I'm a Staff Research Scientist at Riot Games in Los Angeles, CA. Before that, I was a Postdoctoral Researcher at the Max Planck Institute for Intelligent Systems with Georg Martius. I completed my PhD in Machine Learning at Imperial College London advised by Petar Kormushev, MSc in Computer Science at the University of Southern California, and MEng in Electrical Engineering at UCL with a year abroad at Georgia Tech. I have also spent time at Microsoft Research hosted by Harm van Seijen and the University of Alberta hosted by Richard S. Sutton.

Publications

On the Pitfalls of Heteroscedastic Uncertainty Estimation with Probabilistic Neural Networks

M Seitzer, A Tavakoli, D Antic, G Martius

ICLR 2022

Arxiv OpenReview Code Talk
Orchestrated Value Mapping for Reinforcement Learning

M Fatemi, A Tavakoli

ICLR 2022

Arxiv OpenReview Code Talk
On Structural and Temporal Credit Assignment in Reinforcement Learning

A Tavakoli

PhD Thesis, Imperial College London (2021)

PDF
Learning to Represent Action Values as a Hypergraph on the Action Vertices

A Tavakoli, M Fatemi, P Kormushev

ICLR 2021

Arxiv OpenReview Code Talk
A Neural Network Oracle for Quantum Nonlocality Problems in Networks

T Kriváchy, Y Cai, D Cavalcanti, A Tavakoli, N Gisin, N Brunner

npj Quantum Information (2020)

Arxiv Nature Portfolio Code
Using a Logarithmic Mapping to Enable Lower Discount Factors in Reinforcement Learning

H van Seijen, M Fatemi, A Tavakoli

NeurIPS 2019 (oral, top 0.5%)

Arxiv NeurIPS MSR Blog Code Talk
Exploring Restart Distributions

A Tavakoli, V Levdik, R Islam, CM Smith, P Kormushev

RLDM 2019

Arxiv
Time Limits in Reinforcement Learning

F Pardo, A Tavakoli, V Levdik, P Kormushev

ICML 2018

Arxiv PMLR Website Talk
Action Branching Architectures for Deep Reinforcement Learning

A Tavakoli, F Pardo, P Kormushev

AAAI 2018 (oral, top 10.9%)

Arxiv AAAI Code Talk
Crowdsourced Coordination through Online Games

A Tavakoli, H Nalbandian, N Ayanian

HRI 2016

PDF IEEE ICRA Workshop
Seamless Robot Simulation Integration for Education: A Case Study

W Hönig, A Tavakoli, N Ayanian

Workshop on the Role of Simulation in Robot Programming, SIMPAR 2016

PDF Code

Publications

On the Pitfalls of Heteroscedastic Uncertainty Estimation with Probabilistic Neural Networks

Orchestrated Value Mapping for Reinforcement Learning

On Structural and Temporal Credit Assignment in Reinforcement Learning

Learning to Represent Action Values as a Hypergraph on the Action Vertices

A Neural Network Oracle for Quantum Nonlocality Problems in Networks

Using a Logarithmic Mapping to Enable Lower Discount Factors in Reinforcement Learning

Exploring Restart Distributions

Time Limits in Reinforcement Learning

Action Branching Architectures for Deep Reinforcement Learning

Crowdsourced Coordination through Online Games

Seamless Robot Simulation Integration for Education: A Case Study