Bayesian methods promise to fix many shortcomings of deep learning, but they are impractical and rarely match the performance of standard methods, let alone improve them. In this paper, we demonstrate practical training of deep networks with natural-gradient variational inference. By applying techniques such as batch normalisation, data augmentation, and distributed training, we achieve similar performance in about the same number of epochs as the Adam optimiser, even on large datasets such as ImageNet. Importantly, the benefits of Bayesian principles are preserved: predictive probabilities are well-calibrated, uncertainties on out-of-distribution data are improved, and continual-learning performance is boosted. This work enables practical deep learning while preserving benefits of Bayesian principles.

Bayesian methods promise to fix many shortcomings of deep learning, but they are impractical and rarely match the performance of standard methods, let alone improve them. In this paper, we demonstrate practical training of deep networks with natural-gradient variational inference. By applying techniques such as batch normalisation, data augmentation, and distributed training, we achieve similar performance in about the same number of epochs as the Adam optimiser, even on large datasets such as ImageNet. Importantly, the benefits of Bayesian principles are preserved: predictive probabilities are well-calibrated, uncertainties on out-of-distribution data are improved, and continual-learning performance is boosted. This work enables practical deep learning while preserving benefits of Bayesian principles.

Originality: Rather low The main technical novelty lies in applying tricks from the deep learning literature to VOGN. The experiments are fairly standard. Quality: High That being said, the experiments seem to be carefully executed, described in detail and the overall method is technically sound. While not overly ambitious in terms of technical novelty, I think this is a well-executed piece of work. Clarity: High The paper is well-written and easy to follow.

The paper demonstrates that the Variational Online Gauss-Newton (VOGN) method of Khan et al. (2018) can be successfully scaled to deep learning architectures. The authors demonstrated the scalability of Bayesian methods to large scale data such as ImageNet. Extensive experiments on large scale data and models are provided. The main result is an adoption of an existing model (VOGN) to make it practical for deep learning.

Bayesian methods promise to fix many shortcomings of deep learning, but they are impractical and rarely match the performance of standard methods, let alone improve them. In this paper, we demonstrate practical training of deep networks with natural-gradient variational inference. By applying techniques such as batch normalisation, data augmentation, and distributed training, we achieve similar performance in about the same number of epochs as the Adam optimiser, even on large datasets such as ImageNet. Importantly, the benefits of Bayesian principles are preserved: predictive probabilities are well-calibrated, uncertainties on out-of-distribution data are improved, and continual-learning performance is boosted. This work enables practical deep learning while preserving benefits of Bayesian principles.

TensorFlow is by far, the most popular library for deep learning. Backed by Google, it is a solid investment of your time and efforts if you want to succeed in the area of machine learning and AI. The issue most people face is that getting started with Tensorflow guides usually delve too deeply into unnecessary mathematics. That is where this course comes in. While some theory is important, a lot of it is just not needed when you're just getting started!

How to build user-facing AI products. How to train models quickly. How to minimize the code and effort required in prototyping. How to make models more performant and energy-efficient. How to operationalize and scale, and estimate the costs involved.

My infatuation with computers began with an Apple II in 1981. I've been active in machine learning since 2003, and deep learning since before AlexNet was a thing. My background includes a Ph.D. in computer science from the University of Colorado, Boulder (deep learning), and an M.S. in physics from Michigan State University. By day, I work in industry building deep learning systems. By night, I type away on my keyboard generating the books you see here.

Be able to run deep learning models with Keras on Tensorflow 2 backend Run Deep Neural Networks on a real-world scientific protein dataset Understand how to feed own data to deep learning models (i.e. I answer questions on the same day. Understand how to feed own data to deep learning models (i.e. Understand and use Keras' functional API to create models with multiple inputs and outputs I answer questions on the same day. You should be able to use Python (if, while, lists.

This course is for you if you are new to Machine Learning but want to learn it without all the math. This course is also for you if you have had a machine learning course but could never figure out how to use it to solve your own problems. In this course, we will start from very scratch. This is a very applied course, so we will immediately start coding even without installation! You will see a brief bit of absolutely essential theory and then we will get into the environment setup and explain almost all concepts through code.