Salar Abbaspour (@salixabb) header
Salar Abbaspour (@salixabb) profile picture

ML @apple | https://teapawt.org/t.co/AXXxtAwpI4

Joined May 2017
Photo by Salar Abbaspour (@salixabb)
we definitely live in a bubble if jacobian conjecture disproof is not in the top 10 trending topics of today, despite all the attention it's (deservedly) getting on timeline!!
hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final ((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)
3
203
Placebo effect for LLMs? 🤔
Telling mixtral that it is "ChatGPT developed by OpenAI" boosts humaneval score by 6%
3
702
New in @natBME, we present an #AI method that achieves accurate and flexible inference for #brain-computer interfaces. 🎉 It can infer causally in real time, non-causally, & even with missing neural samples, as can happen in wireless #BCI 🧠💻 🧵& code ⬇️ nature.com/articles/s41551-0…
3
38
1
160
34,629
Salar Abbaspour retweeted
Just in time for the holidays, we are releasing some new software today from Apple machine learning research. MLX is an efficient machine learning framework specifically designed for Apple silicon (i.e. your laptop!) Code: github.com/ml-explore/mlx Docs: ml-explore.github.io/mlx/bui…
95
671
135
4,375
962,041
Salar Abbaspour retweeted
Am watching @PyTorch conference keynote and thought to give the torch.compile decorator for NumPy a try. Whoa, in a quick benchmark on my MacBook, I indeed got a ~35x speedup when compiling NumPy code with PyTorch. Exciting times. (Think of the ramifications for scikit-learn).
22
205
9
1,697
266,261
Fall 2023 Internship Join us at NVIDIA and be part of a highly esteemed AIAlgo team at the forefront of AI research and development. @AnimaAnandkumar nvidia.wd5.myworkdayjobs.com…
6
22
1
107
56,020
Salar Abbaspour retweeted
Chain ⛓️ Rule(s) rules! Appreciation thread of one of the most interesting coincidences in machine learning. Two rules, both named "Chain Rule", happen to be absolutely critical to recent advances in ML & AI. A 🧵 on the Chain Rule of Probability & the Chain Rule of Calculus👇
2
43
2
290
95,147
Salar Abbaspour retweeted
Shoutout to the legend who answers literally every single PyTorch question out there!
54
191
42
3,881
645,831
I wrote a VSCode extension that allows you to use ChatGPT within the context of your code 🔥 You can click on code blocks to paste them into the editor, or use a selected code snippet as a reference. The extension is on GitHub if you want to check it out github.com/mpociot/chatgpt-v…
113
879
132
5,148
In #DiffusionModels, often, a continuous function in time is computed to generate data. In this work, we develop a new #NeuralOperator architecture that allows us to directly predicts this function in one model evaluation, which is amazing. It achieves SOTA in both speed and FID.
Fast sampling of diffusion models. Only one model evaluation achieves SOTA! Check out poster at #NeurIPS22 SBM workshop on Friday 2:30pm-4pm at Room 293 - 294. @Kay12400259 @wn8_nie @ArashVahdat @Azizzadenesheli @nvidia @caltech
3
14
Salar Abbaspour retweeted
Curiouser and curiouser #stablediffusion @deforum_art
44
177
34
1,211
A short but insightful summary of variational diffusion models and score-based generative models. Look forward to the blog post!
Excited to share with everyone an accessible, intuitive tutorial on diffusion models! If you're curious about the math behind diffusion models and how their different interpretations can be unified, please check it out! Stay tuned for a blog post soon! arxiv.org/abs/2208.11970
1
1
Salar Abbaspour retweeted
Training large models demands a lot of GPU memory and a long training time. With several training parallelism strategies and a variety of memory saving designs, it is possible to train very large neural networks across many GPUs. lilianweng.github.io/lil-log…
4
158
13
862
Great review on contrastive representation learning. Still amazes me that I had come across this blog post before this was tweeted.
Contrastive learning aims to learn representation such that similar samples stay close, while dissimilar ones are far apart. It can be applied to supervised / unsupervised data and has been shown to achieve good results on various tasks. 📚 A long read: lilianweng.github.io/lil-log…
2
Salar Abbaspour retweeted
Update: Python users can now easily install PSID using pip. Also, we now have a Jupyter notebook with some explanations and usage examples for PSID to help get you started: github.com/ShanechiLab/PyPSI… For more info about PSID, see our @NatureNeuro paper: rdcu.be/b993t
Our @NatureNeuro paper is out, first-authored by my student @omidsani. It develops PSID, a novel dynamic method to dissociate & model behaviorally relevant neural dynamics. Thanks to @salixabb, Y Wong & collaborator @pesaranlab. Summary & code below: 1/n nature.com/articles/s41593-0…
1
2
19
A visual search engine for finding images with similar vibes, and not necessarily similar shapes. So impressive.
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
5
Our paper is out @NatureComms! Huge shoutout to my PhD advisor @MaryamShanechi for her unconditional support and supervision. Also big thanks to the coauthors @yantatwong, Mahdi, and particularly @bijanpesaran/@pesaranlab for their much needed vision & contributions. Summary:👇
2
2
22
Salar Abbaspour 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/
90
899
328
2,872
It'd been for a while that this draft had been sitting in my drawer with no hope to come out. Finally, putting it out. Theme: Avocado Boba Paper: arxiv.org/abs/2011.14251 Label shift is cool, interesting, important & fundamental. Stay tuned to c soon these methods being applied;)
1
3
49