1ª: Recomendando cursos através de redes neurais recorrentes
Com aproximadamente 700 cursos um dos desafios na Alura é guiar cada um de nossos alunos e alunas com o curso ideal para seu momento de vida. Recomendadores tradicionais foram testados e fracassaram miseravelmente.
Usando algoritmos de NLP e redes neurais recorrentes indica gerar uma solução capaz de filtrar o ruído existente nos dados e vencer até mesmo recomendações humanas. Geramos assim recomendações de cursos que se encaixam com o contexto atual de cada aluno e aluna.
Tais algoritmos tem sido a base de recomendadores usados pelo Netflix.
Guilherme co-fundou a Caelum e a Alura. Com mais de 15 anos de experiência no ensino, coordena as equipes de produção de cursos na Alura. É tecnólogo com viés matemático e representou o Brasil em dois mundiais de programação.
2ª: Deep Q-Learning
Today more and more research is targeted at building intelligent agents that are able to solve increasingly difficult tasks from playing Atari games over simple robot control to drones navigating autonomously through the woods.
The focus of this talk is on one of the most important algorithms developed in recent years that gave rise to a whole new research field, Deep Reinforcement Learning, by combining the advantages of Reinforcement Learning with Deep Learning. After a short introduction to Reinforcement Learning, you will see how the classic version of Q-Learning works in a short example and then look in detail at the Deep Q-Learning algorithm to discuss the main characteristics of it before introducing the most important extensions of that algorithm. You will also see many of the successful applications developed based on Deep Reinforcement Learning.
Ruben has a mechatronics engineering background but turned to artificial intelligence where his main interest lies in machine learning (ML) research, particularly in Deep Reinforcement Learning (DRL) in combination with knowledge transfer for cognitive autonomous decision-making agents.
Currently, Ruben is a Ph.D. student at the Escola Politécnica of the University of São Paulo (USP), Brazil. He holds a Brazilian master degree in Mechanical Engineering in the area of controlling mechanical systems from the Universidade Estadual Paulista Júlio de Mesquita Filho, UNESP, and a German Diplom-Ingenieur degree in Mechatronics in the area of sensors and robotics from the Karlsruhe Institut of Technology, KIT.
Apart from his academic experiences, Ruben has acquired years of professional experience before and during his studies while working in the technology sector and has recently become the Lead Organizer at PAPIs.io to host ML conferences all over the world.
Ruben's long-term research intention is to create intelligent agents utilizing scalable Reinforcement Learning architectures for real-world applications.