About me
Hello, I am Aakash Kumar Nain.
I am a Machine Learning Engineer with 8 years of experience in the industry. Currently, I am working as a Senior ML Engineer at Merlyn Mind. I am also a Google Developers Expert in Machine Learning and JAX. I actively contribute to Keras, TensorFlow, and the JAX ecosystem. I am one of the core collaborators for the Keras core project, and I am also one of the maintainers of the TensorFlow-addons package.
In terms of work, my knowledge Machine Learning and Deep Learning spans in T-shape. I have worked in multiple domains, but Computer Vision with Deep Learning is my favorite field. I love simplifying Data Science and Machine Learning concepts, especially the maths behind them. Love and support OSS because, without OSS, we wouldn’t have witnessed so many breakthroughs in Machine Learning and Deep Learning. Python is my favorite language to work with, and I absolutely love coding in it. Always up for a discussion involving anything related to Machine Learning, Deep Learning, MLOps, and API design. Simply put: Python in breakfast, Machine Learning in lunch, and Deep Learning in dinner!
Follow me on Twitter to get the latest updates related to Machine Learning.
Portfolio
Popular Projects
Most researchers and Machine Learning engineers want to remain up-to-date with the latest research work, but the field of machine learning moves crazy fast. The number of papers uploaded on arXiv is witnessing exponential growth. Reading all these published papers is an impossible task. To help with this, I maintain this repository where I annotate and upload the papers that I find exceptionally good. Annotations make it easier to understand and grasp the main concepts explained in the paper.
It is a series of tutorials built to teach the fundamental concepts and the underlying working of two famous libraries: TensorFlow and JAX. These tutorials aren’t typical documentation-type tutorials. It doesn’t matter whether you are a beginner or an advanced user, these tutorials will give you a fresh perspective on building things using TensorFlow or JAX.
Diffusion models are a class of likelihood-based generative models that recently have been used to produce very high-quality images compared to other existing generative models like GANs. It’s hard to find quality resources to learn about diffusion models. The mathematics behind the diffusion models is also a bit harder to understand. The material presented in this repo is enough to make you understand the working of diffusion models and the underlying math.
This is a direct port of Mitral-7B model in JAX and Equinox. This port comes with two implementations: The first one shows how to port models from PyTorch to Equinox and JAX. This is a 1:1 mapping, and not fully optimized. The second implementation is fully optimized for JAX, targeted for more advanced users.
Keras Core Contributions
- JAX NN ops
- Merging layers
- Metrics
- Loss functions
- Applications
- Data adapters
- Code examples