arXiv subjects

Renata Borovica-Gajic

Publications and source records attributed to Renata Borovica-Gajic.

At least 19 recordsLinked to original sources

arXiv2026

High-Throughput LLM-Based Compression and What It Takes: A Systems Study of Inference Engines and Entropy Coding Optimizations

Large language models are attractive text compressors because next-token probabilities can be converted into code lengths. In practice, however, the main bottleneck is often not modeling quality but systems cost: a compressor must obtain logits quickly enough to compete with strong classical baselines. This paper examines systems optimizations for LLM-based compression, including multistream pipelined arithmetic coding and target-interval capture, and evaluates their trade-offs across inference engines, configurations and datasets. We evaluate LLM-based text compression on three different inference engines: Transformers, vLLM, and TensorRT-LLM. Notably, we obtain a speedup of 18.84x compared to prior work's Transformers implementation and a maximum recorded inference speed of 181 kilobytes per second at a compression ratio of 7.17, when running on an NVIDIA A100.

cs.DB↗

arXiv2026

RUBRIC: Realism--Utility Balanced Ranking for Imbalanced Classification

Class imbalance poses a fundamental challenge in risk-sensitive applications such as fraud detection and medical diagnosis, where minority-class samples are scarce yet critical for accurate classification. Existing oversampling methods generate synthetic samples to rebalance class distributions; however, they often produce large numbers of low-quality candidates that distort decision boundaries or introduce artifacts, leading to overfitting and degraded generalization. In this work, we introduce \textbf{RUBRIC}, a generator-agnostic filtering framework that formulates synthetic sample selection as a quality-over-quantity optimization problem. RUBRIC ranks candidates using a realism-utility trade-off: realism is estimated via a neural density-ratio discriminator from each candidate's resemblance to real minority samples, while utility captures proximity to the decision boundary through a concave, margin-based scoring function $g_-(t)=-\log(1+e^{-t/τ})$. The discriminator uses the same architecture and training protocol on every benchmark and is fit independently to that dataset's real minority class versus its synthetic pool. We show that, under mild regularity conditions, the proposed filtering framework monotonically tightens the generalization bound for margin-based classifiers by jointly reducing distribution shift and suppressing near-negative tail contributions. Through extensive experiments on standard public imbalanced-classification benchmarks, we demonstrate that RUBRIC boosts minority-class recall while preserving overall discriminative ability across multiple data generators. Sensitivity analyses in $λ$ and the selection budget $K$ further characterize performance trade-offs oriented toward ranking quality.

cs.LG↗

arXiv2026

Towards Anticipatory Databases Through Shared Data and Workload Semantics

Database management systems increasingly serve dynamic and exploratory workloads, yet many of their decisions still rely on low-level signals such as recency, frequency, and address locality. These signals capture how data was accessed, but not what is being examined or how an analytical focus evolves. We argue for treating workload semantics as a first-class control signal for anticipatory decision making. Central to this view, we introduce semantic locality and semantic trajectories, which capture relationships among nearby queries and how those relationships evolve across a session. We propose a framework that represents semantic context at the data, query, and session levels, models its evolution over time, and translates it into task-specific utility estimates. We instantiate this framework in semantic prefetching and semantic cache eviction, which share a semantic layer to make two separate decisions. Prefetching uses semantic trajectories to anticipate future accesses beyond what address-based locality can capture, while eviction uses semantic relevance to inform block replacement. These systems provide initial evidence that shared semantic context can support multiple DBMS components. We further outline how this principle can extend to other decisions and data systems, and discuss key challenges in representation, cost, adaptation, and evaluation.

cs.DB↗

arXiv2026

HSMLA: Hierarchical Softmax Multi-scale Linear Attention for Efficient Vision Transformers

Vision transformers face significant computational overheads in high-resolution dense prediction due to the quadratic complexity of self-attention. Linear attention offers efficiency but sacrifices local context modeling. We propose \textbf{HSMLA (Hierarchical Softmax Multi-scale Linear Attention)}, which combines ReLU-based linear attention for global context, selective softmax refinement for critical local features, and multi-scale token representations via depthwise convolutions. HSMLA achieves superior accuracy-efficiency trade-offs: up to $4.2\times$ inference-time speedup across dense prediction tasks, $87.3%$ Dice with $3.2\times$ speedup on CT organ segmentation, and $94.2%$ AUC with $4.1\times$ speedup on pathology WSI.

cs.CV↗

arXiv2026

IDSTune: A Multi-Agent Collaborative Framework for Integrated Database System Tuning

Database tuning is critical for achieving high performance in modern database management systems (DBMSs). Existing methods typically optimize a single component---knobs, indexes, or materialized views---without accounting for their interdependencies. This limitation arises because these components require different tuning strategies and are difficult to integrate within a unified framework. As a result, directly extending a method to multiple components or simply combining separate methods often fails to capture cross-component collaboration and shared tuning signals. Moreover, existing methods are insufficient for handling diverse workloads, evolving data, and dynamic query patterns. To address these limitations, we propose IDSTune, an integrated tuning framework that jointly optimizes multiple configuration components through LLM-driven multi-agent collaboration. IDSTune operates in two phases: (i) workload compression, which extracts and selects task-relevant features, and (ii) configuration recommendation, where specialized agents collaboratively generate and refine configurations for knobs, indexes, and materialized views under the supervision of a centralized coordinator. By incorporating feedback and external knowledge retrieval, IDSTune achieves efficient and globally consistent tuning. Extensive experiments show that IDSTune achieves up to 38% performance improvement and 57% faster tuning, with strong adaptability across diverse scenarios.

cs.DB↗

arXiv2026

SemCEB: A Cardinality Estimation Benchmark for Semantic Operators

Modern data systems increasingly expose multi-modal large language models as semantic operators: SQL operators, including filters and joins, whose predicates are defined by a natural-language instruction. Query optimization in these systems still rests on the same foundations as in traditional databases$\unicode{x2013}$plan enumeration and cost models$\unicode{x2013}$yet faces new challenges, e.g., a larger plan space and the lack of efficient cardinality estimates. The elevated per-tuple costs of semantic operators make bad plan choices worse by orders of magnitude. Therefore, precise$\unicode{x2013}$but also fast and cheap$\unicode{x2013}$cardinality estimates for semantic filters and joins are of high importance for optimizing query plans that include semantic operators. In this paper, we introduce SemCEB, the first benchmark for cardinality estimation over semantic operators, based on a real-world dataset of (semi-)structured text and images with 102 hand-curated, diverse queries spanning a wide range of selectivities, assessing cardinality estimation for semantic filters and joins in isolation. We evaluate sampling-based algorithms and Semantic Histograms, a state-of-the-art cardinality estimation algorithm for semantic operators, with respect to their accuracy, cost, latency, and memory overhead. We show that, while sampling is robust across different predicate categories, it does not scale and comes with high costs. Our adaptation of Semantic Histograms, on the other hand, is limited in its applicability, and its performance appears sensitive to the predicate category.

cs.DB↗

arXiv2025

CoLSE: A Lightweight and Robust Hybrid Learned Model for Single-Table Cardinality Estimation using Joint CDF

Cardinality estimation (CE), the task of predicting the result size of queries is a critical component of query optimization. Accurate estimates are essential for generating efficient query execution plans. Recently, machine learning techniques have been applied to CE, broadly categorized into query-driven and data-driven approaches. Data-driven methods learn the joint distribution of data, while query-driven methods construct regression models that map query features to cardinalities. Ideally, a CE technique should strike a balance among three key factors: accuracy, efficiency, and memory footprint. However, existing state-of-the-art models often fail to achieve this balance. To address this, we propose CoLSE, a hybrid learned approach for single-table cardinality estimation. CoLSE directly models the joint probability over queried intervals using a novel algorithm based on copula theory and integrates a lightweight neural network to correct residual estimation errors. Experimental results show that CoLSE achieves a favorable trade-off among accuracy, training time, inference latency, and model size, outperforming existing state-of-the-art methods.

cs.DB↗

arXiv2025

Benchmarking RL-Enhanced Spatial Indices Against Traditional, Advanced, and Learned Counterparts

Reinforcement learning has recently been used to enhance index structures, giving rise to reinforcement learning-enhanced spatial indices (RLESIs) that aim to improve query efficiency during index construction. However, their practical benefits remain unclear due to the lack of unified implementations and comprehensive evaluations, especially in disk-based settings. We present the first modular and extensible benchmark for RLESIs. Built on top of an existing spatial index library, our framework decouples index training from building, supports parameter tuning, and enables consistent comparison with traditional, advanced, and learned spatial indices. We evaluate 12 representative spatial indices across six datasets and diverse workloads, including point, range, kNN, spatial join, and mixed read/write queries. Using latency, I/O, and index statistics as metrics, we find that while RLESIs can reduce query latency with tuning, they consistently underperform learned spatial indices and advanced variants in both query efficiency and index build cost. These findings highlight that although RLESIs offer promising architectural compatibility, their high tuning costs and limited generalization hinder practical adoption.

cs.DB↗

arXiv2025

GrASP: A Generalizable Address-based Semantic Prefetcher for Scalable Transactional and Analytical Workloads

Data prefetching--loading data into the cache before it is requested--is essential for reducing I/O overhead and improving database performance. While traditional prefetchers focus on sequential patterns, recent learning-based approaches, especially those leveraging data semantics, achieve higher accuracy for complex access patterns. However, these methods often struggle with today's dynamic, ever-growing datasets and require frequent, timely fine-tuning. Privacy constraints may also restrict access to complete datasets, necessitating prefetchers that can learn effectively from samples. To address these challenges, we present GrASP, a learning-based prefetcher designed for both analytical and transactional workloads. GrASP enhances prefetching accuracy and scalability by leveraging logical block address deltas and combining query representations with result encodings. It frames prefetching as a context-aware multi-label classification task, using multi-layer LSTMs to predict delta patterns from embedded context. This delta modeling approach enables GrASP to generalize predictions from small samples to larger, dynamic datasets without requiring extensive retraining. Experiments on real-world datasets and industrial benchmarks demonstrate that GrASP generalizes to datasets 250 times larger than the training data, achieving up to 45% higher hit ratios, 60% lower I/O time, and 55% lower end-to-end query execution latency than existing baselines. On average, GrASP attains a 91.4% hit ratio, a 90.8% I/O time reduction, and a 57.1% execution latency reduction.

cs.DB↗

arXiv2025

DriftBench: Defining and Generating Data and Query Workload Drift for Benchmarking

Data and workload drift are key to evaluating database components such as caching, cardinality estimation, indexing, and query optimization. Yet, existing benchmarks are static, offering little to no support for modeling drift. This limitation stems from the lack of clear definitions and tools for generating data and workload drift. Motivated by this gap, we propose a unified taxonomy for data and workload drift, grounded in observations from both academia and industry. Building on this foundation, we introduce DriftBench, a lightweight and extensible framework for generating data and workload drift in benchmark inputs. Together, the taxonomy and DriftBench provide a standardized vocabulary and mechanism for modeling and generating drift in benchmarking. We demonstrate their effectiveness through case studies involving data drift, workload drift, and drift-aware cardinality estimation.

cs.DB↗

arXiv2025

InsightQL: Advancing Human-Assisted Fuzzing with a Unified Code Database and Parameterized Query Interface

Fuzzing is a highly effective automated testing method for uncovering software vulnerabilities. Despite advances in fuzzing techniques, such as coverage-guided greybox fuzzing, many fuzzers struggle with coverage plateaus caused by fuzz blockers, limiting their ability to find deeper vulnerabilities. Human expertise can address these challenges, but analyzing fuzzing results to guide this support remains labor-intensive. To tackle this, we introduce InsightQL, the first human-assisting framework for fuzz blocker analysis. Powered by a unified database and an intuitive parameterized query interface, InsightQL aids developers in systematically extracting insights and efficiently unblocking fuzz blockers. Our experiments on 14 popular real-world libraries from the FuzzBench benchmark demonstrate the effectiveness of InsightQL, leading to the unblocking of many fuzz blockers and considerable improvements in code coverage (up to 13.90%).

cs.SE↗

arXiv2024

SeLeP: Learning Based Semantic Prefetching for Exploratory Database Workloads

Prefetching is a crucial technique employed in traditional databases to enhance interactivity, particularly in the context of data exploitation. Data exploration is a query processing paradigm in which users search for insights buried in the data, often not knowing what exactly they are looking for. Data exploratory tools deal with multiple challenges such as the need for interactivity with no a priori knowledge being present to help with the system tuning. The state-of-the-art prefetchers are specifically designed for navigational workloads only, where the number of possible actions is limited. The prefetchers that work with SQL-based workloads, on the other hand, mainly rely on data logical addresses rather than the data semantics. They fail to predict complex access patterns in cases where the database size is substantial, resulting in an extensive address space, or when there is frequent co-accessing of data. In this paper, we propose SeLeP, a semantic prefetcher that makes prefetching decisions for both types of workloads, based on the encoding of the data values contained inside the accessed blocks. Following the popular path of using machine learning approaches to automatically learn the hidden patterns, we formulate the prefetching task as a time-series forecasting problem and use an encoder-decoder LSTM architecture to learn the data access pattern. Our extensive experiments, across real-life exploratory workloads, demonstrate that SeLeP improves the hit ratio up to 40% and reduces I/O time up to 45% compared to the state-of-the-art, attaining impressive 95% hit ratio and 80% I/O reduction on average.

cs.DB↗

arXiv2023

A Simple Yet High-Performing On-disk Learned Index: Can We Have Our Cake and Eat it Too?

While in-memory learned indexes have shown promising performance as compared to B+-tree, most widely used databases in real applications still rely on disk-based operations. Based on our experiments, we observe that directly applying the existing learned indexes on disk suffers from several drawbacks and cannot outperform a standard B+-tree in most cases. Therefore, in this work we make the first attempt to show how the idea of learned index can benefit the on-disk index by proposing AULID, a fully on-disk updatable learned index that can achieve state-of-the-art performance across multiple workload types. The AULID approach combines the benefits from both traditional indexing techniques and the learned indexes to reduce the I/O cost, the main overhead under disk setting. Specifically, three aspects are taken into consideration in reducing I/O costs: (1) reduce the overhead in updating the index structure; (2) induce shorter paths from root to leaf node; (3) achieve better locality to minimize the number of block reads required to complete a scan. Five principles are proposed to guide the design of AULID which shows remarkable performance gains and meanwhile is easy to implement. Our evaluation shows that AULID has comparable storage costs to a B+-tree and is much smaller than other learned indexes, and AULID is up to 2.11x, 8.63x, 1.72x, 5.51x, and 8.02x more efficient than FITing-tree, PGM, B+-tree, ALEX, and LIPP.

cs.DB↗

arXiv2023

Updatable Learned Indexes Meet Disk-Resident DBMS -- From Evaluations to Design Choices

Although many updatable learned indexes have been proposed in recent years, whether they can outperform traditional approaches on disk remains unknown. In this study, we revisit and implement four state-of-the-art updatable learned indexes on disk, and compare them against the B+-tree under a wide range of settings. Through our evaluation, we make some key observations: 1) Overall, the B+-tree performs well across a range of workload types and datasets. 2) A learned index could outperform B+-tree or other learned indexes on disk for a specific workload. For example, PGM achieves the best performance in write-only workloads while LIPP significantly outperforms others in lookup-only workloads. We further conduct a detailed performance analysis to reveal the strengths and weaknesses of these learned indexes on disk. Moreover, we summarize the observed common shortcomings in five categories and propose four design principles to guide future design of on-disk, updatable learned indexes: (1) reducing the index's tree height, (2) better data structures to lower operation overheads, (3) improving the efficiency of scan operations, and (4) more efficient storage layout.

cs.DB↗

arXiv2022

Testing the Robustness of Learned Index Structures

While early empirical evidence has supported the case for learned index structures as having favourable average-case performance, little is known about their worst-case performance. By contrast, classical structures are known to achieve optimal worst-case behaviour. This work evaluates the robustness of learned index structures in the presence of adversarial workloads. To simulate adversarial workloads, we carry out a data poisoning attack on linear regression models that manipulates the cumulative distribution function (CDF) on which the learned index model is trained. The attack deteriorates the fit of the underlying ML model by injecting a set of poisoning keys into the training dataset, which leads to an increase in the prediction error of the model and thus deteriorates the overall performance of the learned index structure. We assess the performance of various regression methods and the learned index implementations ALEX and PGM-Index. We show that learned index structures can suffer from a significant performance deterioration of up to 20% when evaluated on poisoned vs. non-poisoned datasets.

cs.DB↗

arXiv2022

Energy Efficient Computing Systems: Architectures, Abstractions and Modeling to Techniques and Standards

Computing systems have undergone several inflexion points - while Moore's law guided the semiconductor industry to cram more and more transistors and logic into the same volume, the limits of instruction-level parallelism (ILP) and the end of Dennard's scaling drove the industry towards multi-core chips. We have now entered the era of domain-specific architectures for new workloads like AI and ML. These trends continue, arguably with other limits, along with challenges imposed by tighter integration, extreme form factors and diverse workloads, making systems more complex from an energy efficiency perspective. Many research surveys have covered different aspects of techniques in hardware and microarchitecture across devices, servers, HPC, data center systems along with software, algorithms, frameworks for energy efficiency and thermal management. Somewhat in parallel, the semiconductor industry has developed techniques and standards around specification, modeling and verification of complex chips; these areas have not been addressed in detail by previous research surveys. This survey aims to bring these domains together and is composed of a systematic categorization of key aspects of building energy efficient systems - (a) specification - the ability to precisely specify the power intent or properties at different layers (b) modeling and simulation of the entire system or subsystem (hardware or software or both) so as to be able to perform what-if analysis, (c) techniques used for implementing energy efficiency at different levels of the stack, (d) verification techniques used to provide guarantees that the functionality of complex designs are preserved, and (e) energy efficiency standards and consortiums that aim to standardize different aspects of energy efficiency, including cross-layer optimizations.

cs.AR↗

arXiv2022

Intelligent Autonomous Intersection Management

Connected Autonomous Vehicles will make autonomous intersection management a reality replacing traditional traffic signal control. Autonomous intersection management requires time and speed adjustment of vehicles arriving at an intersection for collision-free passing through the intersection. Due to its computational complexity, this problem has been studied only when vehicle arrival times towards the vicinity of the intersection are known beforehand, which limits the applicability of these solutions for real-time deployment. To solve the real-time autonomous traffic intersection management problem, we propose a reinforcement learning (RL) based multiagent architecture and a novel RL algorithm coined multi-discount Q-learning. In multi-discount Q-learning, we introduce a simple yet effective way to solve a Markov Decision Process by preserving both short-term and long-term goals, which is crucial for collision-free speed control. Our empirical results show that our RL-based multiagent solution can achieve near-optimal performance efficiently when minimizing the travel time through an intersection.

cs.MA↗

arXiv2022

Solving Dynamic Graph Problems with Multi-Attention Deep Reinforcement Learning

Graph problems such as traveling salesman problem, or finding minimal Steiner trees are widely studied and used in data engineering and computer science. Typically, in real-world applications, the features of the graph tend to change over time, thus, finding a solution to the problem becomes challenging. The dynamic version of many graph problems are the key for a plethora of real-world problems in transportation, telecommunication, and social networks. In recent years, using deep learning techniques to find heuristic solutions for NP-hard graph combinatorial problems has gained much interest as these learned heuristics can find near-optimal solutions efficiently. However, most of the existing methods for learning heuristics focus on static graph problems. The dynamic nature makes NP-hard graph problems much more challenging to learn, and the existing methods fail to find reasonable solutions. In this paper, we propose a novel architecture named Graph Temporal Attention with Reinforcement Learning (GTA-RL) to learn heuristic solutions for graph-based dynamic combinatorial optimization problems. The GTA-RL architecture consists of an encoder capable of embedding temporal features of a combinatorial problem instance and a decoder capable of dynamically focusing on the embedded features to find a solution to a given combinatorial problem instance. We then extend our architecture to learn heuristics for the real-time version of combinatorial optimization problems where all input features of a problem are not known a prior, but rather learned in real-time. Our experimental results against several state-of-the-art learning-based algorithms and optimal solvers demonstrate that our approach outperforms the state-of-the-art learning-based approaches in terms of effectiveness and optimal solvers in terms of efficiency on dynamic and real-time graph combinatorial optimization.

cs.LG↗