It can be tempting to spend 2-3 weeks and put together a PoC (proof of concept): a model that can take some of the data your company has and generate some potentially useful predictions. There are millions of tutorials showing how to achieve this in a few lines of code using a few standard libraries, and it's tempting to think that producing your model will be easy. this will not happen. While many might think that building the first model gets them 80% of the way to a machine learning solution, those who have built and produced machine learning solutions before will know that PoC is closer to 10% of the journey.
All machine learning solutions can be divided into three concepts:
Data: Used to find and model existing patterns;
Code: used to define and serve the model, and integrate it with other services;
Model: Used to generate predictions.
A production solution also has many more moving parts. Proof of concept often involves building a simple model and verifying whether it can produce predictions that pass a quick sanity-check. Everything can be done on a single machine. In contrast, this is only the first part of the production workflow. At the production level, you will need a good training server and a good process to keep track of the different models. You'll need to test trained models, estimate scale, and monitor everything to make sure it all holds up before integrating them with your existing production services. In the end, you will iterate over this process several times, as you can improve the data, code, or model components.
Read More : Tensor2Tensor To Accelerate Training Of Complex Machine Learning Models
What kind of team do you need to design a machine learning solution?
When you're building a machine learning team, it's tempting to hire researchers from an academic background. If they invented new machine learning algorithms, they must be the best people to use them, right? not always.
Hiring a researcher to build you a machine learning solution is often like hiring someone who designs KitchenAid equipment for you to cook. While researchers come up with new ways to solve problems and have deep theoretical expertise, they often lack much practical expertise in using existing tools, following good engineering practices, and having a hard time making/quality tradeoffs Is.
More often than not, you want someone who can use existing machine learning algorithms and tools—much like a chef to whip together delicious meals using standard ingredients and kitchen equipment—rather than someone from scratch. To design new equipment from.
Machine learning algorithms are not ultimately a competitive advantage. This is your data. This is why companies like Google and Facebook are happy to open-source their algorithms, but keep their data secret. You'll often find researchers (and salespeople) talking enthusiastically about the latest changes to the neural network algorithm, but in reality the algorithm isn't all that important. Your success is usually determined by how well you choose the right problem, what data you include, and how well you prepare your data.
Read More : Top 10 Highly Recommended Machine Learning Software!
The terminology still isn't completely consistent, but you'll typically be looking for at least one machine learning engineer - someone who specializes in building machine learning solutions for business.
0 Comments