skip to content

Department of Computer Science and Technology

 

EEG Group Research Archives


Not-quite-so-broken TLS in OCaml 2014

Researcher:  Hannes MehnertDavid Kaloper-Mersinjak

Supervisor: Anil MadhavapeddyPeter Sewell

Summary:

This internship focused on developing nqsb-TLS, a re-engineered approach to TLS protocol specification and implementation that serves as both a test oracle for conformance and a usable implementation, aiming to address common security flaws through a modular design and memory-safe programming, while achieving comparable performance to traditional implementations like OpenSSL.

Please click here to view the paper.


Control Flow Analysis for Privilege Separation 2011

Researcher: Chris Harding and Ross Mcllroy

Supervisor: Anil Madhavapeddy, Robert M Watson

Summary:

In the summer of 2011, interns Chris Harding and Ross McIlroy developed tools for the CTSRD/SOAPP project, with McIlroy creating privgrind to track data address interactions using Valgrind, and Harding building a visualizer to generate control flow graphs for compartmentalization efforts, although their work was only partially documented despite its significance.

Please click here to view the paper.


 Consolidating Trust for Client Groups that Use TLS to Secure Connections 2011

Researcher: William Morland

Supervisor: Anil Madhavapeddy, Robert M Watson

Summary:

In the summer of 2011, intern William Morland worked on the CTSRD/CHERI project by enhancing the gxemul MIPS simulator and the LLVM/MIPS backend, validating and improving the CHERI test suite, while also contributing to discussions on LLVM development and producing a poster for his work at the CTSRD project meeting.

Please click here to view the paper.


Datacentre /LLM Energy Use 2024

Researcher: Grant Wilkins

Supervisor: Srinivasan Keshav, Richard Mortier

Summary:

The rapid adoption of large language models (LLMs) has led to significant advances in natural language processing and text generation. However, the energy consumed through LLM model inference remains a major challenge for sustainable AI deployment. To address this problem, we model the workload-dependent energy consumption and runtime of LLM inference tasks on heterogeneous GPU-CPU systems. By conducting an extensive characterization study of several state-of-the-art LLMs and analyzing their energy and runtime behaviour across different magnitudes of input prompts and output text, we develop accurate (R^2>0.96) energy and runtime models for each LLM. We employ these models to explore an offline, energy-optimal LLM workload scheduling framework. Through a case study, we demonstrate the advantages of energy and accuracy-aware scheduling compared to existing best practices.

Please click here to view the paper.


Topic: Buildings 2024

Researcher: Livia Capol

Supervisors: Zoltan Nagy, Srinivasan Keshav

Summary:

As climate change increases the frequency and severity of heatwaves, understanding their impact on the well-being, health, and permissible activities of building occupants becomes crucial. Existing work focuses primarily on the thermal comfort of an average individual, neglecting both the impact of extreme heat on liveability and variability in the resilience of different population groups. To address these gaps, we introduce the Activity hours (Ah) metric, which quantifies the liveability of indoor environments during heatwaves while taking into account air temperature, humidity, occupant age, permissible activity levels, and duration of exposure. We also present the Heatalyzer tool that allows the computation of Ah for different geographies, building archetypes, and heatwave durations. Through a case study of residential housing in London, we compare Ah with established thermal comfort metrics, highlighting Ah's ability to quantify heatwave impacts on occupant liveability for different building types and demographic groups. Our results are made widely accessible through the Heatalyzer dashboard, an intuitive website that enables London residents to evaluate their exposure to heat-related risks.

Please click here to view the paper.


Topic: Insect Monitoring 2024

Researchers: Sachin Matthews, Matteo Redana

PIs: Lynn Dicks, Srinivasan Keshav

Summary from Matteo Redana:

Insects have become a focus for concern about biodiversity loss and conservation, following studies showing strong signals of biomass, abundance and diversity decline (e.g., Hallmann et al., 2017; Klink et al., 2020). Insects provide crucial ecosystem services (e.g., pollination, decomposition) on which human actives directly rely, further, they play a key role in food chains through their interaction with both upper and lower trophic levels. More recent studies have found that trends in insect biomass are better explained if complex, non-linear, relationships with the biophysical structure of their environments, including climatic conditions, land use change and habitat structure, are accounted for (e.g., Müller et al., 2023).

Please click here for more details.


Topic: Real-Time Wildlife Monitoring 2024

Researcher: Tom Ratsakatika

Supervisors: Ruben Iosif, Srinivasan Keshav

Summary:

Machine Learning Analysis of Camera Trap Photos for an Automated Wildlife Alert System in the Romanian Carpathian Mountains.

Please click here for more details.


Topic: Crop Detection 2024

Researcher: Maddy Lisaius

Supervisors: Clement Atzberger, Andrew Blake, David CoomesSrinivasan Keshav

Summary:

Satellite-based monitoring is a key tool for supporting global food security and natural resource management but is challenged by cloud corruption and a lack of labelled training data. To address these issues, self-supervised learning (SSL) techniques have been developed that first learn representations from almost limitless available unlabeled data, before using labelled samples for a specific downstream task. As the learned representations detect, integrate, and compress information in the dataset in a fully unsupervised manner, the downstream tasks require only small labelled datasets. In this study, we present spectral–temporal Barlow Twins (ST-BT), a new pixelwise SSL architecture that generates useful representations designed to be invariant to extensive cloudiness. We demonstrate that ST-BT representations enable stable and high F1 scores on the downstream task of crop classification even with cloud cover reaching 50% of available dates and using only a few labelled samples. The ST-BT representations achieve maximum F1 scores of 0.94 and 0.90 on the two benchmark classification datasets used. These results indicate that ST-BT can create useful representations of pixelwise multispectral Sentinel-2 time series despite cloud corruption.

Please click here to view the paper.


Topic: TMF 2.0

Researcher: Patrick Ferris

Supervisors: Michael Dales, Anil MadhavapeddySrinivasan KeshavThomas Swinfield

PIs: Anil Madhavapeddy, Srinivasan Keshav

Summary:


Topic: Crawling Grey Literature for Conservation Evidence 2024

Researcher: Shrey Biswas and Kacper Michalik

Supervisors: Anil MadhavapeddySadiq Jaffer

Summary:

The Conservation Evidence Copilots project aims to develop a web crawler to search, extract, and make accessible valuable conservation intervention information from grey literature sources, complementing academic research and enhancing the knowledge base for researchers and practitioners.

Please click here for more details.


Topic: Evaluating RAG Pipelines for Conservation Evidence 2024

Researcher: Radhika Iyer

Supervisors: Anil MadhavapeddySadiq Jaffer

Summary:

The Conservation Evidence Copilots project aims to evaluate and optimize RAG (Retrieval Augmented Generation) pipelines for synthesizing conservation evidence, with the goal of developing best practices and recommendations for using these systems to inform policy and practice through natural language query interfaces.

Please click here for more details.


Topic: Reverse Emulating Agent-based Models for Policy Simulation 2023

Researcher: Pedro Sousa

Supervisors: Anil MadhavapeddySadiq Jaffer

Summary:

This project explores reverse emulation using probabilistic machine learning to streamline policy design in agent-based models, enabling more efficient prediction of input parameters that yield desired outcomes and improving upon the traditional iterative parameter tuning process.

Please click here for more details.


Topic: Scalable Agent-based Models for Optimized Policy Design 2022

Researcher: Sharan Agrawal 

Supervisors: Anil Madhavapeddy, Srinivasan Keshav

Summary:

This project develops VDSK-B, a novel agent-based model integrating climate, economy, and biosphere elements, along with SalVO, a scalable and optimizable framework, to address global climate change and biodiversity loss through improved policy design at planetary scales.

Please click here for more details.


Topic: Exploring Concurrency in Agent-Based Modelling with Multicore OCaml 2021

Researcher: Martynas Sinkievič 

Supervisors: Anil Madhavapeddy

Summary:

This project ported the TROLL agent-based forest simulator to OCaml, applying a functional programming style and introducing concurrency, to explore the challenges of refactoring scientific computing codebases and leverage Multicore OCaml's parallelization capabilities for more efficient and accurate ecosystem modeling.

Please click here for more details.


Topic: Assessing High-performance Lightweight Compression Formats for Geospatial Computation 2023

Researcher: Omar Tanner 

Supervisors: Anil MadhavapeddySadiq Jaffer

Summary:

This project aims to enhance geospatial data processing by implementing SIMD-optimized compression techniques for GeoTIFF data, leveraging CPU capabilities to improve cache locality, reduce memory bottlenecks, and accelerate data access.

Please click here for more details.


Topic: Legal Perspectives on Integrity Issues in Forest Carbon  2024

Researcher: Sophie Chapman

Supervisors: Anil MadhavapeddySadiq Jaffer

Summary:

This project explores a novel legal framework for forest carbon credits that distinguishes between carbon tenure and carbon rights, aiming to address integrity issues in carbon finance and promote more accurate, just, and effective benefit-sharing arrangements for forest conservation.

Please click here for more details.


Topic: Legal Perspectives on Integrity Issues in Forest Carbon  2024

Researcher: Sophie Chapman

Supervisors: Anil Madhavapeddy,  Eleanor Toye Scott

Summary:

This project explores a novel legal framework for forest carbon credits that distinguish between carbon tenure and carbon rights, aiming to address integrity issues in carbon finance and promote more accurate, just, and effective benefit-sharing arrangements for forest conservation.

Please click here for more details.


Topic:  Spatial Name System

Researcher: Ryan Gibbs

Supervisor: Anil MadhavapeddyJon Crowcroft

Summary:

The Spatial Name System (SNS) is proposed as an alternative network architecture to address the limitations of the current Domain Name System in supporting emerging hardware like IoT devices and AR headsets. By utilizing a device's location as its unique identifier, SNS aims to enable reliable, low-latency, secure, and private communication for ubiquitous computing and augmented reality interactions.

Please click here for more details.