The kernel trick can be used with any algorithm from the broad class of algorithms known as kernel machines. The most popular kernel machines are the support vector machine and logistic regression. Essentially, the optimisation objective of a generic kernel machine is formulated in such a way that it depends only on dot products between input vectors. This allows us to swap these dot products with a kernel computation of the dot product into some higher-dimensional (possibly infinite dimensional) space. The key to a kernel function is that it MUST have the following property: K(x_i x_j) g(x_i), g(x_j) for some g.

I've created tutorial that shows how to create web service in Python and Django to serve multiple Machine Learning models. There can be several ML models available at the same endpoint with different versions. What is more, there can be many endpoint addresses defined.

Excited to launch a new machine learning competition! The goal is to be able to a better job creating disaster response plans based on detailed maps of communities. In order to do this, we need to understand the risk to structures, which we can do by understanding what kind of roof a building has. Come use your machine learning skills for a good cause! Plus it's got interesting geo data, novel imagery, and the opportunity to develop new methods.

Is reinforcement learning practical at this point for industry work? The most prominent examples we see are from DeepMind (AlphaStar, AlphaGo), but the team are world-class researchers (over 40 of them) who also worked closely with expert Starcraft 2 players with a ton of computing resources. As someone who hasn't had much experience in RL, I see potential applications but am unsure of the amount of work or practicality of it. For example, one potential application for RL is to learn fraudulent behavior in an online retailer system (i.e. Amazon, EBay) and proactively find methods of fraud before they happen.

Microsoft and Nokia collaborate to accelerate digital transformation and Industry 4.0 for communications service providers and enterprises Companies announce their first joint solutions combining Microsoft cloud, AI and machine learning expertise with Nokia's leadership across mission-critical networking and communications REDMOND, Wash., and ESPOO, Finland -- Nov. 5, 2019 -- Microsoft and Nokia today announced a strategic collaboration to accelerate transformation and innovation across industries with cloud, Artificial Intelligence (AI) and Internet of Things (IoT). By bringing together Microsoft cloud solutions and Nokia's expertise in mission-critical networking, the companies are uniquely positioned to help enterprises and communications service providers (CSPs) transform their businesses. As Microsoft's Azure, Azure IoT, Azure AI and Machine Learning solutions combine with Nokia's LTE/5G-ready private wireless solutions, IP, SD-WAN, and IoT connectivity offerings, the companies will drive industrial digitalization and automation across enterprises, and enable CSPs to offer new services to enterprise customers. BT is the first global communications service provider to offer its enterprise customers a managed service that integrates Microsoft Azure cloud and Nokia SD-WAN solutions. BT customers can access this through a customer automated delegated rights service, which enables BT to manage both the customer Azure vWAN and the unique Agile Connect SD-WAN, based on Nokia's Nuage SD-WAN 2.0.

Then left, another left, then left again. That's the typical movement your thumb might go through if you're trying to find "the one" on any dating app. In some circles, it's becoming more common to hear friends say they met their significant other via dating apps Tinder, Bumble or Hinge, but what about Spotify? For two emerging artists, the music-streaming service helped them connect, and fast forward, they're getting married. The lovebirds are trying to find out who's responsible.

I previous posted the write-up on the first part of my project (Github repo) to predict how skaters would rank in the figure skating world championship from earlier scores that they earned in the season. The main idea is to separate the skater effect, the intrinsic ability of each skater, from the event effect, the influence of an event on a skater's performance, so that a more accurate ranking could be built. Unfortunately, this model does not have a closed-form solution to learn the parameters as opposed to the earlier models. Therefore, gradient descent was used to learn them, which resulted in this neat little animation that tracks how the model residuals, RMSE, as well as predicted ranking gets better and better as gradient descent runs. I also explore different strategies to reduce model overfit (so that it can predict skater ranking more accurately), using familiar methods such as model penalization and early stopping. Lastly, note that this hybrid model is nothing but factorizing the event-skater score matrix into an event-specific vector and skater-specific vector, which can multiply together to approximate the score matrix.

I still remember I used to feel uncomfortable when looking at those complicated mathematical concepts needed for machine learning.