2022 𝐑𝐞𝐜𝐚𝐩: 𝐌𝐋 𝐞𝐝𝐢𝐭𝐢𝐨𝐧

1) DALLE-2 by OpenAI uses latent diffusion models to generate realistic images and art from text prompts
openai.com/dall-e-2/

🧵 2) AlphaFold by DeepMind predicts 200 million protein structures, almost every known protein

deepmind.com/research/highl…

𝐁𝐫𝐨𝐚𝐝𝐜𝐚𝐬𝐭𝐢𝐧𝐠 𝐢𝐧 𝐭𝐞𝐧𝐬𝐨𝐫 𝐨𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐬
Numpy, Pytorch and other frameworks use broadcasting in order to execute arithmetic operations between tensors of different shapes.

Understanding broadcasting is essential as it's the source of many bugs.

🧵⬇️ In a nutshell, to execute let's say an addition between two tensors, we have to make sure that they have the same shape. To achieve that, we stretch the dimensions of the smaller tensor to match the size of the corresponding dimensions of the larger tensor.

⬇️

𝐂𝐚𝐥𝐢𝐛𝐫𝐚𝐭𝐢𝐨𝐧 𝐢𝐧 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠
Calibration is the property that tells us how well the estimated probabilities of a model match the actual probabilities, a.k.a the observed frequency of occurrences.

1/7 🧵⬇️ In other words, when building ml models we'd like that the estimated probability of an example belonging to a particular class is as close as possible to the actual frequency of this class.

2/7⬇️

📚Best resources to learn about diffusion models 📚

1) What are diffusion models by @lilianweng lilianweng.github.io/posts/2021-07-…

2) Introduction to Diffusion Models for Machine Learning by @AssemblyAI
assemblyai.com/blog/diffusion…

🧵⬇️ 3) DDPM - Diffusion Models Beat GANs on Image Synthesis (Machine Learning Research Paper Explained) by @ykilcher


⬇️

If your new year's resolution is to learn Deep Learning, here are the top books to read in 2022. We spent time reviewing each one of them so we can help you choose the one that suits your needs.

theaisummer.com/deep-learning-…

🧵👇 The Hundred-Page Machine Learning Book by @burkov is the best one to get you started in ML. Short, with great visualizations, consistent math notation, and a wide range of topics, is an excellent choice for your first contact with the field
1/8 👇