By area

Revision as of 14:57, 13 October 2025 by Keshav (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

This page has papers by area in rough reverse chronological order. Here are papers in Chronological order and a wordcloud.

Earth Systems

Satellite Remote Sensing

These papers develop computational methods to monitor and understand environmental change from space. My recent research has focused on creating better representations from remote sensing data—from developing temporal embeddings that capture surface dynamics (TESSERA) to addressing practical challenges like cloud corruption in time series. A major thread of my work involves using spaceborne lidar (GEDI) to quantify tropical forest disturbance and recovery, developing both the analytical frameworks and computational tools needed to assess forest health at scale. I'm also interested in bridging the gap between raw spectral data and actionable ecological insights, whether through end-to-end learning with physics-based models or multi-modal approaches for tracking restoration efforts in threatened ecosystems like the Atlantic Forest.


  • A. Holcomb, P. Burns, S. Keshav, D.A. Coomes, "Quantifying the Impacts of Tropical Forest Disturbance Using Repeat GEDI Footprints", Proc. American Geophysical Union, December 2024.
  • F. Begliomini, D.A. Coomes, S. Keshav, P. Brancalion, "Using Multi-Modal Orbital Remote Sensing to Track Ecological Restoration in the Atlantic Forest", Proc. American Geophysical Union, December 2024.
  • S. Ghasemitaheri, A. Holcomb, L. Golab, and S. Keshav, On the Data Quality of Remotely Sensed Forest Maps, Proc. VLDB Workshops, April 2023.


UAV and Ground-based Forest Monitoring

Alongside satellite-based monitoring, I've been developing practical tools for forest measurement using accessible technology. A key focus has been enabling accurate tree diameter estimation with mobile phones—moving from proof-of-concept demonstrations to robust algorithms and user-friendly apps that work with coarse optical depth maps. To support this and other computer vision applications in forestry, my students and I have created SPREAD, a large-scale synthetic dataset that provides high-fidelity training data for multiple forest vision tasks. This work is motivated by the need for scalable, low-cost methods to measure forest carbon and monitor forest health on the ground, complementing what we can see from space. By combining UAV imagery, mobile phone technology, and machine learning, I'm working to ease forest monitoring and make it feasible for researchers, land managers, and communities to track changes in their local forests.


Carbon Credits

I'm deeply engaged in addressing the credibility crisis in forest carbon markets and nature-based climate solutions. Much of my work focuses on strengthening the integrity of REDD+ credits through objective assessment methods—including using placebo tests to evaluate counterfactual baselines and developing optimal strategies to anticipate and mitigate the risk of carbon credit reversals. I've contributed to the development of the PACT tropical forest accreditation methodology and explored how to realize the social value of impermanent carbon credits while being forthright about the challenges facing current carbon markets. A key thread of this research involves leveraging computer science and digital technologies to create more robust, comparable, and transparent carbon assets that can support both forest restoration and climate mitigation. Ultimately, I'm working toward trusted marketplaces for nature-based solutions where the environmental claims are verifiable, the risks are properly managed, and the incentives genuinely support forest conservation rather than undermining it.

Climate Science and Policy

My work in climate science and policy spans from foundational questions about data and modeling to practical pathways for decarbonization. I've explored novel approaches to generating long-term climate datasets using expired weather forecasts and contributed to policy discussions on carbon offsetting and nature-based climate solutions. Recognizing that the computing and communications infrastructure itself has climate impacts, I've worked on understanding how to make internet research more climate-friendly. A significant portion of my research has focused on the energy transition, using agent-based modeling to understand the adoption dynamics of solar photovoltaics, battery storage, and electric vehicles across different jurisdictions, and exploring how policy design—from pricing mechanisms to demand response systems—can accelerate decarbonization. I've also applied data-driven methods to understand public discourse around climate action, including sentiment analysis during the COVID-19 recovery period. Throughout this work, I'm interested in how computational tools, behavioral insights, and thoughtful policy design can work together to address the climate crisis.

Energy Systems

Solar PV and Battery Storage

I've spent over a decade working on the technical and economic challenges of integrating solar photovoltaics and battery storage into homes and energy systems. A central focus has been developing robust, practical methods for sizing PV and storage systems—determining how much solar capacity and battery storage is needed to meet household energy needs, particularly as electric vehicles become part of the home energy ecosystem. My recent work includes tools like SOPEVS and SolExplore that optimize the sizing and operation of PV-EV-integrated homes using successive refinement approaches, building on years of research comparing different sizing methodologies and developing synthetic data generators to create realistic usage patterns for system design.

Underpinning these optimization tools is extensive work on battery modeling and control strategies. I've developed tractable lithium-ion storage models that balance accuracy with computational efficiency, enabling them to be used in large-scale energy system optimization. This includes simple specification-based models derived from manufacturer data sheets, adaptive control strategies using neural networks, and practical approaches for battery operation that work in real-world conditions. I've also explored innovations like dynamic storage resizing, grid-friendly solar panel orientations to flatten the "duck curve," and strategies for firming solar power to make it more reliable and dispatchable.

Beyond the technical optimization, I'm interested in understanding what drives adoption of these technologies and how policy shapes deployment. Through agent-based modeling, I've compared solar and battery adoption patterns across different jurisdictions like Ontario and Germany, examining how factors like electricity pricing, incentives, and regulatory frameworks affect profitability and uptake. I've also worked on practical diagnostic tools using data-driven approaches to identify problems with installed solar systems. This body of work aims to bridge the gap between theoretical optimization and real-world deployment, helping accelerate the transition to distributed renewable energy systems.

  • J. Gschwind and S. Keshav, SolExplore: A Successive Refinement Approach for Sizing of PV and Storage Systems in EV-Enabled Homes, To appear, Proc. ACM BuildSys, November 2025.
  • Y. Ghiassi-Farrokhfal, S. Keshav, and C. Rosenberg, Firming Solar Power, Extended Abstract/Poster, Proc. ACM SIGMETRICS, June 2013.

Electric Vehicle Systems

My research on electric vehicles spans from understanding real-world usage patterns to developing practical tools for EV integration into energy systems. A major thread has been the WeBike project, where I collected and analyzed extensive data on electric bicycle usage, examining how e-bikes are actually used in practice and comparing these patterns to electric car usage. This work led to insights on range prediction, data stream management challenges, and the factors that influence adoption. I've also developed non-intrusive methods like EVSense for detecting EV charging without requiring smart meters or instrumentation, and explored how blockchain technology can help mitigate trust issues in public charging infrastructure. On the economic and planning side, I've used agent-based modeling to understand EV adoption dynamics—particularly the relative importance of price versus driving range—and analyzed the return on investment for taxi companies transitioning to electric fleets. I've also worked on optimizing EV charging to take advantage of solar generation and developed methods for sizing finite vehicle pools, all aimed at making the transition to electric mobility more practical and economically viable.


Building and Datacenter Energy Management

My work on building energy management has evolved from developing innovative personal thermal comfort systems to addressing broader challenges of building efficiency and climate resilience. A core thread has been the SPOT series of personalized heating and cooling systems, which allow individuals to maintain thermal comfort while reducing centralized HVAC loads—an approach I've extended by studying how personal environmental comfort systems interact with building-wide climate control. I've developed occupancy detection methods, self-calibrating smart lighting systems, and model predictive control strategies for HVAC that operate across multiple time scales. To support research in this area, I created Beobench, a toolkit that provides unified access to building simulations for reinforcement learning, making it easier to develop and test intelligent building control strategies. Throughout this work, I've also contributed several open datasets on occupancy, thermal comfort, light levels, and snow-covered solar panels to enable broader research in building energy systems.

More recently, my focus has expanded to understanding how buildings perform during climate extremes and how computing infrastructure can be made more energy-efficient. I've developed machine learning approaches for building-level heat risk mapping and fine-grained mapping of urban energy demand during heatwaves, working to assess building liveability through metrics like "activity hours" that capture when indoor conditions remain tolerable during extreme heat events. On the computing side, I've investigated how hybrid heterogeneous clusters can dramatically reduce the energy consumption of large language model inference workloads by developing workload-based energy models and offline optimization strategies. This work reflects a broader interest in making both our built environment and our digital infrastructure more energy-efficient and resilient in the face of climate change.

  • Y. Aussat, A. Rosmanis, S. Keshav, A Power-Efficient Self-Calibrating Smart Lighting System, Energy and Buildings Journal version Author's preprint, January 2022.
  • R. Kalaimani, S. Keshav, and C. Rosenberg, "Multiple Time-scale Model Predictive Control for Thermal Comfort in Buildings," Poster in ACM e-Energy 2016.

Smart Grid and Demand Response

My work on smart grids has focused on developing distributed control algorithms and market mechanisms that enable flexible, responsive electricity systems. A major contribution has been creating real-time distributed control strategies for electric vehicle charging—addressing both the static and dynamic aspects of coordinating large fleets of EVs to avoid grid congestion while meeting user needs. This work, which won Best Paper awards, extends to using EV charging control to provide building load flexibility and exploring how vehicles can provide frequency regulation services through optimal contracting. I've also investigated fundamental concepts like load elasticity and demand response, developing mechanisms such as temperature setpoint markets and analyzing how elastic loads can reduce peak demand and carbon footprints in the residential sector.

Beyond control algorithms, I've worked on the infrastructure and trust mechanisms needed for modern electricity systems. I applied concepts from Internet congestion control and teletraffic theory to power distribution, showing how networking principles can help "green and smarten" the electrical grid. This includes work on efficient demand assignment in multi-connected microgrids, optimal electricity allocation algorithms, and the impact of storage on residential distribution systems. I've also addressed trust and transparency issues by developing efficient, anonymous renewable energy certificate systems using cryptographic techniques, and created tools for processing smart meter data and energy analytics. Throughout this research, I've contributed datasets on high-resolution load measurements and critiqued existing pricing schemes like Ontario's time-of-use tariffs, aiming to make electricity systems both more efficient and more equitable.

  • S. Alamdari, T. Biedl, T. M. Chan, E. Grant, K.R Jampani, S. Keshav, A. Lubiw and V. Pathak, Smart-grid Electricity Allocation via Strip Packing with Slicing, Proc. WADS 2013, August 2013.
  • S. Keshav and C. Rosenberg, On Load Elasticity, in IEEE Comsoc MMTC-E letter, #8, Vol. 7 Nov. 2012.

Computer Networking

My foundational work in computer networking centers on congestion control and fair resource allocation—problems I tackled in my PhD thesis at UC Berkeley, which won both the Sakrison Prize and later a SIGCOMM Test-of-Time Award. I developed fair queueing algorithms and control-theoretic approaches to flow control that became influential in understanding how to manage congestion in high-speed networks. This early work extended into ATM networks, quality of service mechanisms, and rate-based service disciplines, where I contributed to protocol design, network measurement tools, and the ENTRAPID protocol development environment. I also worked on fundamental questions of network architecture, exploring axiomatic foundations for communication and reflecting on paradoxes and design principles that shape how networks evolve. This theoretical foundation has informed all my subsequent networking research.

A major thread of my work has been in wireless networks and mobile systems, spanning from enterprise WiFi optimization to novel RFID sensing applications. I developed CENTAUR, a hybrid data path architecture for centralized WLANs that won a Best Paper Award at MobiCom, and worked extensively on interference mitigation, spectrum sensing, and self-managing wireless architectures. More recently, I've pioneered the use of commodity RFID systems for environmental sensing—developing battery-free soil moisture sensors for sustainable greenhouse monitoring, creating in-class response systems, and demonstrating practical "RFID hacking" techniques that extend the capabilities of these ubiquitous tags. I've also contributed to understanding energy consumption in mobile devices and developed systems like OmniVoice for mobile communications in resource-constrained settings.

My work on opportunistic and delay-tolerant networks was motivated by bridging the digital divide and providing internet access in challenging environments. I led the development of KioskNet, a system that used "mechanical backhaul"—physically transporting data on storage devices carried by buses—to provide low-cost internet access to rural kiosks in developing regions. This work required rethinking fundamental networking assumptions: developing robust communication protocols for intermittent connectivity, addressing security challenges for disconnected nodes, creating fair scheduling algorithms for data ferrying networks, and understanding vehicular opportunistic communication patterns. The insights from this work extend beyond developing regions to any scenario where connectivity is intermittent or opportunistic, and helped establish design principles for robust communication in constrained computing environments.

My more recent networking research has tackled challenges in data center networks and distributed systems, including blockchain technologies. On the data center side, I've worked on cost-effective network upgrades using heterogeneous equipment (LEGUP), low-latency network designs (Quartz), and optimization frameworks for unstructured topologies (REWIRE), as well as exploring how to make data centers more energy-efficient through VM consolidation strategies. In the blockchain space, I've developed FastFabric to dramatically scale Hyperledger Fabric's transaction throughput, explored hybrid execution approaches, created TimeFabric for trusted time in distributed ledgers, and designed the Canopus consensus protocol for massively parallel agreement. I've also applied blockchain to practical problems like creating trustworthy renewable energy certificates and digital carbon assets, while maintaining a skeptical perspective on when blockchain is—and isn't—the right solution.

Wireless Networks and Mobile Systems

Blockchain and Consensus

Network Architecture and Protocols

Data Center Networks

Opportunistic and Delay-Tolerant Networks

Peer-to-Peer Systems

Network Measurement and Performance

ATM and QoS

  • M. Grossglauser and S. Keshav, On CBR Service, Proc. IEEE INFOCOM '96, March 1996.

Security and Privacy

Multimedia and Video Systems

  • A.E. Kaplan and S. Keshav, Talking Heads Made Simple, Presented at the 1993 International Worshop on Facial Animation, Philadelphia, Pennsylvania, November 1993.

Network Tools and Development

Congestion Control and Flow Control

Miscellaneous

Beyond my technical research, I've contributed to the computer science research community through education, community building, and reflections on how we organize and conduct research. I've written two textbooks on computer networking—"An Engineering Approach to Computer Networking" and "Mathematical Foundations of Computer Networking"—and created widely-used educational resources including my guide "How to Read a Paper," which has helped countless students learn to navigate academic literature. I've also developed tutorials on topics like model predictive control to bridge different research communities. My service to the field includes leadership roles as SIGCOMM Chair and editor of ACM's Computer Communication Review, experiences I've reflected on in invited papers that examine the health and evolution of our research community.

I care deeply about how we organize and improve our research processes. I've written extensively about scaling academic publication to match the growth of our field, proposed recommendations for designing effective hybrid conferences in the post-pandemic era, and advocated for better research practices through pieces like "The Value of Weekly Reports." I've also engaged with policy and public discourse, including providing testimony to the U.S. Senate on text messaging pricing and contributing to interdisciplinary venues like the Cambridge Journal of Law, Politics, and Art. This work reflects my belief that computer scientists have both an opportunity and responsibility to contribute beyond our immediate technical domains—improving how we work together as a community and engaging with the broader societal implications of our research.

  • S. Keshav, Taking account, Editorial in ACM SIGCOMM Computer Communication Review, July 2009.
  • S. Keshav, The cost of text messaging, Testimony at the hearing on Cell Phone Text Messaging Rate Increases and the State of Competition in the Wireless Market held by the Senate Subcommittee on Antitrust, Competition Policy and Consumer Rights, June 16, 2009
  • S. Keshav, Bubbles, Editorial in ACM SIGCOMM Computer Communication Review, April 2009.
  • S. Keshav, An Engineering Approach to Computer Networking, Addison-Wesley, 1997.
Retrieved from "https://svr-sk818-web.cl.cam.ac.uk/keshav/wiki/index.php?title=By_area&oldid=1831"