O texto com o título “Como os robôs superaram os humanos no xadrez?” foi publicado hoje no blog da Folha de São Paulo. O texto tem como assinatura a doutoranda do grupo, Roberta Duarte, que trabalha desenvolvendo métodos de inteligência artificial, mais especificamente o uso de deep learning, para compreender a física de buracos negros.
No texto, Roberta escreveu sobre como a inteligência artificial vem crescendo ao longo dos anos, desde quando Alan Turing apresentou seu paper Computing Machinery and Intelligence em 1950. E como, desde Turing, a inteligência artificial tem sido usada para aprender a jogar xadrez e competir com humanos. Ela descreve sobre como ocorreu a evolução dessas máquinas até os dias de hoje, na qual a campeã mundial é a AlphaZero desenvolvida pela empresa DeepMind.
Inteligência artificial está começando a ser usada para o desenvolvimento da Astrofísica como técnicas e métodos que contribuem para análise de dados, simulações e classificações de objetos astronômicos.
No grupo, recentemente, submetemos dois artigos que utilizam do método deep learning para fitar SEDs de AGNs com baixa luminosidade e obter a primeira simulação de buraco negro por uma inteligência artificial. Os artigos estão disponíveis no arXiv.
From November 10th to November 15th, Khipu happened in Montevideo, Uruguay. Khipu was the best biggest event of artificial intelligence and machine learning in Latin America. The speakers of this event are all AI leaders in their fields and big companies such as Google, Deepmind, Apple and Facebook were sponsoring the event. The competition to be in one of the 200 participants of this event was highly competitive. It was one in a lifetime experience that I will try to sum up here.
On the very first day, they welcomed us with a small lecture about motivation in AI fields and with a toast event. In this toast event, I met Nando de Freitas (Principal Scientist at Deepmind and of the main speakers there) and Cho Kyunghyun (Research at Facebook AI and NYU Professor), we talked about my project, about the event and about Brazil! They were really nice and friendly to me and my friend.
All of the lectures were interesting but I will give special attention to some of them: Ian Goodfellow talking about GANs, Yoshua Bengio talking about Perspectives in AI, Nando de Freitas talking about Reinforcement Learning, Cho Kyunghyun talking about Recurrent Neural Networks and Jeff Dean closing the event with an amazing talk about the challenges in AI. All the lectures are available here: http://tv.vera.com.uy/buscar?search=Khipu
Another important lecture was “How To Write a Good Paper”. So important and very helpful ideas!
Another highlight in this event was Nando de Freitas talking about my project with me when he visited my poster and commenting about my project in his lecture. It was a really special moment for sure. He even posted in Twitter and retweet my tweet talking about the project. I received a supportive comment from Cho Kyunghyun as well.
It is impossible to talk about Khipu without talking about Jeff Dean (GoogleAI leader and one of Tensorflow creators). I talk with him for a few minutes in a coffee break, he is also very friendly and we talked about my project and about GoogleAI opportunities. In his awesome lecture about challenges in AI, he also comments about my project and I found that one of the biggest highlights in the week. I took a picture with him and posted on Twitter, he commented.
I even met some amazing people from DeepMind and GoogleAI there, they gave me a lot of advice and talked about the life in DeepMind and in GoogleAI. I even met Feryal Behbahani from DeepMind and Women in ML and Oriol Vinyals from DeepMind (main AlphaStar researcher)
This event was so important with so many big names in AI. I am very thankful for Khipu because this event because I have so many opportunities now and also big thank you because they paid all the expenses with travel and accommodation. I really look forward to Khipu 2020! It is a really life changing experience.
Black Hole Group 