JJ retweeted
Simulations are important for scientific discovery, but they can be high-dimensional and difficult to understand.
We introduce Targeted Causal Reduction (TCR): a method to learn high-level causes of phenomena occurring in scientific simulations. 🧵1/
📄: arxiv.org/abs/2311.18639
JJ retweeted
In the midst of a globally consequential Iranian uprising, some Iranians are using the occasion to shame thoughtful Iranians for impurity of thought. These harassments are disgraceful, and their timing is most unfortunate. (1/3)
#IranProtests #IranRevoIution2022 #MahsaAmini
JJ retweeted
@DistributedDave @kkevsterrr Any ideas of Geneva strategies that may help in Iran?
JJ retweeted
Run packet manipulation to confuse the censor.
- Try github.com/Kkevsterrr/geneva and share working strategies
- or github.com/NullHypothesis/br…
JJ retweeted
This is a thread how to make Outline VPN Servers more resilient to blocking in #Iran:
- Use a domain name and rotate IPs
- Use a 2-hop system
- Do packet manipulation to confuse the censor
Follow along
#IranProtests2022
JJ retweeted
This is how to set up a 2-hop Outline VPN:
github.com/net4people/bbs/is…
This is how to set up a DNS tunnel:
bamsoftware.com/software/dns…
JJ retweeted
This needs validation, but my hope is that we can do 2-hop VPNs with the ASes not blocked and DNS tunnel on the DNS resolvers with external access
JJ retweeted
I'm trying to find a way out of the #InternetShutdown in #Iran
Here is a list of ASes that unaffected by the daily curfew:
github.com/net4people/bbs/is…
And a list of DNS resolvers that could be used for DNS tunneling:
github.com/net4people/bbs/is…
#IranProtests
JJ retweeted
Wo-wo-wo-wow!
More experiments with semantic guidance of a disentangled generative adversarial network for a paper in preparation. Here we start from a GAN-inverted picture of Bertrand Russell, and modify it in various ways using text prompts and a #CLIP-based loss.
JJ retweeted
Can RL algorithms be replaced with transformer-based language models? We’ve looked at this question with our work on Decision Transformer:
Website: sites.google.com/corp/berkel…
Code: github.com/kzl/decision-tran…
1/8
JJ retweeted
Attention-driven Robot Manipulation (ARM)🦾 is our new learning algorithm that can do RLBench tasks, while other baselines fail. Secret sauce is Q-attention🔎, which crops around interesting pixels before giving to an actor-critic method. w/ @AjdDavison
sites.google.com/view/q-atte…
JJ retweeted
I'm not easily impressed & I'm impressed
Results are stunning. Achieves top scores across all major metrics, plus it renders more than 1000× faster than the state of art [NeRF]
BTW: It isn't really neural anymore. It's all about the basis functions.
nex-mpi.github.io/
JJ retweeted
Had a great time teaching "Computational Control Theory" this semester, thanks to my TAs & students! Lecture notes have now reached "spectral filtering", I will add non-stochastic control to notes & course slides. Partial notes + more at:
sites.google.com/view/cos59x…
JJ retweeted
At a minimum, @AnimaAnandkumar should apologize to all those on Twitter she has tried to persecute with her cancel list. Retweet if you agree.
JJ retweeted
My crash course lectures on (Deep) Reinforcement Learning:
MDP: youtu.be/OiEuYkG_B7w
Q-Learning: youtu.be/GvQ2kcPhFgE
Policy Gradient: youtu.be/HsIRC_gZT7g
For more extensive lectures, I would also recommend @svlevine 's lectures on Deep RL: rail.eecs.berkeley.edu/deepr…
JJ retweeted
Stuck at home and bored waiting for the US election results? Cauchy residues come to the rescue with this month blog post! francisbach.com/cauchy-resid…
