the implementation of complex RDF data analytics programs. We demonstrate that Spartex scales to datasets with billions of edges, and show that our core SPARQL engine is at least as fast as the state-of-the-art specialized RDF engines. For complex analytical tasks that combine generic graph processing with SPARQL, Spartex is at least an order of magnitude faster than existing alternatives.

Frequent Subgraph Mining is an essential operation for graph analytics and knowledge extraction. Due to its high computational cost, parallel solutions are necessary. Existing approaches either suffer from load imbalance, or high communication and synchronization overheads. In this paper we propose ScaleMine; a novel parallel frequent subgraph mining system for a single large graph. ScaleMine introduces a novel two-phase approach. The first phase is approximate; it quickly identifies subgraphs that are frequent with high probability, while collecting various statistics. The second phase computes the exact solution by employing the results of the approximation to achieve good load balance; prune the search space; generate efficient execution plans; and guide intra-task parallelism. Our experiments show that ScaleMine scales to 8,192 cores on a Cray XC40 (12× more than competitors); supports graphs with one billion edges (10× larger than competitors), and is at least an order of magnitude faster than existing solutions.

A growing number of applications require combining SPARQL queries with generic graph search on RDF data. However, the lack of procedural capabilities in SPARQL makes it inappropriate for graph analytics. Moreover, RDF engines focus on SPARQL query evaluation whereas graph management frameworks perform only generic graph computations. In this work, we bridge the gap by introducing SPARTex, an RDF analytics framework based on the vertex-centric computation model. In SPARTex, user-defined vertex centric programs can be invoked from SPARQL as stored procedures. SPARTex allows the execution of a pipeline of graph algorithms without the need for multiple reads/writes of input data and intermediate results. We use a cost-based optimizer for minimizing the communication cost. SPARTex evaluates queries that combine SPARQL and generic graph computations orders of magnitude faster than existing RDF engines. We demonstrate a real system prototype of SPARTex running on a local cluster using real and synthetic datasets. SPARTex has a real-time graphical user interface that allows the participants to write regular SPARQL queries, use our proposed SPARQL extension to declaratively invoke graph algorithms or combine/pipeline both SPARQL querying and generic graph analytics.

Distributed RDF systems partition data across multiple computer nodes. Partitioning is typically based on heuristics that minimize inter-node communication and it is performed in an initial, data pre-processing phase. Therefore, the resulting partitions are static and do not adapt to changes in the query workload; as a result, existing systems are unable to consistently avoid communication for queries that are not favored by the initial data partitioning. Furthermore, for very large RDF knowledge bases, the partitioning phase becomes prohibitively expensive, leading to high startup costs. In this paper, we propose AdHash, a distributed RDF system which addresses the shortcomings of previous work. First, AdHash initially applies lightweight hash partitioning, which drastically minimizes the startup cost, while favoring the parallel processing of join patterns on subjects, without any data communication. Using a locality-aware planner, queries that cannot be processed in parallel are evaluated with minimal communication. Second, AdHash monitors the data access patterns and adapts dynamically to the query load by incrementally redistributing and replicating frequently accessed data. As a result, the communication cost for future queries is drastically reduced or even eliminated. Our experiments with synthetic and real data verify that AdHash (i) starts faster than all existing systems, (ii) processes thousands of queries before other systems become online, and (iii) gracefully adapts to the query load, being able to evaluate queries on billion-scale RDF data in sub-seconds. In this demonstration, audience can use a graphical interface of AdHash to verify its performance superiority compared to state-of-the-art distributed RDF systems.