Convolutional Neural Net in Tensorflow

One of the most exciting areas of deep learning is computer vision. Through recent advances in convolutional neural nets we have been able to create self driving cars, facial detection systems and automated medical imagery analysis that out performs specialists just to name a few. In this article I will show you the fundamentals of convolutional neural nets and how you can create one yourself to classify hand written digits.

An Overview of Gradient Descent Optimization Algorithms

Gradient descent optimization algorithms, while increasingly popular, are often used as black-box optimizers, as practical explanations of their strengths and weaknesses are hard to come by. This article aims to provide the reader with intuitions with regard to the behaviour of different algorithms that will allow her to put them to use. In the course of this overview, we look at different variants of gradient descent, summarize challenges, introduce the most common optimization algorithms, review architectures in a parallel and distributed setting, and investigate additional strategies for optimizing gradient descent.

Are people overly infatuated with Deep Learning, and can it really deliver?

One of the factors often credited for this latest boom in artificial intelligence (AI) investment, research, and related cognitive technologies, is the emergence of Deep Learning AI algorithms, and the corresponding large volumes of big data and computing power that makes Deep Learning a reality. However, deep learning is just an approach to machine learning (ML), that while having proven much capability across a wide range of problem areas, is still just one particular approach. Increasingly, we’re starting to see news and research showing the limits of deep learning, and some of the downsides to the deep learning approach. So we have to ask, are people’s enthusiasm of AI tied to their enthusiasm of deep learning, and is deep learning really able to deliver on many of its promises?

Weave – Web-based Analysis and Visualization Environment

Weave (Web-based Analysis and Visualization Environment) is a visualization platform designed to enable visualization of any available data by anyone for any purpose. Weave is an application development platform supporting multiple levels of user proficiency – novice to advanced – as well as the ability to integrate, disseminate and visualize data at ‘nested’ levels of geography.

Stemming and Lemmatization in Python

Stemming and Lemmatization are Text Normalization (or sometimes called Word Normalization) techniques in the field of Natural Language Processing that are used to prepare text, words, and documents for further processing. Stemming and Lemmatization have been studied, and algorithms have been developed in Computer Science since the 1960’s. In this tutorial you will learn about Stemming and Lemmatization in a practical approach covering the background, some famous algorithms, applications of Stemming and Lemmatization, and how to stem and lemmatize words, sentences and documents using the Python nltk package which is the Natural Language Tool Kit package provided by Python for Natural Language Processing tasks.

Machine learning logistic regression for credit modelling in R

Machine learning logistic regressions is a widely popular method to model credit modeling. There are excellent and efficient packages in R, that can perform these types of analysis. Typically you will first create different machine learning visualizations before you perform the machine learning logistic regression analysis.

Fluid Annotation: An Exploratory Machine Learning–Powered Interface for Faster Image Annotation

The performance of modern deep learning-based computer vision models, such as those implemented by the TensorFlow Object Detection API, depends on the availability of increasingly large, labeled training datasets, such as Open Images. However, obtaining high-quality training data is quickly becoming a major bottleneck in computer vision. This is especially the case for pixel-wise prediction tasks such as semantic segmentation, used in applications such as autonomous driving, robotics, and image search. Indeed, traditional manual labeling tools require an annotator to carefully click on the boundaries to outline each object in the image, which is tedious: labeling a single image in the COCO+Stuff dataset takes 19 minutes, while labeling the whole dataset would take over 53k hours!

Introduction to Active Learning

An extensive overview of Active Learning, with an explanation into how it works and can assist with data labeling, as well as its performance and potential limitations.

Announcing MCHT: An R Package for Bootstrap and Monte Carlo Hypothesis Testing

I am very excited to announce my first (public) package (and the second package I’ve ever written, the first being unannounced until the accompanying paper is accepted). That package is MCHT, a package for bootstrap and Monte Carlo hypothesis testing, currently available on GitHub. This will be the first of a series of blog posts introducing the package. Most of the examples in the blog posts are already present in the manual, but I plan to go into more depth here, including some background and more detailed explanations. MCHT is a package implementing an interface for creating and using Monte Carlo tests. The primary function of the package is MCHTest(), which creates functions with S3 class MCHTest that perform a Monte Carlo test.

Maximized Monte Carlo Testing with MCHT

I introduced MCHT two weeks ago and presented it as a package for Monte Carlo and boostrap hypothesis testing. Last week, I delved into important technical details and showed how to make self-contained MCHTest objects that don’t suffer side effects from changes in the global namespace. In this article I show how to perform maximized Monte Carlo hypothesis testing using MCHT.

Mysterious Ellipsis: Tutorial

If you have any basic experience with R, you probably noticed that R uses three dots ellipsis (…) to allow functions to take arguments that weren’t pre-defined or hard-coded when the function was built. Even though R beginners are usually aware of this behavior, especially due to some common functions that implement it (for example, paste()), they are often not using it enough in their own functions. In other cases, the ellipsis is just not used properly or not fully taken advantage of. In this tutorial we will go through some common mistakes in using the ellipsis feature, and some interesting options to fully utilize it and the flexibility that it offers.

Searching for R Packages

Searching for R packages is a vexing problem for both new and experienced R users. With over 13,000 packages already on CRAN, and new packages arriving at a rate of almost 200 per month, it is impossible to keep up. Package names can be almost anything, and they are rarely informative, so searching by name is of little help. I make it a point to look at all of the new packages arriving on CRAN each month, but after a month or so, when asked about packages related to some particular topic, more often than not, I have little more to offer than a vague memory that I saw something that might be useful. Fortunately, package developers have provided some very useful tools, if you know where to look. 🙂 This post presents a search strategy based on some relatively new packages I have come across in my monthly review.

Transfer learning from pre-trained models

Deep learning is fast becoming a key instrument in artificial intelligence applications (LeCun et al. 2015). For example, in areas such as computer vision, natural language processing, and speech recognition, deep learning has been producing remarkable results. Therefore, there is a growing interest in deep learning. One of the problems where deep learning excels is image classification (Rawat & Wang 2017). The goal in image classification is to classify a specific picture according to a set of possible categories. A classic example of image classification is the identification of cats and dogs in a set of pictures (e.g. Dogs vs. Cats Kaggle Competition). From a deep learning perspective, the image classification problem can be solved through transfer learning. Actually, several state-of-the-art results in image classification are based on transfer learning solutions (Krizhevsky et al. 2012, Simonyan & Zisserman 2014, He et al. 2016). A comprehensive review on transfer learning is provided by Pan & Yang (2010). This article shows how to implement a transfer learning solution for image classification problems. The implementation proposed in this article is based on Keras (Chollet 2015), which uses the programming language Python. Following this implementation, you will be able to solve any image classification problem quickly and easily.

Random forests and decision trees from scratch in python

Random forest is the prime example of ensemble machine learning method. In simple words, an ensemble method is a way to aggregate less predictive base models to produce a better predictive model. Random forests, as one could intuitively guess, ensembles various decision trees to produce a more generalized model by reducing the notorious over-fitting tendency of decision trees. Both decision trees and random forests can be used for regression as well as classification problems. In this post we create a random forest regressor although a classifier can be created with little alterations in the following code.

Reconstructive Principle Component, Global Contrast, Ranged Normalization Layer in Tensorflow

Today I wanted to practice my matrix calculus as well as be creative with creating some layers. And I thought it would be a good idea to implement different layers that can be used later. Below is the list of layers I wanted to practice.

Deep Learning Cloud Service Providers

An Introduction to Bayesian Inference in PyStan

The many virtues of Bayesian approaches in data science are seldom understated. Unlike the comparatively dusty frequentist tradition that defined statistics in the 20th century, Bayesian approaches match more closely the inference that human brains perform, by combining data-driven likelihoods with prior beliefs about the world. This kind of approach has been fruitfully applied in reinforcement learning, and efforts to incorporate it into deep learning are a hot area of current research. Indeed, it has been argued that Bayesian statistics is the more fundamental of the two statistical schools of thought, and should be the preferred picture of statistics when first introducing students to the subject.

Reinforcement Learning: Towards General AI

So, How Does Reinforcement Learning Work? Imagine a robot (Also known as an agent) is trying to pick up a pen, and it fails. It tries again, fails. After repeating this process 1000 times, it finally succeeds. The agent has now learned how to pick up a pen. This is Reinforcement Learning in a nutshell, it’s a lot like how living creatures learn.


GraphPipe is a protocol and collection of software designed to simplify machine learning model deployment and decouple it from framework-specific model implementations.

What is AI Alignment?

In an earlier blog, I discussed the control problem. The challenge we face controlling a machine which thinks in an entirely different way to us and may well be far more intelligent than we are. Even if we have a perfect solution to the control problem we are left with a second problem, what should we ask the AI to do, think and value? This problem is AI alignment.

A gentle introduction to OCR

OCR, or optical character recognition, is one of the earliest addressed computer vision tasks, since in some aspects it does not require deep learning. Therefore there were different OCR implementations even before the deep learning boom in 2012, and some even dated back to 1914 (!). This makes many people think the OCR challenge is ‘solved’, it is no longer challenging. Another belief which comes from similar sources is that OCR does not require deep learning, or in other words, using deep learning for OCR is an overkill. Anyone who practices computer vision, or machine learning in general, knows that there is no such thing as a solved task, and this case is not different. On the contrary, OCR yields very-good results only on very specific use cases, but in general, it is still considered as challenging. Additionally, it’s true there are good solutions for certain OCR tasks that do not require deep learning. However, to really step forward towards better, more general solutions, deep learning will be mandatory.

‘Logit’ of Logistic Regression; Understanding the Fundamentals

At the very beginning of my journey to learn fundamentals of machine learning, I remember spending a lot of time in clearly understanding the basics of logistic regression. Hopefully this meditation will leave you with more answers and correct concepts than confusions related with logistic regression.

Finding Similar Quora Questions with BOW, TFIDF and Random Forest

In September 2018, Quora reported hitting 300 million monthly users. With over 300 million people visit Quora every month, it’s no surprise that many people ask duplicated questions, that is, questions that have same intent. For example, questions like ‘How can I be a good geologist?’ and ‘What should I do to be a great geologist?’ are duplicate questions because they all have the same intent and should be answered once and once only.

Learning neural network architectures

Last time we talked about the limits of learning and how eliminating the need for design of neural network architecture will lead to better results and use of deep neural networks. Here we will analyze all recent approaches to learning neural network architectures and reveal pros and cons. This post will evolve over time, so be sure to log in from time to time to get the updates.