CUDA

Matrix Multiplication in CUDA

Matrix multiplication is at the heart of deep learning. In this evolving world of LLMs, the need for fast and efficient matrix multiplications is paramount. Nvidia CUDA allows...

Computer Vision

PyTorch Basic Tutorial

PyTorch libraries torchvision: for computer vision torchtext: for NLP torchaudio: for speech

Deep Learning

Introduction to Panoptic Segmentation: A Tutorial

In semantic segmentation, the goal is to classify each pixel into the given classes. In instance segmentation, we care about segmentation of the instances of objects separately. The...

Deep Learning

Evaluation metrics for object detection and segmentation: mAP

Read about semantic segmentation, and instance segmentation.

Deep Learning

Quick intro to Instance segmentation: Mask R-CNN

This is the third post in the Quick intro series: object detection (I), semantic segmentation (II).

Deep Learning

Quick intro to semantic segmentation: FCN, U-Net and DeepLab

Suppose you’ve an image, consisting of cats. You want to classify every pixel of the image as cat or background. This process is called semantic segmentation.

Deep Learning

Converting FC layers to CONV layers

It is worth noting that the only difference between FC and CONV layers is that the neurons in the CONV layer are connected only to a local region...

Deep Learning

Data augmentation

Deep Learning

Generative Adversarial Networks variants: DCGAN, Pix2pix, CycleGAN

First, make sure you read the first part of this post, Generative models and Generative Adversarial Networks. This post is its continuation.

Deep Learning

Layer-specific learning rates

Note: This post answers the question I asked on Cross-Validated: What are the benefits of layer-specific learning rates?

Deep Learning

Quick intro to Object detection: R-CNN, YOLO, and SSD

Object detection deals with the detection of object instances in an image. There are a number of methods to accomplish it. The following post summarizes few important object...

Deep Learning

Attention

Attention is the technique through which the model focuses itself on a certain region of the image or on certain words in a sentence just like the same...

Deep Learning

Backpropagation Through Time

In the earlier post The gradient problem in RNN, you came to know that RNNs have exploding and vanishing gradient problems. This post discusses how exactly the gradients...

Deep Learning

Autoencoder: Downsampling and Upsampling

Note: Read the post on Autoencoder written by me at OpenGenus as a part of GSSoC.

Deep Learning

Weight initialization in neural nets

In deep learning, the goal is to find the optimum weights of the model to get the desired output. In transfer learning, the network is initialized using the...

Deep Learning

Image captioning using encoder-decoder

Note: This post discusses the project I’m doing currently. Check it on kHarshit/image-captioner. The working demo is available on https://imagecaptioner.herokuapp.com.

Deep Learning

The gradient problem in RNN

Much as Convolutional Neural Networks are used to deal with images, the Recurrent Neural Networks are used to process sequential data. The key idea in recurrent neural networks...

Deep Learning

Why Batch Normalization?

As mentioned in Scaling Vs Normalization, applying normalization to the input features increases the convergence rate of our algorithm i.e. it speeds up the learning process.

Deep Learning

Filters in Convolutional Neural Networks

Note: This post is inspired by the answer I gave on stackoverflow.

Deep Learning

Loss vs Accuracy

A loss function is used to optimize the model (e.g. a neural network) you’ve built to solve a problem.

Deep Learning

Generative models and Generative Adversarial Networks

In supervised learning, we have data x and response (label) y and the goal is to learn a function to map x to y e.g. regression, classification, object...

Deep Learning

Skip connections and Residual blocks

Deep neural networks are difficult to train. They also have vanishing or exploding gradient problems. Batch normalization helps, but with the increase in depth, the network has trouble...

Data Science

Loss functions

In machine learning, the difference between the predicted output and the actual output is used to tune the parameters of the algorithm. This error in prediction, so called...

Data Science

Optimizers

Building neural networks, we compute gradients with the backpropagation algorithm. These gradients are used to perform the parameter updates. The default way of doing this is to update...

Deep Learning

Transfer learning: How to build accurate models

A good Covolutional Neural Network model requires a large dataset and good amount of training, which is often not possible in practice. Transfer learning provides a turn around...

Data Science

Methods of Hyperparameter optimization

The parameters, called hyperparameters, that define the performance of the machine learning algorithm (model), depends on the problem we are trying to solve. Thus, they need to be...

Deep Learning

word2vec: The foundation of NLP

Natural Language Processing deals with the task of making computer understand the human language. The computer understands only 0s and 1s. A language consists of words. So, the...

Data Science

Dropout: Prevent overfitting

Dropout is a regularization technique that prevents neural networks from overfitting. Regularization methods like L2 and L1 reduce overfitting by modifying the cost function. Dropout, on the other...

Data Science

How deep should neural nets be?

How do we decide on what architecture to use while solving a problem using neural networks? Should we use no hidden layers? One hidden layer? Two hidden layers?...

Data Science

Don't use sigmoid: Neural Nets

In neural networks, activation functions are used to introduce non-linearity in the model. There are several activation functions to choose from. Traditionally, people have been using sigmoid as...

Deep Learning

The magic behind ConvNets

Ordinary neural networks consists of neurons that have learnable weights and biases. The input is a single vector (of features) that is transfomed through a number of hidden...

Data Science

Computational graphs: Backpropagation

Backpropagation is an efficient method of computing gradients in directed graphs of computations, such as neural networks. This is not a learning method, but rather a nice computational...

Data Science

Gradient descent: The core of neural networks

As discussed in the post linear algebra and deep learning, the optimization is the third and last step in solving image classification problem in deep learning. It helps...

Data Science

Linear algebra: The essence behind deep learning

Mathematics lies behind every algorithm; if not mathematics then mathematical thinking. In case of deep learning algorithms, linear algebra is the driving force.