Twitter | Search | |
hardmaru
research scientist at google brain 🧠 香港人 🏴
20,454
Tweets
890
Following
93,344
Followers
Tweets
hardmaru 2h
Replying to @hardmaru
Bloomberg article on Imma As Covid-19 leads to the cancellation of product launches and sponsored travel, computer generated “virtual influencers” and models are becoming a fast growing and increasingly profitable business.
Reply Retweet Like
hardmaru 21h
Replying to @hardmaru
NaN's have real consequences
Reply Retweet Like
hardmaru Oct 29
Replying to @tkasasagi
Credit: using pre-trained PULSE () with KaoKore dataset ()
Reply Retweet Like
hardmaru Oct 29
Machine learning bias as an art form
Reply Retweet Like
hardmaru Oct 29
Someone forgot to load the model.
Reply Retweet Like
hardmaru Oct 29
The balancing wheel is kind of like a Segway. The author called it an inverted pendulum robot:
Reply Retweet Like
hardmaru Oct 29
This inverted pendulum robot by is quite a magnificent project.
Reply Retweet Like
hardmaru retweeted
Emtiyaz Khan Oct 29
We will be starting to interview soon for this post-doc / RA position for BayesML. We have updated the call so that people with Masters degree can also apply for the RA (officially called tech staff) position. Updated link: Help me spread the word!
Reply Retweet Like
hardmaru Oct 29
This is what the future of the Internet may look like.
Reply Retweet Like
hardmaru Oct 27
Replying to @hardmaru
A recent work looks at this in an offline RL setting, by using an unsupervised learning objective (simple variational autoencoder) on undirected experience data to learn primitive skills that inform what kind of behaviors are possible in the agent’s world.
Reply Retweet Like
hardmaru Oct 27
Replying to @roydanroy @NeurIPSConf
Your post-retirement career?
Reply Retweet Like
hardmaru Oct 27
Replying to @GitHubPolicy
Statement from on the matter:
Reply Retweet Like
hardmaru Oct 27
Replying to @natfriedman @github
Looks like stepping in to try to bring youtube-dl () back up:
Reply Retweet Like
hardmaru Oct 27
If people still read a paper of yours 25 years later, the last thing on their minds is where you had presented or published the paper 25 years ago. You‘re not the first to have a paper rejected and you won’t be the last :)
Reply Retweet Like
hardmaru retweeted
Yujin Oct 26
pfrl from is a great library for RL training that is based on PyTorch. I used it to reproduce the results from OpenAI’s Procgen paper:
Reply Retweet Like
hardmaru retweeted
/MachineLearning Oct 27
What do you think of JAX? Is it solid enough to start building long term research products on?
Reply Retweet Like
hardmaru Oct 26
Replying to @mooopan
What about tasks outside of MuJoCo? I had a go using SAC for SlimeVolleyGym but had disappointing results (even with continuous inputs/outputs)
Reply Retweet Like
hardmaru Oct 26
“Constraints can spur creativity and incite action, as long as you have the confidence to embrace them.” Amazing portfolio of 1000 programs that fit in a tweet of code and runs on an 8-bit virtual computer:
Reply Retweet Like
hardmaru Oct 26
An opportunity to become an Ethics Research Scientist Also, check out 's interview with (before applying) The true dangers of AI are closer than we think. Forget Superintelligent AI: Algorithms are already creating real harm.
Reply Retweet Like
hardmaru retweeted
@artnome Oct 26
New Artnome post. Nigerian soldiers are killing peaceful unarmed protestors for demanding an end to police brutality. Help artists raise the visibility of these human rights violations by getting the word out.
Reply Retweet Like