Can you deploy your scikit learn models in C++?

I've had many people reach out to me to ask me about deploying ML models in C++.
​
Most of them have read my article on Medium about the subject and then contacted me for further questions.
​
I've deployed many deep learning models in a C++ production environment but most of them were focused on computer vision tasks (image recognition, object detection, ...).
​
But I noticed that many of those who contacted me were working on regression models for other industries. Specifically, they were using scikit learn library to build their models in Python.
​
So to answer the question above, yes, If you are building ML models using scikit-learn then you can still deploy your models in C++.
​
You can do this using ONNX Runtime which is a C++ library that can read ONNX models.
​
But you have to first convert your scikit-learn models to ONNX format. You can do this using the Python package skl2onnx.
​

​

How do diffusion models work?

With all the news lately about image generation models such as DALL-E and Stable Diffusion I've become very curious as to what are these diffusion models. So I’ve been reading lately about them to try to understand how they differ from GANs. Here’s what I learned so far :

• Diffusion models and GANs belong to the same category in deep learning which is generative modeling.
• Diffusion models are based on “Markov Chains” but GANs are based on a generator-discriminator paradigm.
• Diffusion models are relatively more stable than GANs for generating images.
• Diffusion models are trained by adding Gaussian Noise to the training images in a series of timesteps, and then learning to undo this noising process.
• Once diffusion models are trained, the denoising model can then be iteratively applied to a randomly sampled Gaussian noise.
• “Denoising” this Gaussian noise will generate novel images.

​

​