I am from Sri Lanka 🇱🇰, currently living in Oulu Finland 🇫🇮.
I am an engineer with over 7 years of experience developing automation and digitization solutions in the telecommunications industry.

I will soon receive a Master's Degree in Wireless Communications Engineering from the University of Oulu, where I have been exploring and studying applications of AI, RL and DL in Wireless Communications Networks.

When I am not building dreams or volunteering to teach kids, you'll find me running outdoors 🏃🏽 - rain, shine, or snow!, 📺 Binge-watching Netflix's finest, and spending quality time with my beautiful wife Jero 🩵.

Adaptive multi-agent systems by modeling other agents' communications and actions.

• Discrete State/Action Spaces

• Continuous State/Action Spaces This is a work in Progress - I am extending the discrete results to continuous state and action spaces using the MPE gym environment. On the left - Base Policy. On the right - Sequential Rollout.

• NVIDIA Sionna
  As a part of the coursework for 521322S Telecommunication Engineering Project, I am leveraging Sionna from NVIDIA Labs - the open-source library to simulate the physical layer of wireless systems. Specifically, I am attempting to implement OFDM signal with sync and explore channel and data estimation comparing results with implementations on MATLAB.

• Building RAG Agents with LLMs

Here is a summary of my project compilation.

• RL, DL and ML

  • Reinforcement Learning
    • World Models
    • Multi Robot Repair
    • CartPole
    • Cleanup and Harvest - PPO Agents in MARL
    • DM Walker - Vanilla Policy Gradient Agent for Base Policy
    • DM Walker - MPC on MPPI Agent with TD3
  • Deep Learning
  • Machine Learning

• Communication Networks

  • Signal Processing
  • Signal Detection and Estimation
  • L1 Simulations
  • Network Virtualization and Containerization
  • Antenna Design

• World Models
  

  Original implementation by ctallec
  My contribution is the compatibility with Apple Silicon processors.

  Pytorch implementation of the "World Models" for ARM processors (Apple Silicon). I reimplemented the Paper: Ha and Schmidhuber, World Models compatible with Apple Silicon processors. On the right, you can see the agent's regenerated dream of the real observation (left) by using the VAE.
  

• Multi Robot Repair
  
  This is a work in progress. I am developing the simulator described in the paper - Multiagent Rollout and Policy Iteration for POMDP with Application to Multi-Robot Repair Problems.
  
  

• CartPole
  

  • DQN Agent This is an attempt to understand how a DQN agent can be trained for a cartpole game.
  • Vanilla Actor-Critic and PPO This is my attempt to understand how to train a vanilla actor-critic and PPO agent for a cartpole game.
    
    
  

• Cleanup and Harvest - PPO Agents in MARL
  
  This is a self-study exploration of how to train PPO agents in a multi-agent setting. On the left you can see the PPO agents at work and on the right side compares to random actions.
  
  

  

• DM Walker - Vanilla Policy Gradient Agent for Base Policy
  
  This is a self-study of how to train a vanilla policy gradient agent that can be used as a base policy for running MPC later on.

• DM Walker - MPC on MPPI Agent with TD3
  
  This is a self-study of how to learn physics dynamics and a policy for walker-stand using TD3 and MPC based on MPPI.
  Left - Vanilla PG Agent | Right - MPC Agent

  

• Multi-Modal Physical Exercise Classification Decision-level fusion for multimodal classification. Feature extraction. Feature-level fusion.

• Transfer Learning A group project on transfer learning-based image classification for the course Deep Learning 2023. Collaborators :

  • Suranga Wengappuli Arachchige : suranga.wengappuliarachchige@student.oulu.fi
  • Madhusanka, Manimel Wadu : madhusanka.manimelwadu@oulu.fi

• Linear Regression Load data and create a train/test split. Build a Pytorch model for a simple linear regression problem. Training the model with gradient descent algorithm in Pytorch.

• Fashion-MNIST classification Neural Network, Deep Neural Network, Loss Function and Optimization. Building a simple NN using numpy to understand the backpropagation. Gradient check using finite-difference approximation. Stochastic Gradient Descent (SGD). Regularization and simple hype-parameters tuning methods to improve NN performance.

• CNN

• Generative Adversarial Network

• Neural Network from scratch An interesting approach to explaining neurons, layers, and how weights and biases work using basic principles. YOUTUBE

• Optimal Wiener Filter
• Kalman Filter
• Extended Kalman Filter
• LMS Algorithm for Channel Equalizations
• RLS Algorithm
• Estimator Correlator

• Efficient Signal Estimator by Monte Carlo Simulation
• Maximum likelihood estimator. Linear Estimator
• MLE and LSE Estimators
• LMMSE Estimator. Signal Detection
• Linear algebra, Probabilities, Multivariate densities, and Matrices

• Simulation of AWGN Channel
• BER Simulations on AWGN with BPSK
• Gray and Binary Coding with QPSK on AWGN
• Adaptive Gain Controller for LTE20 Signal
• Ray Tracing
• MPSK in AWGN Channel
• DPSK in Rayleigh Fading Channel
• Maximal Ratio Combining for Diversity
• Channel Coding
• Convolutional Encoder and Decoder

• CoAP | IoT | Hypervisor
• Mininet

• Single Element Antenna and Antenna Array Design