Projects

Experiments in code. From reinforcement learning agents to consciousness simulations, each project explores a different aspect of intelligence and reasoning.

RL AgentsActive

2024-01-10

Multi-Agent RL Behavior Analysis

This project investigates how multiple RL agents develop sophisticated cooperative behaviors without explicit programming for cooperation.

Description

Analyzing emergent behaviors in multi-agent reinforcement learning environments. Focusing on communication protocols and cooperative strategies.

Technologies

PythonStable-Baselines3PyTorchGymnasium

Key Learnings

Discovered that agents develop implicit communication through action patterns even without explicit communication channels.

Code

Mind ModelingPlanning

2024-01-05

Consciousness Simulation Framework

An attempt to create a computational framework that exhibits properties associated with consciousness.

Description

Building computational models to simulate aspects of conscious experience, including attention, self-awareness, and qualia representation.

Technologies

PythonTensorFlowCognitive Architectures

Key Learnings

Still in research phase - exploring Integrated Information Theory and Global Workspace Theory as foundations.

BCI & EEGCompleted

2023-12-20

EEG Pattern Recognition for Mental States

Exploring the feasibility of real-time mental state detection using affordable EEG hardware.

Description

Machine learning model to classify mental states (focus, relaxation, stress) from EEG signals using consumer-grade hardware.

Technologies

Pythonscikit-learnMNEOpenBCI

Key Learnings

Consumer EEG can detect basic mental states with 70-80% accuracy. Meditation experience correlates with cleaner signal patterns.

Demo Code

OthersCompleted

2023-12-15

Logic Puzzle Generator

A systematic approach to generating and solving logic puzzles for cognitive training.

Description

Algorithmic generation of logic puzzles with varying difficulty levels. Includes solver and hint system.

Technologies

JavaScriptReactAlgorithm Design

Key Learnings

Puzzle difficulty is highly subjective. What's hard for humans isn't necessarily hard for algorithms.

Demo Code

OthersActive

2024-01-01

Transformer Attention Visualization

Making the 'black box' of transformer attention more interpretable through interactive visualization.

Description

Interactive visualization of attention patterns in transformer models to understand what they 'focus' on during reasoning tasks.

Technologies

PythonTransformersD3.jsFlask

Key Learnings

Attention patterns reveal surprising similarities to human cognitive focus, especially in logical reasoning tasks.

Demo Code