Mike.W -

Latest Work

Here is a list of some of my lastest works or projects, enjoy.

Real or Realistic? What is more important? Let's imagine me

Are these pictures real? Is it ok to use AI for your next passport picture? for your linkedin? In a world where AI blurs the line between real and artificial, can you distinguish between genuine and AI-generated images? Does it even matter?

GenAI Diffusion stablediffusion Flex Midjourney Photos

JPEG + Fourier Transform + Efficient CNN?

An intriguing idea about using JPEG compression to make neural networks learn faster.

JPEG CNN Fourier

What is it missing from LLMs to automate complex tasks?

I am exploring all the limitations and challenges left to solve in the area of LLMs in order to develop a fully automated smart assistant

LLM GPT Automation SmartAssistant TaskManagement

Autonomous Parking Takes Over FORD Factory!

This is the latest demo to date of our product in a fully operation vehicle manufactoring plant

Ford AVP AVP-type2 Kopernikus KopernikusAuto

About transformers

A quick exercise of coding a classic transformer from zero to get a full understanding of the logic behind it

Transformers Self-attention Math

International Motor Show (IAA 2021)

Some interesting videos of our product in IAA (The biggest mobility show in the world)

IAA IAA-2021 International-Motor-Show Internationale-Automobil-Ausstellung

Update: ML Lead

Some reflexions on my time as ML Lead at Kopernikus Automotive. I did not update this page for a while, because being involved in all internal projects demands time.

Lead Machine-Learning Team

A quick review of Differentiable Renderers

A quick review of Differentiable Renderers, more exactly a dive deep into SoftRas and a line-to-line understanding of the model

Maths Differentiable-renderers SoftRas

Shapes3D - Blender-python minimalist API

Open sourced library to simplify the work flow with blender for Computer Vision problems

Computer-Vision Rendering

Error State - Extended Kalman Filter

A "quick" review of how to derive the Error State - Extended Kalman Filter

Maths Sensor-Fusion Kalman-Filter

Unsupervised Depth Estimation Explained

Explaining how unsupervised depth estimation works and a re-implementation of the original paper

Maths Deep-learning Computer-vision

Hierarchical Reinforcement Learning for Robustness, Performance and Explainability

Using Hierarchical Policies in order to improve the general performance of RL models

Hierarchical-RL Python DQN

Visual Perception for Autonomous Racing Car

Edinburgh University Formula Student - Perception Team

Perception ROS Object-detection

Visual Perception for Autonomous Racing Car

RL applied to Self-Driving Race Cars

Using TRPO in order to solve EUFS problem (Driving around a track)

RL TRPO ROS

Solving CarRacing with PPO

An easy, robust and fast solution for the CarRacing-v0 environment

RL OpenAI CarRacing-v0 PPO

CarRacing-v1

An significant improvement over CarRacing-v0

Gym OpenAI Python

Autonomous Guided Vehicle

Autonomous Guided Vehicle in an Experimental Autonomous Warehouse

Robotics ROS perception

Cross-Convolutional Neural Networks applied to Finance

Cross-Convolutional Neural Networks applied to Finance (working on ...)

Movement-prediction DNN

Agent-based Models applied to transportation

Agent-based Models applied to transportation

Trueque Gamer

This is a project that a friend and I came up with when I found out how tedious it is to exchange games through Facebook groups

Maths Rubik's-cube

SomosUN

Platform to help all students to build and manage their curriculum during the undergraduate degree.

Maths Rubik's-cube