The tech giant plans to speed up ML proficiency by publicizing its long-internal material
The What and Why
Amazon has long been striving to fix the issue of excess demand (vs supply) of individuals who have proficiency across the fields both Machine Learning and Software Engineering. To date, they have developed sloths of internal resources to get employees up to speed on the essentials. This is typically referred to as OJT, for “on the job training.”
OJT only goes so far — the size of your workforce. Aside from hired workers, companies depend on the education system to routinely supply capable talent to the workforce. This system has performed sufficiently for hundreds of years. However, the tide is turning. The speed of machine learning’s integration into industry workflows has largely outpaced the education system’s ability to provide fully-equipped talent. This is partially due to large systems necessarily moving slowly, but also due to a lack of convergence of dominant algorithms and tools in the field. Education systems are basically faced with a choice between overfitting on current trends versus sticking with classical techniques and allowing for OJT to solve the last-mile problem.
Amazon has a great idea — meet halfway.
Academic institutions will largely lean towards proven classical techniques for education, and that is the correct move. To help the last-mile OJT problem even more than post-hire education, Amazon is now making available course materials from their internal “ML University”. By doing this, they will be able to educate many eventual employees even before it is interview time. This helps both sides of the table. Prospective employees can learn much more relevant material ahead of job applications and feel more equipped in job selection and commitment. On the flip side, Amazon and similar companies can then judge talent more directly in interviews than they have been able to. Since so much learning material is publicly available, there is less room for “the benefit of the doubt” when an applicant does not have experience in a certain sub-area.
Just three courses have been released for immediate use: natural language, computer vision, and tabular data. However, more will be rolling out through the end of 2020, with the start of 2021 having all the material public.
“By going public with the classes, we are contributing to the scientific community on the topic of machine learning, and making machine learning more democratic,” Werness adds. “This field isn’t limited to individuals with advanced science degrees, or technical backgrounds. This initiative to bring our courseware online represents a step toward lowering barriers for software developers, students and other builders who want to get started with practical machine learning.”
– Amazon Science
Check out the intro to the “Accelerated Computer Vision” course below. The entire course is available on similar Youtube pages.Introduction for the Computer Vision course of Amazon’s ML University
Opinions and Cautions
This is great for the democratization of machine learning in the industry. Academic has long been very open and cooperative with ML research. The same can be said for the open-source software movement. Recently, in the past decade or so, we have seen these ideologies extend into the ML industry space. Its continuation will ensure that the economy’s aggregate output will rise, while still fostering healthy competition.
I’ll add a word of caution, however. The phenomenon referred to as “vendor lock-in” occurs when a service provider produces so much incentive to continue acquiring its own products across its ecosystem that the consumer effectively becomes stuck buying the provider’s goods and services, lest he/she suffer either lackluster integrations or the switching cost of starting over with a new provider. Look no further than a comparison of Apple vs Microsoft vs Google products for examples of vendor lock-in at work.
The courses at ML University indeed appear at the outset to provide a lot of general applicability across the ML and software space. It is likely that 80–90% of all of its material will do so, which is great!
However, as you go through the courses, remain keen on staying up-to-date on how other providers are accomplishing similar products and services. To be a truly marketable ML practitioner in this evolving workforce, one must stay flexible in showing ML proficiency independent of algorithm, language, framework, and platform provider.