Bio

Luckeciano is a DPhil student in the OATML group, supervised by Yarin Gal and Alessandro Abate. His research interests lie in designing agents that learn behaviors through interactions in an efficient, generalist, safe, and adaptive way. He believes that such agents emerge from three main pillars: semantically rich representations of entities in the world; self-supervised World Models with inductive biases for memory and counterfactual reasoning; and fast policy adaptation mechanisms for out-of-distribution generalization.

Previously, he worked as an Applied Scientist at Microsoft, working with multi-modal representation learning and large language models for web data semantic understanding. He also led the RL Research group at the Center of Excellence in AI in Brazil, working with real-world RL applications in scalable digital platforms with industry partners.

Interests

  • Reinforcement Learning (Meta-RL, Offline RL, Model-Based RL)
  • Representation Learning

Education

  • PhD in Computer Science, 2026

    University of Oxford

  • MSc in Artificial Intelligence, 2019

    Aeronautics Institute of Technology

  • BSc in Computer Engineering, 2018

    Aeronautics Institute of Technology

Experience

Major pieces of my research/industry professional experience. For full list, check my CV.

 
 
 
 
 

Applied Scientist

Microsoft

Oct 2021 – Sep 2023 Seattle, US

Works with multi-modal representation learning for web data semantic understanding.

Responsibilities include:

  • Tech Leader for the Semantic Document project in the Bing Document Understanding team.
  • Leading efforts to scale HTML-based deep learning models for 400 bilion documents
  • Developed an LLM-based pipeline for Semantic Document label extraction
  • Data Engineering, Data Analysis, Feature Engineering, Model development, Deployment, Monitoring
 
 
 
 
 

Software Engineer

Microsoft

Feb 2020 – Oct 2021 Vancouver, Canada

Works building up microservices that enable data transfer/processing from SQL databases to Azure Data Lake at scale in Dynamics 365 Finance and Operations. Responsibilities include:

  • Delivering high quality, scalable code for asynchronous, distributed, and multi-threaded applications in the context of SaaS in the cloud
  • Architectural Discussions
  • Code Reviews, DevOps/Livesite.
 
 
 
 
 

Head of RL Research

Center of Excellence in AI (Deep Learning Brazil)

Jun 2019 – Present Remote

Responsibilities include:

  • RL core research (Offline RL, Meta-RL)
  • RL applications with industry partners
  • Manage/Supervise BSc/MSc students
 
 
 
 
 

Software Development Engineer Intern

Amazon Web Services

May 2018 – Aug 2018 Cape Town, South Africa
Worked at EC2 Core Platform, in Host Placement team. Developed a Continuous Deployment Pipeline for instances metering service based on several testing mechanisms that uses metering data.
 
 
 
 
 

Software Engineer

VTEX IO E-commerce cloud platform

Feb 2018 – May 2018 Rio de Janeiro, Brazil
Developed Logs and Monitoring features for Go/.NET web services. Developed HTTP cache and throttling mechanisms. Developed AB Testing support in VTEX IO infrastructure.
 
 
 
 
 

Research Intern

Software Engineering Research Group - ITA

Mar 2017 – Nov 2017 Sao Jose dos Campos, Brazil
Worked using Deep Learning to Facial Recognition for Biometric systems, replacing from Eingenfaces’ solution to FaceNet (improved Identification Rate from 40% to 90%); also worked in feature engineering for credit cards anti-fraud systems.

News

  • Oct 2023: Pleased to announce that I am starting my PhD in Computer Science at University of Oxford, supervised by Yarin Gal and Alessandro Abate!
  • May 2022: Transformers are Meta-Reinforcement Learners was accepted as an spotlight talk in the International Conference on Machine Learning!
  • October 2021: PulseRL: Enabling Offline Reinforcement Learning for Digital Marketing Systems via Conservative Q-Learning was accepted as an spotlight talk in the Offline Reinforcement Learning Workshop at Neural Information Processing Systems!
  • October 2020: MARS-Gym: Offline Reinforcement Learning for Recommender Systems in Marketplaces was accepted as an spotlight talk in the Challenges of Real World Reinforcement Learning Workshop at Neural Information Processing Systems!
  • September 2020: We presented Contextual Meta-Bandit for Recommender Systems Selection in the ACM Conference on Recommender Systems 2020!
  • February 2020: Joined Microsoft as Software Engineer!
  • October 2019: Pleased to announce that my MSc dissertation Imitation Learning and Meta-Learning for Optimizing Humanoid Robot Motions was awarded as the best Brazilian MSc work in AI in the AI Awards 2019!
  • October 2019: Pleased to announce that my MSc dissertation Imitation Learning and Meta-Learning for Optimizing Humanoid Robot Motions was awarded as the best Brazilian MSc work in Robotics in the V Best MSc Dissertation and PhD Thesis Contest in Robotics!
  • October 2019: Pleased to announce that our paper Learning Humanoid Running Skills Through Proximal Policy Optimization was awarded as the best paper in the Latin America Robotics Symposium!

Awards

Main recent awards. For full list (which comprises code/research competitions, hackathons and high school scientific competitions), check my CV.

AI Awards

DATA-H Dec 2019
Winner of Best MSc dissertation with ‘Imitation Learning and Meta-Learning for Optimizing Humanoid Robot Motions

V Best MSc Dissertation and PhD Thesis Contest in Robotics

Brazilian Robotics Society Nov 2019
Winner of Best MSc dissertation with ‘Imitation Learning and Meta-Learning for Optimizing Humanoid Robot Motions’

Khipu Attendee

Khipu Nov 2019
Accepted to attend to Khipu: Latin American Meeting for AI 2019

Best Computer Engineering Thesis

Aeronautics Institute of Technology Dec 2018
Winner of Best Undergraduate Thesis (Class of 2018) with ‘A Deep Reinforcement Learning Method for Humanoid Kick’

Intel AI Student Ambassador

Intel Oct 2018 – Feb 2020
As part of the Intel® AI Developer Program, get access to newly optimized frameworks and technologies, hands-on training, and technical resources for graduate and PhD students from top universities worldwide to further innovation in AI.

Recent Publications

For up-to-date, full list of publications, check my google scholar.

Quickly discover relevant content by filtering publications.
Luckeciano C Melo* (2022). Transformers are Meta-Reinforcement Learners. 39th International Conference on Machine Learning (ICML).

PDF

Luckeciano C Melo*, Luana Martins*, Bryan Oliveira*, Bruno Brandao, Douglas Soares, Telma Lima (2021). PulseRL: Enabling Offline Reinforcement Learning for Digital Marketing Systems via Conservative Q-Learning. Offline Reinforcement Learning Workshop at the 35th Conference on Neural Information Processing Systems (Spotlight talk).

PDF

Luckeciano C Melo*, Marlesson Santana*, Fernando Camargo*, Bruno Brandao, Renan M Oliveira, Sandor Caetano, Anderson Soares (2020). MARS-Gym: Offline Reinforcement Learning for Recommender Systems in Marketplaces. Challenges of Real World Reinforcement Learning at Neural Information Processing Systems 2020 (Spotlight talk).

PDF

Luckeciano C Melo*, Marlesson Santana*, Fernando Camargo*, Bruno Brandao, Renan M Oliveira, Sandor Caetano, Anderson Soares (2020). Contextual Meta-Bandit for Recommender Systems Selection. ACM Conference on Recommender Systems 2020.

PDF

Luckeciano C Melo*, Marcos R O A Maximo, Adilson M Cunha (2019). Bottom-Up Meta-Policy Search. Workshop on Deep Reinforcement Learning at the 33rd Conference on Neural Information Processing Systems (NeurIPS 2019).

PDF

See all publications