Opens profile photo
Follow
Click to Follow erfaneshrati
Erfan
@erfaneshrati
PhD, Research Engineer
Los Angeles, CAamirerfan.comJoined September 2013

Erfan’s Tweets

Picking up inspiring indications that the protests in Iran are gathering strength. The BBC reporter concluded, “Who knows where it will end?” Decent people all over the world must hope it will end with the toppling of the disgusting Ayatollahs and their Islamic tyranny.
804
15.2K
Get faster, more flexible inference on GPUs using our newly open-sourced AITemplate, a revolutionary new inference engine that delivers up to 12X performance improvements on NVIDIA GPUs & 4X on AMD GPUs compared to eager-mode within Pytorch. Learn more: bit.ly/3rl8F3b pic.twitter.com/3HMYZuYJ52
12
777
According to Bernard Lewis, after the Arab Conquest, "Iran was indeed Islamized, but it was not Arabized." But Iranians turned to Islam to save their lives, and by inventing Shi'ite they gradually expelled the Arab version of Islam.
1
2
Show this thread
GPT-3 is amazing at complex tasks like creative writing and summarizing. But it's surprisingly bad at reversing words. 🤔 The reason is that GPT-3 doesn't see the world the way we humans do. 👀 If you teach it to reason, it can get around its limitations to get really good. 💡
Image
31
1,422
Show this thread
A variant I always try: import cudf as pd Not kidding. Many times pandas code runs out of the box with cudf, only way faster.
Quote Tweet
Want to speedup Pandas DataFrame operations? Let me share one of my Kaggle tricks for fast experimentation. Just convert it to cudf and execute it in GPU using this 1 liner. #Kaggle #MachineLearning #Python #pandas #Nvidia #rapidsai #cudf
Show this thread
Image
9
652
DALL.E can be viewed as a world model represented in the joint space of vision and language. Really impressive work 🤩
Quote Tweet
Our newest system DALL·E 2 can create realistic images and art from a description in natural language. See it here: openai.com/dall-e-2/
Embedded video
0:30
873.9K views
2
I’ve been following for years, and on and off participating in the tournament. I have always been a very big fan of what they are trying to accomplish. And now after attending #NUMERCON, I can confidently say that they have the best competitive ML setup in the World.
Image
3
123
deep learning research has been mostly focused on unstructured data (image, video, text, audio). xgboost is still on top for tabular data.
Quote Tweet
XGBoost Is All You Need Deep Neural Networks and Tabular Data: A Survey arxiv.org/pdf/2110.01889
Image
research in academia is economically broken and academics often have competitve mindset. this imbalance in the desire and available resources have eventually lead to creating abusive phd research environments so that many talents are leaving academia.
Quote Tweet
I’ve decided to leave my PhD after a lot of personal challenges due to mental health/COVID, which also revealed the frustrating ways that academia and I are less-than-compatible. Not sure what’s next (other than continuing to do math, because I literally cannot help myself)!
4