Research
Our mission is to make a secure, performant, and affordable Internet accessible to everyone. We pursue this mission along two paths, each of which requires bridging fundamental data gaps in how we measure and operate networks:
- Path I, Data-Driven Policymaking: building independent, decision-grade broadband data so that public investment in connectivity can be measured, audited, and held accountable.
- Path II, Self-Driving Networks: developing AI-powered network operations that let networks adapt, generalize, and stay trustworthy with limited infrastructure and operational resources.
Both paths turn on the same conviction: we cannot fix what we cannot measure, and we cannot safely automate what we cannot generate and reason about as data.
Path I: Data-Driven Policymaking
Public investment in broadband runs into a fundamental data gap: we cannot tell whether tens of billions of dollars in funding actually reach the communities they target. Our work closes that gap, building the intellectual foundations and operational infrastructure to measure affordability, competition, and performance at scale.
Central to this effort is the Broadband Query Tool (BQT). BQT has informed regulatory and policy discussions across states and agencies, shaping how competition and pricing are evaluated in large-scale funding and oversight decisions. The broader vision is set out in What We Can’t See, We Can’t Fix.
BQT has enabled a series of research contributions grounded in measurement and statistical modeling:
- Contextualizing Speed Test Measurements (IMC 2022, Distinguished Paper Award): principled modeling of user-perceived performance.
- Decoding the Divide (SIGCOMM 2023): address-level broadband pricing measurement for market analysis.
- Assessing the Efficacy of the Connect America Fund (SIGCOMM 2024, ANRP Award): scalable methods for extracting and synthesizing ISP plan data across heterogeneous interfaces.
Projects
BQT
Broadband Query Tool: an address-level broadband pricing and availability platform for reproducible analysis of market structure and affordability. BQT decomposes ISP interaction processes into reusable, adaptive components that scale across providers. BQT is used by Cal Advocates, the Institute for Local Self-Reliance (ILSR), and Merit Network.
- Robust and Extensible Measurement of Broadband Plans with BQT+ (arXiv preprint arXiv:2602.16969, 2026)
- Enabling Data-Driven Policymaking using Broadband-Plan Querying Tool (BQT+) (ACM IMC PRIME Workshop, 2025)
- From Measurement to Policy Design: A Convex Optimization Framework for Broadband Affordability Using Address-Level Data Collected with the Broadband-Plan Querying Tool (BQT+) (TPRC, 2026)
- Assessing the Efficacy of the Connect America Fund in Addressing Internet Access Inequities in the US (ACM SIGCOMM, 2024)
- Decoding the Divide: Analyzing Disparities in Broadband Plans Offered by Major US ISPs (ACM SIGCOMM, 2023)
TurboTest
Learns when an Internet speed test has collected enough data to stop early. A companion workshop paper examines the policy implications of early termination.
- TURBOTEST: Learning When Less is Enough through Early Termination of Internet Speed Tests (USENIX NSDI, 2026)
- The Policy Implications of Early Termination in Internet Speed Testing (ACM IMC PRIME Workshop, 2025)
NetVibe
A longitudinal latency platform that connects infrastructure-level performance measurements with user-level experience.
Path II: Self-Driving Networks
Operating networks with limited infrastructure and expertise runs into a second data gap: the labeled, representative data needed to build trustworthy models is scarce, and the tools to generate it at scale are missing. We close that gap through controllable data generation, structured representation learning, and principled evaluation, so that networks can adapt, generalize, and stay trustworthy under dynamic conditions.
We build a programmable experimentation infrastructure (Pramana, NetReplica, NetGent, netUnicorn) for scalable and verifiable network research. These systems share one design principle: complex operational tasks become reliable and scalable when decomposed into smaller, verifiable components, a principle mirrored in BQT. On top of this substrate we build network foundation models (netFound, NetBurst). We also build tools to evaluate and interpret learned models. netFound and NetBurst are in tech-transfer at ESnet and Google within the DOE Genesis Mission.
We engage with the Energy Sciences Network (ESnet) to explore how AIOps frameworks can support large-scale scientific infrastructure, including adaptive traffic management, scalable telemetry analysis, and AI-driven operational intelligence in production environments.
Projects
Pramana
A composable, domain-specific backend for empirical networking research. Its intent specification separates what data to generate, where to generate it, and how to produce it, so one specification runs on any substrate.
- Pramana: A Composable, Domain-Specific Backend for Empirical Networking Research (ACM HotNets, 2026)
NetReplica
A bottleneck-aware emulation framework for controlled experimentation and network data generation.
- NetReplica: A Programmable Substrate for Bottleneck-Centric Network Data Generation (arXiv preprint arXiv:2507.13476, 2025)
- NetReplica: A Bottleneck-Regime Substrate for Reproducible Network Data Generation (Workshop on Network Traffic Classification — Reloaded (NTC-R), co-located with ACM CoNEXT, 2026)
NetGent
An agentic workflow automation system that compiles high-level specifications into deterministic, reusable execution pipelines.
- NetGent: Agent-Based Automation of Network Application Workflows (NeurIPS Workshop (MLforSys), 2025)
netUnicorn
A data-collection platform for distributed orchestration across heterogeneous environments; adopted as a reproducibility substrate.
netFound
A network foundation model that learns spatial, temporal, and hierarchical structure directly from packet-level telemetry.
- netFound: Foundation Model for Network Security (arXiv preprint arXiv:2310.17025, 2023)
NetBurst
Event-centric forecasting of bursty, intermittent network time series.
- NetBurst: Event-Centric Forecasting of Bursty, Intermittent Time Series (arXiv preprint arXiv:2510.22397, 2025)
Intrinsic Evaluation Framework
A representation-level method to validate network foundation models, which separates embedding quality from downstream task artifacts.
- Demystifying Network Foundation Models (NeurIPS (Dataset and Benchmarking Track), 2025)
Trustee
An interpretability framework for learning-based network systems.
- AI/ML for Network Security: The Emperor has no Clothes (ACM CCS, 2022)
Impact and engagement
Our broadband data work informs regulatory, legislative, and community decisions beyond the research literature.
Policy briefs and reports
- Create Independent Broadband Data for Public Accountability, Policy Brief, UC Presidential Faculty Fellows Program, 2026. Presented at the Congressional Staff Briefing “Building Trust in Public Institutions,” Dirksen Senate Office Building, Washington DC, May 2026.
- Broadband Affordability and the BEAD Program: Analysis and Policy Recommendations, MEDIA Project Phase 3 Report, USC Annenberg, 2026.
- Beyond Access: Broadband Affordability & Adoption, Research Study, Virginia Joint Commission on Technology and Science (JCOTS), 2025.
Amicus briefs
- Brief of Amici Curiae in Support of Respondents, Wisconsin Bell v. U.S. ex rel. Heath (No. 23-1127), Supreme Court of the United States, 2024.
Op-eds and public commentary
- What We Can’t See, We Can’t Fix, Benton Institute for Broadband & Society, 2026.
- Measuring Broadband Policy Success, Harvard Law Review Blog, 2024.
- NetGent: Agent-based Automation of Network Application Workflows, APNIC Blog, 2026.
Data contributions
- Broadband Competition and Pricing Strategies in California’s Urban Markets, California Public Advocates Office (Cal Advocates), CPUC, 2026. Competition and pricing analysis conducted using UCSB’s Broadband Query Tool (BQT).
- Dollars to Megabits: How We Uncovered Disparities in Internet Deals, The Markup, 2022.
Community and policy engagement
- BEAD Challenge Process (2024): provided evidence for Merit Network Inc. to challenge the FCC’s National Broadband Map for underserved regions in Michigan.
- City of Los Angeles (2024): our data helped catalyze a motion empowering CHRED to act against digital discrimination.
- #OaklandUndivided (2023 to present): used the BQT tool to identify underserved addresses and bridge data gaps for policymaking in Oakland, CA.
- Institute for Local Self-Reliance (ILSR) (2023 to present): gathered BQT data to strengthen the case for community networks.
- Affordable broadband for multi-dwelling units (2023 to present): assessing broadband offerings in MDUs using the BQT tool.
- Santa Barbara County (2023): report on the state of broadband affordability shared with the county.
Funding
Our research is supported by the National Science Foundation (NSF), the Department of Energy (DOE), the California Public Utilities Commission, the Commonwealth of Virginia, the Benton Foundation, the Pew Charitable Trusts, Google, Cisco, Intel, Verizon, and ViaSat.
| Project | Sponsor | Period | Amount | Status |
|---|---|---|---|---|
| Making Agentic AI Safe for DOE User Facilities | DOE / LBNL LDRD (Co-PI) | Oct 2026 | $250k | Active |
| Effectively Measuring Broadband Affordability in California | California Public Utilities Commission | Jan 2026 to Dec 2026 | $275k | Active |
| Marjorie & Charles Benton Opportunity Fund | Benton Foundation | Feb 2025 to Jan 2027 | Active | |
| Bridging the Representation–Semantics Gap for Production-Ready AI-Powered Network Operations | Cisco | Winter 2026 | $75k | Active |
| Low Infrastructure ML | Jul 2025 | $100k | Active | |
| Network Foundation Model for Enabling AI-powered Network Operations (AIOps) | Jul 2025 | $60k | Active | |
| Developing Generalizable ML Models for Diverse Learning Problems in Network Operations | NSF | Apr 2025 to Mar 2030 | $700k | Active |
| Characterizing Broadband Pricing in California | California Public Utility Commission | Summer 2025 | $125k | Active |
| Telemetry-driven Foundation Models for Self-Driving Networks | Cisco Research | Sep 2024 | $90k | Active |
| netFound: Network Foundation Model | DoE | Sep 2024 | Active | |
| IMR: MT: NetFlex: A Flexible Scalable & Privacy-Preserving Network Measurement Platform to Iteratively Collect Multi-modal Multi-view Network Data from Access Networks | NSF | Oct 2023 to Sep 2027 | $600k | Active |
| AIOps Roadmap Development for ESnet | DOE / LBNL (ESnet) | Jul 2025 to Aug 2026 | $65k | Completed |
| Measuring the Effectiveness of Digital Inclusion Approaches (MEDIA), Phase 3 | Pew Charitable Trusts (via USC) | Dec 2024 to Oct 2025 | $42k | Completed |
| Measuring the State of Broadband Affordability in Virginia | Virginia JCOTS | Jul 2025 to Dec 2025 | $30k | Completed |
| IMR: RI-P: Programmable Closed-loop Measurement Platform for Last-Mile Networks | NSF | Oct 2022 to Sep 2024 | $100k | Completed |
| IMR: MM-1A: ADDRESS: Augment, Denoise and Debias Crowdsourced Measurements for Statistical Synthesis of Internet Access Characterization | NSF (Co-PI) | Oct 2022 to Sep 2025 | $600k | Completed |
| The Estimation and Monitoring of Quality of Experience Delivered over Internet Services | ViaSat | Jan 2022 | $200k | Completed |
| CC* Integration-Large: Democratizing Networking Research in the Era of AI/ML | NSF | Oct 2021 to Sep 2025 | $1M | Completed |
| CC* Integration-Large: Bringing Code to Data: A Collaborative Approach to Democratizing Internet Data Science | NSF (Co-PI) | Oct 2021 to Sep 2025 | $1M | Completed |
| Workshop on Next-G Mobile Security | NSF | Jul 2020 to Jun 2023 | $49k | Completed |
| MLWiNS: RL-based Self-driving Wireless Network Management System for QoE Optimization | NSF & Intel | Aug 2020 to Jul 2024 | $820k | Completed |
| Scaling Cybersecurity Infrastructure using Programmable Data Planes | Verizon | Sep 2019 | $200k | Completed |