Marlos C. Machado

@MarlosCMachado

Research Scientist at Brain Montréal. Former intern and . Research interests: AI, ML, RL. Opinions are my own.

Montréal, Québec
Vrijeme pridruživanja: travanj 2011.

Tweetovi

Blokirali ste korisnika/cu @MarlosCMachado

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @MarlosCMachado

  1. Prikvačeni tweet
    16. lip 2019.

    My Ph.D. thesis is (finally!) available at University of Alberta's website. In it, I discuss different ways one can use the process of learning (time-based) representations to guide option-based and count-based exploration in reinforcement learning.

    Poništi
  2. 23. sij

    Totally useless piece of information: I learned today you can change your Twitter handle. Therefore, I am now instead of @cholodovskis since no one knows my first last name in North America. Yeah, the ‘C’ stands for Cholodovskis ¯\_(ツ)_/¯

    Poništi
  3. proslijedio/la je Tweet
    15. sij

    When neuroscience and AI researchers get to chatting, cool stuff happens! My first, and I hope not last, trip into neuroscience has been published in Nature. 1/

    Prikaži ovu nit
    Poništi
  4. 14. sij

    This is a fantastic initiative that continues to spearhead. I'll be attending and I'm planning on attending this luncheon as well. Because I think this is such a great idea, below is the advertisement 😉

    Poništi
  5. 10. sij
    Poništi
  6. 3. sij

    Sometimes one might feel science is “stuck” and that there isn’t that much progress nowadays. I find it refreshing to see articles like this, looking back at what humanity has accomplished. It’s amazing how much was done in the last decade!

    Poništi
  7. 2. sij

    Em 2006, eu passei no vestibular para começar no 2o semestre. Meu projeto para minhas férias de 1 semestre foi colocar todos os mapas dos suplementos do Reinado num único mapa gigante, na escala certa!

    Poništi
  8. 2. sij

    Mas além das Dragão Brasil, tinha também os livros-jogo e as Holy Avengers. Sigo ansioso para a entrega do meu Tormenta 20! Pena que voltarei para o Canadá antes dele chegar. Busco da próxima vez 🤷🏻‍♂️

    Prikaži ovu nit
    Poništi
  9. 2. sij

    Essa é a época na qual visitamos nossos pais. Enquanto arrumava a minha antiga estante, olha as relíquias que resgatei! Aqui estão as fotos das Dragão Brasil que fizeram minha infância.

    Prikaži ovu nit
    Poništi
  10. 30. pro 2019.

    My 2010s: 👨🏻‍💻Worked as SWE; decided to become a researcher 💒Got married 🌎Lived in 3 countries 👨🏻‍🎓Finished BSc, MSc, PhD 🚗Had major car accident 👨🏻‍🔬Interned @ 3 awesome places 🐈Adopted 2 cats 📖Published in top tier venues 👼🏻Lost a baby 👶🏻Had a daughter 🧠Started RS job @ Brain

    Poništi
  11. 20. pro 2019.

    A visualization I really like is the one below. When we use the Laplacian to iteratively discover options, the agent learns options that take it to the boundary of its knowledge! We discover options that allow the agent to go deeper in the state space at each iteration.

    Prikaži ovu nit
    Poništi
  12. 20. pro 2019.

    One of the reasons I'm excited about this paper is that it continues the line of work related to eigenoptions. After showing we can use the Laplacian in a clever way to reduce the # of options one needs, we introduce a fully online method for option discovery in deep RL

    Prikaži ovu nit
    Poništi
  13. 20. pro 2019.

    We're pleased to let you know that your submission, Exploration in Reinforcement Learning with Deep Covering Options, has been accepted at ! This work was led by , with Jee Won Park and George Konidaris. More👇🏼

    Prikaži ovu nit
    Poništi
  14. 20. pro 2019.

    I'd say that a lot of our results suggest that we have been overfitting to a single domain for a while now. They also highlight the *importance of proper/fair comparisons* between algorithms. Recent methods, under a fair comparison, often don't outperform older methods!

    Prikaži ovu nit
    Poništi
  15. 20. pro 2019.

    In this paper we make an apples to apples comparison between popular bonus-based exploration methods. We show that while these methods lead to higher score on Montezuma's Revenge, most of them don't provide meaningful gains in the other Atari games.

    Prikaži ovu nit
    Poništi
  16. 20. pro 2019.

    We're pleased to let you know that your submission, On Bonus Based Exploration Methods In The Arcade Learning Environment, has been accepted at ! This huge endeavor was led by . W/ , & . More👇🏼

    Prikaži ovu nit
    Poništi
  17. 12. pro 2019.

    Starting today, I’m officially on vacation. Let the season of feeling-guilty-because-I’m-not-working start! Tweeter makes it even better as I’ll keep seeing everyone else tweeting about their work ¯\_(ツ)_/¯

    Poništi
  18. 21. stu 2019.

    I’m always happy of seeing AI papers at places like Science because it means AI is becoming mainstream science. I’m even happier when the authors are friends. But I got say, the coolest thing here, by far, is seeing a reference to Hari Seldon & Foundation!

    Poništi
  19. 16. stu 2019.

    After all this time, they are all mine. I’m sorry, Frieza, I got there first 🤷🏻‍♂️

    Poništi
  20. 10. stu 2019.

    We are pleased to inform you that your paper, "Count-Based Exploration with the Successor Representation", has been accepted for presentation at the (with and ). Link for the old version (a new one is coming soon):

    Poništi
  21. proslijedio/la je Tweet
    7. stu 2019.

    An update: has been in contact with Canadian immigration officials. They told him that anyone who has been denied a visa to attend can request their case to be reconsidered via this form:   No guarantees, but please pass along!

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·