Seq2seq architectures process input elements sequentially (the output upto the previous element is used in the processing of the current element). This is slow, can’t be parallelised, suffers from vanishing gradient and has trouble with context when processing long sequences. Attention comes to solve all of this in the form…

EfficientDet is an improvement upon EfficientNet, so we’ll look at the latter first. EfficientNet The paper sets out to explore the problem of given a baseline model i.e. CNN architecture how can we scale the model to get better accuracy. There are 3 ways to scale a model. Resolution — Input…

Understanding ROC and AUC for binary classification models — I’m not a big fan of pretentious statistical terms. Everything should have a need and use case. In this article I try to explain the ROC curve and the ROC-AUC (area under the curve) by comparing different models and developing intuition to satisfy myself. Consider the example of testing a…

How Deep Learning libraries handle gradients — How do you compute derivatives of a function? By hand, you’d find it analytically. You can then write a function to compute it. Or you can write a function to approximate it. def f(x): return x*x // function def df(x): return 2*x // analytical derivative def df(x,f,e): return (f(x+e) - f(x))/e // approximation This can get quite cumbersome…

Grand Central Dispatch or GCD is a framework built by Apple to allow for concurrent execution of code. It abstracts a lot of painful and tedious code necessary to smoothly do parallel computation. I have seen many stack-overflow questions over the terms in the title of the article and caused…

Comes in handy during interviews. For beginners and experts alike — Defining a Neural network The way we define the architecture of a fully connected feed-forward neural network is using hyperparameters, mentioning the number of layers, the number of neurons in each layer and the activation function for each layer.

Looking up human pose estimation online will eventually lead you to one of these papers: OpenPose, Stacked Hourglass Networks, Convolutional Pose Machines. We will look at CPMs as it is a very simple and neat approach. Pose Machines Apparently, pose machines existed before deep learning took off and used random forests for…

In a previous post, I had discussed the basics of the Metal framework as intuitively as possible. Please do give it a read before beginning with this article. You’ll need a device that supports Metal to run the code as Metal is not supported by the Xcode’s device simulator The…

If you have clicked on this article I'm assuming you know what a GPU is and since it’s Apple, a bit of Swift too. Apple made Metal to give developers low-level access to the GPUs on iPhones and Mac for rendering 3D graphics and do high-performance computing. This is an…