Current Projects

Detailed information about my active research projects, focusing on computational optimization, distributed systems, and machine learning applications to scientific computing problems.

Active Research Projects

1. Reinforcement Learning for Strategic Game Optimization: Secret Hitler AI Agent

Project Overview
This project explores the application of reinforcement learning (RL) techniques to develop an AI agent capable of playing the social deduction game Secret Hitler at an expert level. The project combines large language models (LLMs) with RL training and Low-Rank Adaptation (LoRA) fine-tuning to create an agent that can learn optimal strategies for both fascist and liberal roles.

Objectives

  • Train an LLM-based agent using reinforcement learning to master Secret Hitler gameplay

  • Implement LoRA adaptation techniques for efficient model fine-tuning

  • Develop optimal strategies for information gathering, deception, and coalition building

  • Create a framework for evaluating strategic decision-making in incomplete information environments

Technical Approach

  • Base Model: Large language model adapted for strategic reasoning

  • Training Method: Reinforcement learning with policy gradient methods

  • Optimization: LoRA (Low-Rank Adaptation) for parameter-efficient fine-tuning

  • Environment: Custom Secret Hitler game simulation with multi-agent interactions

  • Evaluation: Tournament-style play against human and AI opponents

Current Status: Research and development phase Expected Outcomes: Novel insights into RL applications for social deduction games and strategic AI decision-making

Applications

  • Game theory research

  • Strategic AI development

  • Multi-agent systems

  • Social interaction modeling

2. MCMC-Enhanced LSA Kernels for Reusable Semantic Analysis

Project Overview
This project focuses on developing reusable computational kernels by applying Markov Chain Monte Carlo (MCMC) methods to Latent Semantic Analysis (LSA) models. The goal is to create more robust and transferable semantic representation systems that can be efficiently applied across different domains and datasets.

Objectives

  • Implement MCMC sampling techniques for LSA model parameter estimation

  • Develop reusable kernel architectures for semantic analysis tasks

  • Improve model generalization across different text corpora and domains

  • Create efficient sampling strategies for large-scale semantic spaces

Technical Approach

  • Core Method: Markov Chain Monte Carlo sampling for LSA parameter inference

  • Architecture: Modular kernel design for cross-domain applicability

  • Optimization: Efficient sampling algorithms (Gibbs sampling, Metropolis-Hastings)

  • Evaluation: Cross-domain semantic similarity and classification tasks

  • Implementation: High-performance computing optimizations

Current Status: Algorithm development and initial testing phase Expected Outcomes: More robust semantic models with improved transferability and computational efficiency

Applications

  • Natural language processing

  • Information retrieval systems

  • Cross-domain semantic analysis

  • Document classification and clustering

3. Radio Wave Propagation Modeling for Atmospheric Mapping

Project Overview
This interdisciplinary project combines radio frequency engineering with atmospheric science to develop advanced models for radio wave propagation. The research aims to use radio signal characteristics to map atmospheric conditions and improve our understanding of atmospheric dynamics and structure.

Objectives

  • Develop accurate models for radio wave propagation through varying atmospheric conditions

  • Create mapping techniques that use radio signal data to infer atmospheric properties

  • Improve weather prediction and atmospheric monitoring capabilities

  • Establish relationships between radio wave behavior and atmospheric parameters

Technical Approach

  • Modeling Framework: Physics-based propagation models with machine learning enhancement

  • Data Sources: Radio frequency measurements, atmospheric sensor data, satellite observations

  • Methods: Signal processing, inverse modeling, atmospheric physics simulations

  • Validation: Comparison with traditional atmospheric measurement techniques

  • Implementation: High-performance numerical simulations

Current Status: Literature review and model architecture design phase Expected Outcomes: Novel atmospheric mapping techniques and improved radio communication systems

Applications

  • Weather forecasting and climate monitoring

  • Radio communication system optimization

  • Atmospheric research and modeling

  • Remote sensing technology development

Project Timeline and Milestones

Short-term Goals (3-6 months)

  • Complete initial implementations for Secret Hitler RL agent

  • Develop MCMC sampling framework for LSA models

  • Establish baseline radio propagation models

Medium-term Goals (6-12 months)

  • Conduct comprehensive evaluations of all three projects

  • Publish preliminary findings and seek peer review

  • Develop proof-of-concept demonstrations

Long-term Goals (12+ months)

  • Submit research papers to relevant conferences and journals

  • Explore commercial applications and partnerships

  • Expand projects based on initial results and community feedback

Collaboration Opportunities

These projects offer excellent opportunities for collaboration with researchers in:

  • Machine Learning and AI: RL techniques, LLM fine-tuning, MCMC methods

  • Game Theory: Strategic decision-making, multi-agent systems

  • Atmospheric Science: Radio meteorology, atmospheric modeling

  • Signal Processing: RF engineering, propagation modeling

  • High-Performance Computing: Optimization and scalability

Funding and Resources

Currently seeking funding opportunities through:

  • Research grants for AI and machine learning applications

  • Atmospheric science and meteorology research programs

  • Industry partnerships for practical applications

  • Academic collaborations with relevant departments

🏭 LUStores: Intelligent Stocking System for Lancaster University Physics

Duration: 2024 - 2025 (18 months)
Collaborators: Lancaster University Physics Department
Type: Applied research and systems development

Project Overview

Development of an intelligent inventory management system for the Lancaster University Physics Department, incorporating predictive analytics and automated procurement workflows. This project applies machine learning and optimization techniques to solve real-world resource management challenges in academic research environments.

Specific Aims

  1. Automated Inventory Tracking

    • Approach: Computer vision-based item recognition and RFID integration

    • Timeline: Q3 2024 - Q1 2025

    • Status: Prototype deployed, testing CV recognition accuracy

  2. Predictive Demand Modeling

    • Approach: Time-series analysis and usage pattern recognition

    • Timeline: Q4 2024 - Q2 2025

    • Status: Historical data analysis complete, models in development

  3. Optimization Engine

    • Approach: Multi-objective optimization for cost minimization and availability maximization

    • Timeline: Q1 2025 - Q3 2025

    • Status: Initial optimization framework implemented

Key Achievements to Date

  • System Architecture: Complete system design and database schema deployed

  • Integration: Successfully integrated with existing departmental procurement systems

  • User Interface: Web-based management interface with real-time inventory tracking

Technical Innovations

  • Real-time Analytics: Dashboard providing instant visibility into stock levels and usage patterns

  • Automated Reordering: procurement suggestions based on usage forecasts

  • Multi-criteria Optimization: Balancing cost, availability, and storage constraints

Expected Outcomes

  • Operational Impact: 30-40% reduction in stockouts and 20% cost savings

  • Software: Open-source inventory with long term security updates

🌐 SuperDARN Chord-DHT: Scalable Atmospheric Data Processing

Duration: 2024 - 2026 (2 years)
Collaborators: SuperDARN research community
Type: Distributed systems research for scientific computing

Project Overview

Development of a chord-based distributed hash table (DHT) backend for processing SuperDARN (Super Dual Auroral Radar Network) atmospheric data. This project addresses the growing computational and I/O challenges in processing large-scale radar data for space weather research through novel distributed computing approaches.

Research Objectives

Primary Goal: Create an I/O efficient, scalable processing framework for SuperDARN data that enables real-time analysis and historical data mining across the global radar network.

Technical Approach

  1. Chord-DHT Architecture

    • Design: Distributed hash table optimized for scientific data workflows

    • Timeline: Q4 2024 - Q2 2025

    • Status: Core DHT implementation complete, testing data distribution algorithms

  2. I/O Optimization Pipeline

    • Approach: Custom data serialization and compression for radar time-series data

    • Timeline: Q1 2025 - Q3 2025

    • Status: Benchmarking compression algorithms, developing streaming interfaces

  3. Scalable Processing Framework

    • Approach: Map-reduce style processing with fault tolerance and load balancing

    • Timeline: Q2 2025 - Q4 2025

    • Status: Initial processing node implementation, designing task scheduling

Key Innovations

  • Data Locality Optimization: Intelligent data placement considering temporal and spatial correlation in radar measurements

  • Adaptive Load Balancing: Dynamic workload distribution based on computational requirements and data access patterns

  • Fault-Tolerant Processing: Resilient processing pipeline with automatic recovery and data replication

Progress Report

Completed:

  • βœ… Chord protocol implementation and testing

  • βœ… SuperDARN data format analysis and parsing libraries

  • βœ… Initial performance benchmarks on historical datasets

In Progress:

  • πŸ”„ Integration with existing SuperDARN data pipelines

  • πŸ”„ Development of query optimization for common analysis patterns

  • πŸ”„ Collaborative testing with multiple SuperDARN institutions

Upcoming:

  • πŸ“… Deployment on SuperDARN production systems (Q3 2025)

  • πŸ“… Performance evaluation with global radar network data (Q4 2025)

  • πŸ“… Conference presentation at Space Weather symposium

Expected Impact

  • Research Acceleration: Enable new classes of large-scale atmospheric studies

  • Community Benefit: Provide scalable infrastructure for the global SuperDARN community

  • Technical Contribution: Novel DHT applications to scientific data processing

Related Links

πŸ€– Tactile Linear Sum Assignment with ML-Guided Gradients

Duration: 2024 - 2025 (15 months)
Type: Fundamental research in optimization and machine learning
Focus: Novel approaches to NP-hard optimization through tactile computing paradigms

Project Overview

Investigation of linear sum assignment problems through tactile and haptic computing approaches, using machine learning models to guide gradient-based optimization. This project explores whether physical intuition and haptic feedback can improve optimization algorithm design and performance.

Research Questions

  • Can tactile/haptic interfaces provide intuitive insights into the optimization landscape of assignment problems?

  • How can machine learning models learn and reproduce successful β€œtactile strategies” for optimization?

  • What are the fundamental connections between physical manipulation and mathematical optimization?

Methodological Innovation

  1. Tactile Optimization Interface

    • Approach: Haptic feedback systems for visualizing and manipulating assignment cost matrices

    • Timeline: Q4 2024 - Q2 2025

    • Status: Initial haptic hardware setup complete, developing cost matrix visualization

  2. ML-Guided Gradient Methods

    • Approach: Neural networks trained on tactile optimization sessions to guide gradient descent

    • Timeline: Q1 2025 - Q3 2025

    • Status: Collecting training data from human optimization sessions

  3. Hybrid Optimization Framework

    • Approach: Integration of learned tactile strategies with traditional assignment algorithms

    • Timeline: Q2 2025 - Q4 2025

    • Status: Designing architecture for human-AI optimization collaboration

Technical Components

Hardware Integration:

  • Haptic feedback devices for cost matrix manipulation

  • Real-time visualization of assignment solutions and costs

  • Multi-modal interface combining visual, tactile, and auditory feedback

Machine Learning Pipeline:

  • Sequence models for learning optimization trajectories

  • Reinforcement learning for strategy development

  • Transfer learning across different problem instances

Optimization Algorithms:

  • Modified Hungarian algorithm with ML-guided pivoting

  • Gradient-based methods with learned step size adaptation

  • Hybrid human-AI optimization protocols

Preliminary Findings

Human Optimization Patterns:

  • Identification of consistent human strategies for small assignment problems

  • Analysis of tactile feedback preferences during optimization tasks

  • Characterization of successful vs. unsuccessful optimization approaches

ML Model Performance:

  • Initial neural networks showing ability to reproduce human optimization strategies

  • Promising results in guided gradient methods for medium-scale problems

  • Evidence for transferability of learned strategies across problem instances

Expected Contributions

  • Theoretical: New insights into the connection between physical intuition and mathematical optimization

  • Algorithmic: Novel ML-guided optimization methods for assignment problems

  • Interface Design: Tactile computing paradigms for optimization problem solving

  • Publications: Conference papers on human-AI collaboration in optimization

Challenges and Risks

Technical Challenges:

  • Scaling tactile approaches to larger problem instances

  • Ensuring ML models generalize beyond training scenarios

  • Integration complexity between haptic hardware and optimization software

Research Risks:

  • Tactile approaches may not provide advantages over traditional methods

  • Human optimization strategies may not be systematically learnable

  • Hardware limitations may constrain experimental scope

Related Links

Cross-Cutting Research Activities

πŸ”— Optimization-Systems Integration

Common themes across projects involve the application of optimization theory to practical systems:

  • Shared Algorithms: Linear sum assignment methods developed for tactile research applied to inventory optimization in LUStores

  • Performance Benchmarking: Consistent evaluation frameworks across SuperDARN and LSA projects for scalability analysis

  • Hardware Acceleration: GPU optimization techniques shared between DHT processing and assignment algorithms

πŸ’» Software Development and Open Source

Commitment to reproducible research through open-source software development:

Active Development:

  • LUStores Framework: Modular inventory management system for academic institutions

  • SuperDARN-Chord: Distributed processing framework for atmospheric data

  • TactileLSA Toolkit: Experimental framework for tactile optimization research

Community Contributions:

  • Documentation: Comprehensive guides and tutorials for all developed tools

  • Testing: Automated CI/CD pipelines and extensive test suites

  • Collaboration: Active engagement with user communities and external contributors

οΏ½ Data Management and Sharing

Responsible data stewardship across all projects:

Data Standards:

  • Standardized formats for optimization problem instances and benchmarks

  • Metadata schemas for experimental results and performance evaluations

  • Privacy-preserving protocols for human subject data in tactile research

Infrastructure:

  • Distributed storage solutions for large-scale radar data

  • Version-controlled datasets for reproducible research

  • Secure data sharing protocols for collaborative projects

Resource Management and Infrastructure

Computational Resources

Current Allocations:

  • University HPC Cluster: GPU nodes for LSA algorithm development and testing

  • Cloud Infrastructure: Distributed testing environment for SuperDARN Chord-DHT

  • Local Development: High-performance workstations for tactile research and prototyping

Resource Utilization:

  • LUStores Development: 20% (primarily web services and database operations)

  • SuperDARN Chord-DHT: 50% (distributed systems testing and data processing)

  • Tactile LSA Research: 20% (ML model training and optimization experiments)

  • Method development and benchmarking: 10%

Specialized Equipment

Tactile Research Hardware:

  • Haptic Devices: Force feedback interfaces for optimization visualization

  • Sensors: Pressure and motion tracking for human optimization behavior

  • Display Systems: Multi-modal feedback interfaces for assignment problem visualization

Development Infrastructure:

  • Testing Environment: Multi-node cluster for distributed systems development

  • Data Storage: High-speed storage for large-scale atmospheric datasets

  • Networking: High-bandwidth connections for SuperDARN data ingestion

Team and Collaboration

Current Collaborators:

  • Lancaster University Physics: Domain experts for inventory management requirements

  • SuperDARN Community: International radar network researchers and engineers

  • Human-Computer Interaction Researchers: Collaborators on tactile optimization interfaces

Research Support:

  • Undergraduate Researchers: 3 students working on various project components

  • Graduate Collaboration: Joint work with students in adjacent research areas

  • International Partnerships: Remote collaboration with SuperDARN institutions globally

Timeline and Upcoming Milestones

Next 6 Months (July - December 2025)

  • August 2025: LUStores system deployment and initial user training

  • September 2025: SuperDARN Chord-DHT integration testing with live data feeds

  • October 2025: Tactile LSA preliminary results and first conference submission

  • November 2025: Cross-project optimization algorithm benchmarking study

  • December 2025: Year-end progress reviews and planning for 2026 extensions

Next 12 Months (2025-2026)

  • Q3 2025: LUStores operational assessment and optimization refinements

  • Q4 2025: SuperDARN Chord-DHT production deployment across multiple sites

  • Q1 2026: Tactile LSA methodology paper submission to optimization conference

  • Q2 2026: Integrated system demonstrations and community workshops

Long-term Outlook (2026-2027)

  • Project Completions: LUStores and Tactile LSA projects completion by mid-2026

  • SuperDARN Extension: Potential multi-year extension for operational support and enhancement

  • New Initiatives: Exploration of tactile methods for other optimization problems

  • Community Impact: Establishment of open-source communities around developed tools

For more information about any of these projects or collaboration opportunities, please see the contact page.