About

Nalini Venkatasubramanian is a Professor of Computer Science in the Donald Bren School of Information and Computer Sciences at the University of California, Irvine. She is known for her work in effective management and utilization of resources in the evolving global information infrastructure. Her research interests include Multimedia Computing, Networked and Distributed Systems, Internet technologies and Applications, Ubiquitous Computing, and Urban Crisis Responses. Dr. Venkatasubramanian's research focuses on enabling effective management and utilization of resources in the evolving global information infrastructure, as well as addressing the problem of composing resource management services in distributed systems.

Research topics

• Computer Science
• Data Mining
• Artificial Intelligence
• Machine Learning
• Mathematics
• Computer Security
• Medicine
• Engineering
• Data science
• Virology

Selected publications

• EB-Refine: Improving Suffix Prediction Accuracy for Long and Non-Pareto Event Logs through Evidence-Based Refinement

  2026-02-02

  article

  Predictive Process Monitoring (PPM) seeks to anticipate the future behavior of ongoing process instances, an area that has seen rapid progress with deep learning techniques. Yet, accurately predicting future activity sequences (suffix) remains difficult, particularly on long and Non-Pareto event logs. In longtail suffix prediction, stepwise errors accumulate, significantly reducing sequence-level accuracy. This challenge is amplified in Non-Pareto logs, where activity frequencies are more evenly distributed and dominant behavioral patterns are absent, limiting statistical cues for generalization. To address these limitations, we propose EB-Refine, a refinement framework that integrates deep learning with evidence-based refinement mechanism to improve suffix prediction, especially on long and Non-Pareto event logs. EB-Refine employs a two-stage design: (1) fine-tune a BERTbased model to generate an initial suffix prediction; (2) this prediction is then refined by comparing it against frequency-indexed database(FDB) through the proposed evidence-based refinement mechanism. By combining learned predictions with historical evidence, EB-Refine efficiently retrieves the most relevant historical trace to guide the final revision. Crucially, the refinement operates solely at inference, introducing no additional training overhead while yielding substantial accuracy gains. Extensive experiments demonstrate that EB-Refine consistently outperforms state-of-the-art baselines by $\mathbf{1 0 \% - 3 0 \%}$ across diverse public benchmark datasets. Additional evaluations on synthetic datasets simulating varying noise types in a wastewater treatment process further confirm its robustness to data imperfections.

  PublisherDOI

• VOILA: Value-of-Information Guided Fidelity Selection for Cost-Aware Multimodal Question Answering

  Open MIND · 2026-02-03

  preprintSenior author

  Despite significant costs from retrieving and processing high-fidelity visual inputs, most multimodal vision-language systems operate at fixed fidelity levels. We introduce VOILA, a framework for Value-Of-Information-driven adaptive fidelity selection in Visual Question Answering (VQA) that optimizes what information to retrieve before model execution. Given a query, VOILA uses a two-stage pipeline: a gradient-boosted regressor estimates correctness likelihood at each fidelity from question features alone, then an isotonic calibrator refines these probabilities for reliable decision-making. The system selects the minimum-cost fidelity maximizing expected utility given predicted accuracy and retrieval costs. We evaluate VOILA across three deployment scenarios using five datasets (VQA-v2, GQA, TextVQA, LoCoMo, FloodNet) and six Vision-Language Models (VLMs) with 7B-235B parameters. VOILA consistently achieves 50-60% cost reductions while retaining 90-95% of full-resolution accuracy across diverse query types and model architectures, demonstrating that pre-retrieval fidelity selection is vital to optimize multimodal inference under resource constraints.

  DOI

• VOILA: Value-of-Information Guided Fidelity Selection for Cost-Aware Multimodal Question Answering

  ArXiv.org · 2026-02-03

  articleOpen accessSenior author

  Despite significant costs from retrieving and processing high-fidelity visual inputs, most multimodal vision-language systems operate at fixed fidelity levels. We introduce VOILA, a framework for Value-Of-Information-driven adaptive fidelity selection in Visual Question Answering (VQA) that optimizes what information to retrieve before model execution. Given a query, VOILA uses a two-stage pipeline: a gradient-boosted regressor estimates correctness likelihood at each fidelity from question features alone, then an isotonic calibrator refines these probabilities for reliable decision-making. The system selects the minimum-cost fidelity maximizing expected utility given predicted accuracy and retrieval costs. We evaluate VOILA across three deployment scenarios using five datasets (VQA-v2, GQA, TextVQA, LoCoMo, FloodNet) and six Vision-Language Models (VLMs) with 7B-235B parameters. VOILA consistently achieves 50-60% cost reductions while retaining 90-95% of full-resolution accuracy across diverse query types and model architectures, demonstrating that pre-retrieval fidelity selection is vital to optimize multimodal inference under resource constraints.

  PublisherOA PDF

• Microservice provisioning and event-driven adaptation in heterogeneous IoT settings

  Pervasive and Mobile Computing · 2026-02-13

  articleOpen accessSenior author

  In mission-critical Internet-of-Things environments, where data from diverse data sources (stationary and mobile) must be analyzed rapidly, ensuring reliable service provisioning for the execution of analytics is critical. Under dynamic conditions and changing infrastructure capabilities, adaptive methods are required to handle surges in data volumes and execute complex services for IoT applications. In this paper, we propose AMPHI–an adaptive microservice provisioning framework that handles various mission-aware workflows in combined stationary-mobile IoT environments to enable flexible deployment of containerized microservices. AMPHI exploits the variability in cost and performance among operators that implement similar functionalities but with distinct cost/quality/time tradeoffs, and provides solutions for intelligent selection, placement, and sharing of operators across a hybrid set of devices (stationary sensors and unmmaned aerial/ground vehicle) to maximize Quality of Service (QoS) and resource efficiency under dynamic situations. We formulate the microservice provisioning and instantiation as an (NP-hard) optimization problem and design efficient heuristic approaches to deploy microservices by utilizing application and system context. The system’s robustness and resilience are guaranteed by two variant paradigms in AMPHI’s design, one provides with optimal microservice provisoining solutions, and the other makes adaptation operations on existing solutions when abnormal events occur. We study a basic version of AMPHI that focuses on optimal microservice provisioning; the basic version is then enhanced to create a robust and extensible variant of AMPHI that ensures dependable operation through operators that can deal with unexpected failures and urgent service requests. AMPHI is evaluated in the context of smart firefighting with high-rise fires utilizing building sensors and autonomous aerial mobile units. Through real-world testbeds and extensive simulations, we show how AMPHI allows flexible and cost-effective execution of dynamically changing IoT workflows.

  PublisherDOI

• SEQUIN: A Network Science and Physics-based Approach to Identify Sequential N-k Attacks in Electric Power Grids

  2025-05-06

  articleOpen accessSenior author

  The electric grid is a vital infrastructure that supplies power on which modern society depends, and maintaining its reliable service and resilient operation is essential. The grid's performance typically relies on a few key components. However, efficiently finding these components is challenging, due to the geo-distributed scale of the grid, complex physics governing power flows, and automated network response. Realistically identifying these key components must also consider the temporal aspect of how failures affect the network. In this paper, we address the problem of identifying worst-case disruptions to the grid, under the sequential failure of components. We present SEQUIN, a framework leveraging network science principles and physics-based constraint optimization to explore such failures in the grid. We formulate the problem using a sequential N-k interdiction model, which provides a methodology to explore and capture interactions between the failures and network response. Our approach defines several network properties to assess the contribution of each component towards its operation, and provides an efficient guided exploration of attacks. We also provide a toolkit to help reason about the impact on the grid. Extensive experiments on multiple benchmark grid networks are conducted to show the efficacy of our approach and demonstrate how the varying the sequence of attacks can result in different levels of disruption.

  PublisherOA PDFDOI

• SEQUIN: A Network Science and Physics-based Approach to Identify Sequential N-k Attacks in Electric Power Grids

  2025-05-06

  articleOpen accessSenior author

  Electric grid infrastructures are critical systems that provide power to communities, but face issues of reliability and resilience due to extreme events, e.g., natural disasters and man-made attacks. In general, the performance of the grid depends on a few key components. However, efficiently identifying them can be challenging. This demo paper presents the SEQUIN toolkit for exploring the evolution of the grid under the sequential failure of components. Our tool uses the specification of an electric power grid network, and provides an interface on which domain experts and practitioners can explore hypothetical attacks and their impacts. The underlying logic in the SEQUIN tool relies on network science-based principles and physics-based optimization. We envision that this tool can be used to help reason about the impact of such failures for the grid.

  PublisherOA PDFDOI

• Meaningful Data Erasure in the Presence of Dependencies

  Proceedings of the VLDB Endowment · 2025-06-01 · 1 citations

  articleOpen accessSenior author

  Data regulations like GDPR require systems to support data erasure but leave the definition of "erasure" open to interpretation. This ambiguity makes compliance challenging, especially in databases where data dependencies can lead to erased data being inferred from remaining data. We formally define a precise notion of data erasure that ensures any inference about deleted data, through dependencies, remains bounded to what could have been inferred before its insertion. We design erasure mechanisms that enforce this guarantee at minimal cost. Additionally, we explore strategies to balance cost and throughput, batch multiple erasures, and proactively compute data retention times when possible. We demonstrate the practicality and scalability of our algorithms using both real and synthetic datasets.

  PublisherOA PDFDOI

• Microservice Provisioning and Event-Driven Adaptation in Heterogeneous Iot Settings

  SSRN Electronic Journal · 2025-01-01

  preprintOpen accessSenior author
  PublisherDOI

• Experiences with Composability for Resilient IoT Systems

  Lecture notes in computer science · 2025-09-24

  book-chapter1st authorCorresponding
  PublisherDOI

• In-Context Example Ordering for LLM-Based API Sequence Generation

  2025-12-15

  articleOpen accessSenior author

  We demonstrate OptiSeq [1], an edge-cloud middleware system that converts natural language queries into executable API sequences with high reliability in distributed environments. A central challenge in such systems is that large language models (LLMs) can generate incorrect API sequences, causing costly retries that consume edge bandwidth and add cloud latency. Our demonstration shows how OptiSeq, a lightweight inference-time example-ordering optimization engine, reduces these failures by intelligently arranging in-context examples before prompting the LLM. In our distributed architecture, (1) natural language queries originate at the edge, (2) relevant examples are retrieved from an edge-hosted vector store, (3) OptiSeq runs within the middleware pipeline to optimize prompt construction, (4) a cloud-hosted LLM generates API sequences, and (5) a distributed API execution layer carries out the calls. Through live demonstrations on real-world API sequencing tasks, we show that OptiSeq-guided prompting reduces retry attempts by 35-45 % while maintaining high accuracy. This work presents the first demonstration of ordering-aware middleware for natural-language-to-API sequence generation deployed across edge-cloud systems.

  PublisherDOI

Recent grants

• CPS: Medium: Collaborative Research: Dependability Techniques for Instrumented Cyber-Physical Spaces

  NSF · $516k · 2010–2015

• SCC-IRG-Track2: Creating an Extensible Data Exchange and Analytics Sandbox for Smart Water infrastructures

  NSF · $1.5M · 2020–2026

• Collaborative Research: CSR-EHS: Modeling and Exploiting Cross-Layer Timing in Distributed Embedded Systems

  NSF · $125k · 2006–2008

• EAGER: Exploring Resilience in SmartCity Water Infrastructure

  NSF · $150k · 2015–2020

• PC3: Collaborative Research: Pervasive Computing for Disaster Response

  NSF · $164k · 2011–2014

Frequent coauthors

• Sharad Mehrotra

  184 shared

• Roberto Yus

  University of Maryland, Baltimore County

  60 shared

• Andrew Chio

  University of California, Irvine

  58 shared

• Daokun Jiang

  University of California, Irvine

  54 shared

• Cheng-Hsin Hsu

  National Tsing Hua University

  50 shared

• Nikil Dutt

  39 shared

• Carolyn Talcott

  39 shared

• Daniela Nicklas

  University of Bamberg

  36 shared

Education

• Ph.D., Computer Science

  University of California, Los Angeles

  1995

• M.S., Computer Science

  University of California, Los Angeles

  1991

• B.S., Electrical and Electronics Engineering

  University of Madras

  1987

Awards & honors

• AAAS Fellow (2025)
• IEEE Fellow (2025)
• ACM 2021 Distinguished Member