Paperspace Blog

Our latest guide details setting up Arize AI's ML observability platform with Gradient Deployments. Follow this guide to learn how to integrate Arize within Gradient Deployments to monitor data drift, traffic, and other model monitoring metrics. http://blog.paperspace.com/monitoring-deployments-with-arize-ai/

Having worked with YOLO, I really recommend this intro: https://blog.paperspace.com/how-to-implement-a-yolo-object-detector-in… And in general, YOLO is performant and at the same time, it has a simpler architecture than the Fast(er) R-CNN family. And in general, due to its head, it is WAY more readable in PyTorch than in TensorFlow; to the point, I use it as an example in Keras vs PyTorch example https://deepsense.ai/keras-or-pytorch/ (was here at some point).

Looks like Paperspace announced Volta support yesterday: https://blog.paperspace.com/tesla-v100-available-today/ One nice thing here is you can do monthly plans instead of reserved on AWS which is a minimum $8-17k upfront. Really great to see the cloud providers adopting modern GPUs.

Looking at today, the Tensor-TFLOPS is the unit of measurement that Nvidia is trying to push as the relevant benchmark for hardware performance with ML models. https://blog.paperspace.com/understanding-tensor-cores/amp/

Paperspace offers such a service, though I've only ever used it with a Windows server and Linux client: https://blog.paperspace.com/setting-up-your-cloud-gaming-rig-with-pape… I haven't checked recently, but the blog post claims it works with all games.

Transfer learning to emulate artistic style [0] has been around for the better part of a decade and has had zero impact on artists’ livelihoods. People consume art because they enjoy admiring the human talent that creates it, celebrating that some individuals are capable of extraordinary feats the vast majority of people are incapable of. It’s the same reason people watch sports—they enjoy admiring the top echelon of human physical ability. Very few people would watch Olympic Games performed by realistic androids. I do agree that tools like this could eliminate mediocre graphic designers, or anyone else creating visual products that are so mundane that their viewers never bother to consider the artist. Corporate Memphis [1] designers’ days are numbered. [0] https://blog.paperspace.com/art-style-transfer-neural-networks/ [1] https://en.m.wikipedia.org/wiki/Corporate_Memphis

This tutorial examines the DALL-E family of image generation frameworks from OpenAI -https://blog.paperspace.com/dall-e-image-generator/

Dillon from Paperspace here. If anyone is interested, this is actually part of a longer multi-part series covering GauGAN in-depth. Here's the link to the full series: https://blog.paperspace.com/tag/series-gaugan/

Paperspace also announced NVidia Volta VMs available today. As well as a contest for 100 compute hours credit ;) https://blog.paperspace.com/tesla-v100-available-today/

> the only way to fix them... This kind of claim is almost always suspect. There was a time that rotating or other manipulations caused issues for the models and those were largely taimed. Hinton's work with Capsule Networks [1] produced much more resilient models, though they are computationally much more intensive. It's also interesting that this is coming from Japan, where they also have a law regarding copyright and training that is very permissive. [2] > That said, it's kind of bonkers to me that publishing an image with alterations like this might be considered illegal, particularly if an image is being published by the copyright holder. It's like saying you can't watermark anything you publish. This should be generally true, but in law, intent typically matters. If you are doing something with the primary reason being to cause harm, then normal protections often don't apply. [1] https://blog.paperspace.com/capsule-networks/ [2] https://finance.yahoo.com/news/ai-art-wars-japan-says-185350499.html

What makes a thing funny? At its core, humor seems to be all about incongruity. Comic situations set up a context where something is marked or out of place. This oddness, far different to what we were led to expect or what we blithely assume is normal, is what makes things funny. [...] Dad jokes play with incongruity largely through linguistics and wordplay, rather than subject matter. The much-maligned pun is a mainstay of the dad joke. Puns, bad or good, have long fascinated researchers for their playful ability to tell a micro mystery, with its red herring clues in plain sight. . . . Through a trick of linguistics, words cleverly disguised like other words because of the way they sound or their different semantic senses can lead us in the wrong direction of meaning resolution, before we “get it.” After recently making a dad-joke ("Why does the string section have so much security? Because of all the viol[ins|ence].") I started a project to map the CMUdict phonemes [0] by Levenshtein distance in order to gather more word sets that are not precisely homophones for more jokes. The "Dubious Art" article was a top read from my literature review. 0. https://github.com/cmusphinx/cmudict 1. https://blog.paperspace.com/measuring-text-similarity-using-levenshtei…

Wow LFE is really cool, thanks for showing me another new thing! Admittedly I got pretty depressed about the lack of progress on this stuff and kinda checked out over the years. So my knowledge has gaps since the 90s. - I'll use this as an excuse to talk about another learning model besides NNs and GAs: Boltzmann Machines (BMs). For anyone unfamiliar with them, they can be thought of as a temperature minimization or gas diffusion model. Rather than gradient descent, simpler formulas like summations and random sampling can be used, so they are "big dumb algorithms" like GAs as opposed to "fickle algorithms" like NNs. They borrow concepts from Markov chains and simulated annealing. I'm having trouble finding a single approachable summary though: https://en.wikipedia.org/wiki/Boltzmann_machine https://www.geeksforgeeks.org/types-of-boltzmann-machines/ https://medium.com/@soumallya160/a-complete-guide-to-boltzmann-machine… https://blog.paperspace.com/beginners-guide-to-boltzmann-machines-pyto… https://deepai.org/machine-learning-glossary-and-terms/boltzmann-machi… The main problem is that their fully-connected nature makes them difficult to train beyond a certain size. Restricted Boltzmann Machines (RBMs) try to overcome that, but their increased complexity makes them more like NNs. I think that GAs could be useful for training fully-connected BMs by using evolution to decide which weights to update, similarly to your original question of…