About Experience Projects Interests Contact Blog

Fraser Mince

Full Stack Engineer - React, Rails, and Elixir

ML Researcher - RL, and NLP

About

As a Senior Fullstack Engineer and ambitious Machine Learning Researcher, I am passionate about personal growth and expanding my knowledge.

In recent years, I have focused on mastering the field of Deep Reinforcement Learning and Natural Language Processing. My current endeavors include engaging in cutting-edge research projects and actively seeking out new career opportunities.

In my free time, I am deeply interested in the fields of politics, economics, and urbanism. I enjoy running and staying active. Reading fiction and non-fiction literature is one of my favorite pastimes.

Projects

Explore a curated selection of personal projects that I have undertaken throughout my career, each serving as an opportunity for growth and skill development.

From startup ideas event to my unfinished endeavors, these projects reflect my dedication to continuous learning and self-improvement.

Illustration of a stylized game of pong combined with go

Muzero Implementation - 2022

A recent project where I successfully implemented the Muzero algorithm using Jax and Ray for distributed training on Atari games. The project required a deep understanding of the underlying theory and the development of skills in neural network building, distributed training orchestration, and the use of Jax and Ray.

Acquired proficiency in using Jax for building neural networks that are easy to use and device portable
Developed expertise in using Ray for distributed training
Learned how to orchestrate distributed training effectively
Gained a thorough understanding of the AlphaZero family of algorithms and its underlying theory.

Illustration of a man inventing AI to be used in games

Reinforcement Learning Paper Implementations - 2021

A compilation of RL algorithms implemented in Pytorch, representing an initial step in my journey towards proficiency in reading and implementing RL algorithms.

Gained familiarity and comfort with Pytorch
Developed the ability to understand the fundamentals of RL papers and the mathematical concepts behind them
Acquired a foundational understanding of value networks, policy networks, and actor-critic networks
Implemented DQN, DDQN, A2C, and initiated an implementation of Muzero

Illustration of a book that is combined with a computer

Unchart - 2020

A startup idea I implemented to facilitate reading books in other languages, featuring an Elixir GraphQL backend and Purescript React Native frontend.

Enhanced my expertise in Purescript during a period of focus on functional programming
Acquired proficiency in GraphQL and Apollo
Gained experience in reaching out to potential users through cold emailing and messaging to gather feedback on the product
Became proficient in React Native
Learned how to use spaCy for NLP tasks related to processing epub files

Fast.ai course and beginning interest in machine learning - 2018

A period of learning and exploration in the field of AI and deep learning, including participation in the fast.ai course and Kaggle competitions, as well as delving into the underlying theory of deep learning and AI.

Acquired proficiency in using Pytorch
Developed expertise in using fast.ai
Gained a foundational understanding of neural networks, and distilled my learning in a blog post: https://thoughtbot.com/blog/neural-nets
Became familiar with convolutional neural networks and attention mechanisms
Began studying the "Spinning Up In Deep RL" guide: https://spinningup.openai.com/

Illustration of a computer higher than trees showing off code

Interactive Code Portfolio - 2017

A project developed with Purescript that aimed to create an interactive code-first portfolio.

Developed proficiency in Purescript
Gained experience in creating a product to meets my specific needs

Make a Lisp - 2016

A project to implement a Lisp dialect using Haskell

Advance my skills in Haskell
Became familiar with the Parsec library
Gained deeper understanding of the homoiconic nature of Lisp

Current Interests and Readings

Explore my current interests and thoughts.

This includes both my professional research pursuits and personal interests, including books and various subjects

Machine Learning

For AI Community Research Project

A research project I am working on as part of the cohere.for.ai community advised by Sara Hooker of For AI and Neil Thompson of MIT. More details to come.

RLHF

As reinforcement learning has found more production use cases through RLHF, I have begun to explore the state of the art.

Currently Reading

Learning to Summarize From Human Feedback - https://openai.com/blog/learning-to-summarize-with-human-feedback/
Training a helpful and harmless assistant with reinforcement learning from human feedback - https://arxiv.org/abs/2204.05862
Constitutional AI - https://www.anthropic.com/constitutional.pdf
Deepmind Sparrow - https://storage.googleapis.com/deepmind-media/DeepMind.com/Authors-Notes/sparrow/sparrow-final.pdf

Muesli

A new algorithm from deepmind. I am currently reading the theory to fully understand the inner workings

Currently Reading

Muesli: Combining Improvements in Policy Optimization - https://arxiv.org/abs/2104.06159
Reinforcement Learning and Control as Probabilistic Inference: Tutorial and Review - https://arxiv.org/abs/1805.00909
Monte-Carlo tree search as regularized policy optimization - https://arxiv.org/pdf/2007.12509.pdf
Maximum A Posteriori Policy Optimisation - https://arxiv.org/pdf/1806.06920.pdf