Samuel Oliveira

Samuel Oliveira

Ph.D. Student in Reinforcement Learning

University of Alberta

Amii

Biography

Hi! I’m Samuel. I am a Ph.D. student at the RLAI Lab, supervised by Rupam Mahmood and affiliated with Amii. Previously, I graduated with a MSc in Machine Learning at University College London (UCL), and completed my undergraduate degree at Imperial College London. I am interested in creating decision-making systems that can learn continually. In particular, I am curious about studying these under a RL framework, as well as through improvements in current neural network architectures.

Interests
  • Reinforcement Learning
  • Continual Learning
  • Robot Learning
Education
  • Ph.D. in Computer Science, 2025-2030

    University of Alberta

  • MSc in Machine Learning, 2023 - 2024

    University College London

  • MEng in Computational Bioengineering, 2019 - 2023

    Imperial College London

Experience

 
 
 
 
 
UCL
MSc Thesis Student
UCL
May 2024 – May 2025 London, UK

Investigated the use of diffusion models for Inverse Reinforcement Learning, in Ilija Bogunovic’s Group.

  • Created a model to learn a reward/guide function from a classifier-guided trajectory-level diffusion model.
  • Benchmarked the framework against state of the art IRL methods in multiple Gymnasium environments.
  • Thesis available here. Code available here.
 
 
 
 
 
Imperial College London
Research Assistant in Statistical ML
Imperial College London
October 2022 – September 2023 London, UK

Undergraduate Thesis on creating a computationally efficient algorithm to predict eczema severity. Worked on it as part of my Thesis (October 2022 - June 2023), and as an Intern Research Assistant (June 2023 - September 2023). Work done at the Tanaka Group, and in collaboration with Dr Rob Moss of the University of Melbourne.

  • Implemented a framework using Sequential Monte Carlo methods to predict symptom severity in eczema patients.
  • The framework is capable of outputting predictions for each patient in under a second.
  • Taught a 2-hour tutorial on SMC methods for the entire lab group.
  • Currently writing it as a paper “EczemaPF: a computationally efficient algorithm for predicting the evolution of eczema severity” (to be published in Q1 2025).
 
 
 
 
 
Goldman Sachs
Software Engineering Intern
Goldman Sachs
June 2022 – August 2022 Birmingham, UK

Worked for 10 weeks as a Software Engineer in the Technology division of Goldman Sachs.

  • Built from scratch a pen testing framework for a new web app hosted on AWS.
  • Presented my work to 100+ colleagues.
  • Received a returning offer.
  • Tech Stack: Python, AWS, Terraform.

Achievements

Imperial College London
Dean’s List 4th Year
Top 10% of cohort during the 4th year of my undergraduate degree.
ETS
GRE

Score: 338/340.

  • Verbal Reasoning: 169/170 (99th percentile).
  • Quantitative Reasoning: 169/170 (93rd percentile).
  • Analytical Writing: 5.0/6.0 (91st percentile).
Imperial College London
Dean’s List 3rd Year
Top 10% of cohort during the 3rd year of my undergraduate degree.

Projects

*
Inverse Reinforcement Learning using Generative Planning Models in Trajectory Space
MSc Thesis in Ilija Bogunovic’s Group, as part of the MSc in Machine Learning at UCL.
Inverse Reinforcement Learning using Generative Planning Models in Trajectory Space
Impact of the pre-training data distribution on the fine-tuned performance of Masked Autoencoders
Group Project as part of the Applied Deep Learning (COMP0197) module at UCL.
Impact of the pre-training data distribution on the fine-tuned performance of Masked Autoencoders
Multi-Task Multi-Agent RL using shared distilled policies
Adapting DeepMind’s Distral to a Multi-Agent setting, as part of the Multi-Agent AI (COMP0124) module at UCL.
Multi-Task Multi-Agent RL using shared distilled policies
Maximum Likelihood Estimation in Gaussian Process Regression is Ill-Posed
Paper Review as part of the Machine Learning Seminar (COMP0179) module at UCL, taught by Marc Deisenroth and Brooks Paige.
Maximum Likelihood Estimation in Gaussian Process Regression is Ill-Posed
Gatsby Unit PhD Modules Coursework
Coursework as part of the PhD-level modules at the Gatsby Unit.
Gatsby Unit PhD Modules Coursework
GPU Ray Tracing
Ray Tracing task as part of the Computer Graphics module at Imperial College London. Received an honorable mention for “scene composition” for Part 2.
GPU Ray Tracing

Gallery

Some of the photos I have taken with my film camera.