Colab is phenomenal for beginning deep learning, but how does it stack up against an eGPU + Ultrabook?

Image for post
Image for post
Photo by Nana Dua on Unsplash

Deep learning is expensive. GPUs are an absolute given for even the simplest of tasks. For people who want the best on-demand processing power, a new computer will cost upwards of $1500 and borrowing the processing power with cloud computing services, when heavily utilized, can easily cost upwards of $100 each month. That is absolutely fine for businesses, but for the average individual, that adds up.

Because of this, 2 months ago, I made my first major purchase to give myself reasonable computing power. I already own an older XPS 15 with a small GPU (a GTX 960m that just was not cutting it), so I decided to buy a Razer core + NVIDIA GTX 1080, and I immediately had around 4x the processing power I had before, for under $450 (I bought them both used). …


Using Machine Learning to generalize the effects of Covid-19 on the lungs

Image for post
Image for post
Photo by CDC on Unsplash

If you have tuned into the news this past year, the leading cause of death among Covid-19 patients is related to the inability to breathe through fluid filling the lungs. While this seems pretty intuitive, how do we know what a Covid-19 lung looks like? How does this compare to a healthy lung?

In order to accomplish this, we can turn to a class of algorithms known as autoencoders. Autoencoders are a type of neural network used to learn a representation (encoding) for a set of data. Once we learn this representation, we can reconstruct this into a viewable image (decoding). This allows us to ignore the noise in images and focus on the similarity between all of the images. …


Whether you are using it for an eGPU and machine learning, gaming, or for external monitors… Living without one may become painful.

Image for post
Image for post
Photo by Sabri Tuzcu on Unsplash

The Thunderbolt 3 was crowned as groundbreaking for computers. So much so that Apple decided to make it every single port on newer MacBooks (2016+). Other laptop manufacturers advertise it as an absolute premium on laptops, sometimes charging hundreds of dollars more for this premium slot. Many people at the time were frustrated with Apple’s abrupt switch, but have we ever asked why they made this change and never looked back?

The first laptop I bought with a Thunderbolt 3 was a Dell XPS 15, the 2016 model, one of the first that offered this new technology. However, to my surprise, this was not actually a Thunderbolt 3, which supports speeds up to 40 Gbps (Or 4x USB 3.1s), …


My story of building an algorithmic trader for $0, analyzing free APIs, datasets, and web scrapers. Part 2: Datasets and Web Scrapers

Image for post
Image for post
Photo by Stephen Dawson on Unsplash

Finding data for a news based trader is an exceptionally challenging task. Getting historical data going back more than a year is either a daunting challenge or one that will cost hundreds to thousands of dollars to purchase. In part 1, I analyzed the best news APIs for accessing data. If you have not read that article, I recommend first doing so here. To summarize the results of what I discovered about free APIs, they are excellent for providing real-time news for traders but fall exceptionally short in providing users with a backlog of data. In fact, I could not find data that went further back than a year. They are also limited in the number of requests a user can make in any individual month. Luckily, we can take steps to mitigate the drawbacks of APIs. …


My Story building an algorithmic trader for $0, analyzing free APIs, Datasets, and web scrapers. Part 1: APIs

Image for post
Image for post
Photo by M. B. M. on Unsplash

Algorithmic trading, using either news or stock signals, has blossomed in popularity over recent years. An entire industry has developed from giants like Bloomberg and Webhose, to thousands of smaller companies all vying to have the fastest, most accurate, and most expansive news coverage. The problem with most of these services, however, is they are targeting large firms, and therefore often cost hundreds to an individual, most likely pricing themselves out of most people’s algorithms, and at the very least eating away at potential returns. For this reason, I decided to try and piece together the most realistic solution for the at-home developers, and after exploring dozens of sources, I have narrowed my search down to the 10 most viable options. None of these 10 free sources provide a home run for algorithmic trading, but it's quite realistic to think that a combination of these sources can get most people to a great place! In part 1, I will be covering the first 5, which are all APIs! …


The surprisingly simple journey of transforming an Ultrabook for a fraction the cost of a new PC

A couple of years ago in early 2016, part of the way through my undergraduate at Georgia Institute of Technology, I was trying to get my computer science degree with a 5-year-old laptop with a failing hard drive. I needed to upgrade. The new Intel Skylake processors and the Dell XPS 15 9550 blew me away with how form-fitting and powerful an ultrabook could be, and all of that, with a dedicated NVIDIA mobile GPU, I figured I would be set for years.

And I was set. In fact, I am writing this story on that very same laptop, it has not slowed down a bit since the day I have gotten it. That was, until, I began to pursue my masters coursework in AI and machine learning. I was assigned a classic deep learning task that has been solved now but proves a great learning opportunity in computer vision, to train a CNN to solve CIFAR-10. CIFAR-10 is a dataset of 10 categories of low resolution (32x32) images, and the task is to train a model to correctly classify a new image to one of those 10 categories. It is widely used as teaching material because the low resolutions allow quick training. Or so I thought. A CNN with multiple layers began to take hours at best to a night at worst to train, and this was supposed to be a simple task. Suddenly, that dependable Skylake I7, now 3+ generations old, and those 2 gigabytes of dedicated GPU RAM began to show their signs of wear. …

About

Jeremy DiBattista

I am a graduate student in Machine Learning at Georgia Tech!

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