Image for post
Image for post
Photo by Florian Olivo on Unsplash

Machine Learning Production Vs Machine Learning Research

Most people who are experts in Machine Learning gained their experience through academia: taking courses, doing research and reading research papers. Most companies nowadays use Machine Learning, although it is still relatively new compared to traditional software engineering. If you are one of the people who learned ML through academia, you might have a very different industry experience. ML Production is very different from ML Research. Let's talk about some of the differences

1. Objectives/Goals/Intention

Often, ML Research has one single objective: model performance and achieving state-of-the-art (SOTA) results. In contrast, there are many stakeholders involved in bringing an ML system into Production. Also, each stakeholder might have a different objective. …


Image for post
Image for post
Photo by chris panas on Unsplash

The main topic of the article is the privacy-transparency trade-off and how it affects a huge number of issues. This article walks through some of the most important challenges to society and identifies how the privacy-transparency trade-off underpins them. Improving information flows, by solving this trade-off, can help us in many areas like disinformation, scientific innovation, and even democracy itself.

Every part of the human experience is soaked in information flows. Since the beginning of human collaboration.

We share our medical information with our doctor. We share our location with an app to get directions. We share our heart rates and sleeping patterns in hopes of improving our well-being. Every day, we share personal information to exchange goods, receive services, and in general, to collaborate. …


Image for post
Image for post
Photo by Nick Chong on Unsplash

The Stock Market is a fascinating area that everyone wants to jump into, especially with the rise in Deep Learning techniques like Transformers and Attention Mechanism.

Why is the Stock Market Difficult to Predict?

There are too many factors to take into account which affect stock prices. It is really impossible to build a precise model which would rely on all of those factors, and one of the main reasons is that most of the factors are not known beforehand: even if some events affecting the stock market have happened in the past, you never know what else would happen in the future.

Think of a quite common situation: just one public statement from a high-calibre politician and a certain company looses a billion in its value in one day because the market reacts to that. All kinds of that sort of unpredictable thing could happen in the future: political moves, catastrophes, climate cataclysms, whatever else, generally known as ‘acts of God.’


Despite recent progress in artificial intelligence (AI) research, human children are still by far the most skilled learners we know of, learning noble skills like language and high-level thinking from minimal data. An efficient, hypothesis-driven exploration helps children’s learning. In fact, they explore so well that numerous ML researchers have been encouraged to put videos like the one below in their speeches to motivate study into exploration methods. However, because using results from studies in developmental psychology can be challenging, this video is often the degree to which such research correlates with human cognition.

Image for post
Image for post
A time-lapse of a baby playing with toys (https://www.youtube.com/watch?v=8vNxjwt2AqY)

Why is directly employing research from developmental psychology to problems in Artificial Intelligence so tricky? Taking motivation from developmental studies can be challenging because the environments that human children and RL agents are typically investigated in can be very diverse. Traditionally, reinforcement learning (RL) investigation takes place in grid-world-like environments or other 2-dimensional games, whereas children act in the rich and 3-dimensional real world. Moreover, similarities between children and AI agents are challenging because the tests are not regulated and often have an objective mismatch; much of the developmental psychology study with children takes place with children involved in the free exploration, whereas most AI research is goal-driven. Lastly, it can be difficult to ‘close the loop’ and build agents inspired by children and learn about human perception from AI research results. By examining children and artificial agents in the identical controlled, 3D environment, we can probably mitigate several of these difficulties above and ultimately advance research in AI and cognitive science. …


Image for post
Image for post
Photo by CDC on Unsplash

Before anything. I’d like to disclose that I’m nowhere near close to an epidemiologist or a doctor. I am a Master’s Student at the University of Toronto working on Deep Learning based Recommendation System. With that said, most of the studies, commentaries, and conclusions in this work are the results of my own experience with machine learning as well as some online articles and some very shallow research on Covid-19 dataset, but most come from basic math and stats knowledge.

These results are not the actual results. I am just posting this to encourage others. I truly believe we can always make better decisions when we have data and appropriate tools on our side. …


Image for post
Image for post

In this article, we will recreate the style transfer method that is outlined in the paper, Image Style Transfer Using Convolutional Neural Networks, in PyTorch.

Image style transfer is a technique that aims to render the content of one image with the style of another, which is essential and exciting for both practical and scientific reasons. The style transfer techniques are widely used in image processing applications such as mobile camera filters and creative image generation.

In the article, we are going to use the pre-trained 19 layer VGG (Visual Geometry Group) Network to achieve style transfer task. VGG network consists of a series of convolutional, pooling, and fully connected layers. In the below image, we name the convolutional layer by stack and the order of the stack. For example, conv_1_1 represents the first convolutional layer in the first stack; conv_2_1 represents the first convolutional layer in the second stack. …


Image for post
Image for post

“The most important thing is to try and inspire people so that they can be great in whatever they want to do”

— Kobe Bryant.

Well, what to say, Kobe is my hero! The Black Mamba is my idol, the man that inspired me for everything in life on and off the Basketball court. He had such a great impact on my life. I am heartbroken and devastated on hearing the news that he and his little baby girl Gianna Bryant (Gigi), died on a helicopter crash. I still remember as a high school student, I used to watch Kobe’s game and get excited and mesmerized with his fade-away buzzer beat shots, game-winning dunks, etc. I watched him play basketball, and that encouraged me to play basketball as well. …


Image for post
Image for post

Natural Language Model

Language modeling is fundamental to major natural language processing tasks. Lately, deep-learning-based language models have shown better results than traditional methods. These models are also a part of more challenging tasks like speech recognition and machine translation. Language modeling is generally built using neural networks, so it often called Neural Language Modeling (NLM). Originally, feedforward neural network models were applied to introduce this language modeling procedure. …


Image for post
Image for post
Photo by Alina Grubnyak on Unsplash

For humans, vision feels so easy since we do it all day long without thinking about it. But if we think about just how hard the problem is, and how amazing it is that we can see. To see the world, we have to deal with all sorts of “nuisance” factors, such as a change in pose or lighting. Amazingly, the human visual system does this all so seamlessly that we don’t even have to think about it. Computer Vision is a very active field of research, which tries to help the machines to see the world as humans do. This field made tremendous progress in the last decade because of modern Deep Learning techniques and the availability of a large set of images online. …

About

Karthik Bhaskar

Vector Institute | Deep Learning | MASc ECE — University of Toronto

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store