In before the predictable lame Google data slurp comment
Tensorflow Lite: Neat, but an ordeal to get running on your mobe
Late last year, Google announced a new mobile library for Tensorflow on mobile devices: Lite. If you didn't know by now, Tensorflow is Google's framework for developing machine-learning models on a large scale and is proving popular with the community – even if many developers choose to use a higher level abstraction on top of …
COMMENTS
-
Wednesday 31st January 2018 11:41 GMT Anonymous Coward
"suddenly you are deep in the field of machine learning – not an easy subject to just pick up"
You can't just pick it up - its an entire field of knowledge, not just a new API. If you really want to understand it you'll need to do a course either through a book or an actual one. Sure, you could bang together something that vaguly works using a *very* high level library but you wouldn't have a clue how it was actually doing it and so wouldn't have a clue what to do if it didn't work properly.
I'm in no way an ML expert , but I've learnt enough to know I don't know nearly enough to do justice to any ML program I tried to write.
-
Wednesday 31st January 2018 23:36 GMT Anonymous Coward
Compared to where I started in AI/ML in comp. sci. in the early 70's? Now the field is quite similar in nature to picking up BI skills. To do your best you need some type of map to guide you to the right signposts along the way. Getting it via coursework helps immensely simply due to the interactive nature of the course. Book knowledge, not so great. I'd put the various multimedia tutorials in between the two. Gifted amateurs can do pretty well even if they lack specific maths skills.
As I stated above I started working in this field back in the 70's and despite, or perhaps because of, my dozen years working in various engineering fields, I've been using ML quite a bit along the way. There were always problems that required a different approach to working with the data sets that were beyond existing methodologies in statistics, other types (long list) of analysis (long list), etc. "Wiring-Up" a neural net was useful, even if the mainframe or later microcomputers, took a good loooong while to complete its training datasets. I even came up with a couple that I've never seen anyone approach, ever. Self-allocating/wiring neural nets was one; the other was non-linear neural nets. [I do like to take a walk on the wild side from time to time.]
So, I'm not depending on my university work before or after my wandering off to play (nuclear) engineer. It was my time spent in economics, specifically econometrics, and taking every damn statistical/mathematical analysis, experimental design and/or modeling courses the university had, no matter what the department, that allowed me to rip up some of the rules and try something different. And this is one point I'd like to really make clear. We do need people that are also ready to take a walk on that wild side as the whole field, and all the domains that ML can be utilized in, as this universe hasn't been entirely mapped.
Yes, it's not down to pure API plays but it's also not something you need eight plus years, or a lifetime, in maths to do. It's also quite fun seeing what you can do. [And in the various engineering fields I've worked professionaly in over the years, let's say the laws were suggestive rather than concrete.]
-