Sentim retweeted
Visualizing Weights — A new Distill article by @csvoss, @nickcammarata, @gabeeegoooh, @mpetrov, @ludwigschubert, Ben Egan, Swee Kiat Lim, and @ch402. This is the seventh article in the circuits thread.
distill.pub/2020/circuits/vi…
Sentim retweeted
High-Low Frequency Detectors — A new Distill article by
@ludwigschubert, @csvoss, and @ch402. This is the fifth article in the circuits thread.
distill.pub/2020/circuits/fr…
Sentim retweeted
We’ve developed two neural networks which have learned by associating text and images. CLIP maps images into categories described in text, and DALL-E creates new images, like this, from text.
A step toward systems with deeper understanding of the world. openai.com/multimodal/
Sentim retweeted
I'm happy to announce that our MuZero paper has been published in Nature: go.nature.com/38qb2I8 !
Please see our official blog post https://xitter.cf/t.co/buMXAhMarh for more; also see my personal blog for a high level description of MuZero: furidamu.org/blog/2020/12/22…
Sentim retweeted
Language models like GPT-2 memorize text from their training data.
This blog discuss the legal and privacy implications. Are models:
* subject to GDPR's "right to be forgotten" law?
* infringing on copyright?
* illegally misusing data?
bair.berkeley.edu/blog/2020/…
Sentim retweeted
Interpretability of AI systems is crucial for their adoption, but how can it be achieved, and where is the research at?
Read our new overview by Owen Shen on that:
thegradient.pub/interpretabi…
Sentim retweeted
Google's `Hum to Search`
ai.googleblog.com/2020/11/th…
Sentim retweeted
Adapting to your data distribution changing after you train your model remains one of the big challenges of Machine Learning. Check out our latest blog post on work addressing this challenge -
ai.stanford.edu/blog/adaptiv…
Sentim retweeted
A system capable of answering any factual questions can enable many useful applications. My new post delves into how we can build an open-domain question answering model, with neural networks or with access to a powerful pre-trained language model. lilianweng.github.io/lil-log…
Sentim retweeted
New blog post!
Fine-tuning ResNet-18 for Audio Classification: Reproducible experiments using fastai, fastaudio, and wandb sweeps to explore hyperparameters for achieving high accuracy on the ESC-50 dataset.
wandb.ai/jhartquist/fastaudi…
Special thanks to @fastdotai @wandb
Sentim retweeted
Transformer’s attention mechanism can be linked to other cool ideas in AI
- Indirect Encoding in Neuroevolution
attentionagent.github.io
- Hopfield Networks with continuous states
arxiv.org/abs/2008.02217
- Graph Neural Networks with multi-head attention
thegradient.pub/transformers…
Sentim retweeted
Generalizing Randomized Smoothing for Pointwise-Certified Defenses to Data Poisoning Attacks – Machine Learning Blog | ML@CMU | Carnegie Mellon University blog.ml.cmu.edu/2020/09/25/g…
Sentim retweeted
New preprint alert! 🚨
Introducing GeDi (pronounced Jedi): A Powerful New Method for Controlling Language Models.
Paper: arxiv.org/abs/2009.06367
Code: github.com/salesforce/GeDi
Blog: blog.einstein.ai/gedi/
This paper has a bunch of really cool results. Here are a few.
Sentim retweeted
Evan wrote an excellent blog post on this work, which is now out on the SAIL blog:
ai.stanford.edu/blog/meta-ex…
This work changed how I think about exploration in meta-RL & sheds light on why end-to-end optimization is not a good idea in this setting.
Want your robot to explore intelligently? We study how to learn to explore & introduce a *efficient* meta-learning method that can lead to optimal exploration.
Paper: arxiv.org/abs/2008.02790
w Evan Liu, Raghunathan, Liang @StanfordAILab
Thread👇🏼(1/5)
youtube.com/watch?v=EiIC0Rkz…
Sentim retweeted
Self-classifying MNIST Digits: Achieving Distributed Coordination with Neural Cellular Automata
Each cell trying to spread what they believe to be the objective truth to their neighbors.
Sometimes the truth prevails, but they are collectively misguided.
distill.pub/2020/selforg/mni…
Sentim retweeted
Ever wondered how to do machine learning on sets? Here's a fresh blog post on the topic, enjoy! akosiorek.github.io/ml/2020/…
Sentim retweeted
Zero-Shot Learning in Modern NLP buff.ly/3glmk2A
Sentim retweeted
Does adv. robustness help transfer learning? W/ @hadisalmanX @andrew_ilyas @logan_engstrom @akapoor_av8r @MSFTresearch we show l2-robust ImageNet models transfer better, despite lower accuracy: arxiv.org/abs/2007.08489.
Blog: gradientscience.org/transfer…
Code: github.com/microsoft/robust-…
Sentim retweeted
There is a lot to wrap your head around in LSTMs🤯. One way of thinking that helped me a lot is the 'conveyor belt' metaphor of the cell state 🧑🏭 by @ch402. I put together a little animation 🖼️ Check out the amazing blog post by Chris Olah here✍️: colah.github.io/posts/2015-0…


























