Human Stories in AI: Simon Stochholm

Overview

Simon Stochholm, a lecturer at UCL Denmark, details his unconventional journey into AI, starting with early fascination with ELIZA and computational linguistics. He transitioned from theoretical linguistics to practical application, learning to code under pressure for a speech recognition switchboard project and later applying deep learning techniques to diverse problems like ferry battery optimization, medical imaging for foot amputations, pig welfare monitoring, and satellite cloud movement prediction.

Key takeaways

Simon Stochholm transformed time series data from ferry operations into images using GANS and Markov transition fields to identify individual captains by their thrust patterns.
He is working on a project using instance segmentation on MRI scans to analyze blood flow and potentially prevent leg amputations.
A pig welfare project uses YOLOv8 to track and classify pig behaviors (drinking, eating, lying down) from ceiling-mounted cameras.
Efforts to generate synthetic pig images for training using Stable Diffusion encountered issues with realism and prompt interpretation, leading to unexpected outputs like barbecued pigs.
Predicting cloud movement from satellite images for solar energy forecasting is a computationally intensive project requiring substantial cloud infrastructure.
Simon's career advice emphasizes embracing opportunities, persistence through challenges (like transferring to five universities), and following one's passion.

Chapters

0:00 Early Fascination with ELIZA and Computational Linguistics

Simon Stochholm's interest in AI began at age 10 with the ELIZA chatbot.
He initially pursued Linguistics, later discovering Computational Linguistics.
An administrative mix-up led him to switch universities to study Computational Linguistics for his Master's.

5:02 Developing a Speech Recognition Switchboard

Simon was tasked with building a speech-activated switchboard within six months.
He learned to code on the fly, adapting Dragon speech libraries for Danish.
This project involved phonetic pronunciation mapping for word recognition.

12:07 Transition to Machine Learning and Academia

A financial crisis led to company-wide layoffs, prompting Simon to explore teaching.
He gained certifications and improved his coding skills while teaching individuals with Asperger's Syndrome.
Simon joined UCL nine years ago, initially teaching computer science before focusing on machine learning.

13:48 Applied Deep Learning and Fast.ai

He utilized DataCamp for learning and StatQuest videos for teaching applied deep learning.
The Fast.ai course was a turning point, making deep learning accessible.
Simon shifted from hands-on application to deeper mathematical and algorithmic understanding.

18:29 Project 1: Ferry Battery Optimization via Time Series Images

A ferry company needed to transition from diesel to electric, requiring optimized charging times.
Simon converted time series data (ship thrust over time) into images using GANS and Markov transition fields.
This allowed identification of individual captains based on their thrust patterns, enabling a shift schedule optimization.

35:17 Projects 2 & 3: Medical Imaging and Pig Welfare

Project 2: Identifying blood flow in MRI scans to prevent leg amputations using instance segmentation.
Project 3: Monitoring pig welfare (drinking, eating, posture) using cameras and YOLOv8 for behavior classification and tracking.
Synthetic data generation with Stable Diffusion was explored for pig behavior analysis, facing challenges with image realism and prompt interpretation.

Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, StatQuest with Josh Starmer.