Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google launches an end-to-end AI platform

techcrunch.com

Post reply on HN
Print view View original

41–50 of 132 posts

Re: Google launches an end-to-end AI platform

#41

Post by perturbation »

silasdavis wrote: post #38

Paraphrased from the article: AutoML tables takes a generic table and predicts can predict a columns value (for an unseen partial row, I assume). Their product page emphasises how easy it is: for developers with limited machine learning experience. What are the consequences of using such structured data predictions devoid of interpretation or quantified uncertainty? Presumably such predictions could be from a set of…

AutoML is essentially training a ML model using some heuristics or optimization algorithm to select model architecture and train a model. Feature engineering / feature synthesis as well as interpretability remain open challenges.

If I'm understanding your questions correctly, the main problems I see with this are:

- Using raw data instead of feature engineering (less of a problem given feature synthesis libraries like https://www.featuretools.com/ and other heuristic methods). I'd expect Google to do a good job of basic things like normalization of raw input features before training.

- Using features that it really shouldn't (if you just throw ML at your database for say, loan applications, then sensitive / personally identifying information can/will be used as features)

- Lack of insight / understanding as to what is driving the model. This can be partially overcome with post-training methods like LIME, Shapley values, etc.

I wouldn't expect predictions to be from a set of discrete values - if (say) predicting housing values and training a NN, the output should be continuous and based on the input features.

Re: Google launches an end-to-end AI platform

#42

Post by RosanaAnaDana »

streetcat1 wrote: post #36
pplonski86 wrote: post #21

Earlier quoted context omitted.

I'm running a startup which offers the same solution as Google AuotML Tables. Recently I decided to go open source. I will need to compare my solution with Google AutoML Tables (compare in terms of final model accuracy). But anyway I think many times the best model accuracy is not the most important in ML solutions. Any ideas what can I do with such a situation with my solution? Can I compete with Google?

I am too building a new automl platform on top of kubernetes, I do not think that you can compete with google on alg accuracy, since most of the underlying ML alg are open source (scikit learn or tensor flow). This is not a secret sauce. That said, to get better models, you will probably need to find better hyper parameters tuning method, which depends on the number of models that you are willing to run per model tun…

Yeah I don't think I agree with the assessment that better models are a function of finding better hyper parameters (entirely). Like, in my limited experienced,stacking and packing models (the 'lift' required by individual models), adjusting ones thinking regarding 'how' they go from data-in >> product out and 'what' it is were trying to predict, pipe-lining annotation and scaling its distribution, are far more effective at generating better ML models than tuning hyper parameters.

Generally, if a model is not performing or generating unexpected results, its almost always the data or how the question is being structured.

Re: Google launches an end-to-end AI platform

#43

Post by jimrandomh »

siavash wrote: post #35

Make sure you've read the service terms[1] if you plan on building apps for speakers, cars, TVs or smart watches... 12.1 The following terms apply only to current and future Google Cloud Platform Machine Learning Services specifically listed in the "Google Cloud Platform Machine Learning Services Group" category on the Google Cloud Platform Services Summary page: Customer will not, and will not allow third parties to…

I have a hard time imagining who this wouldn't be a dealbreaker for. These terms also mean you can't use it for open source, and you can't use it if you don't know what the ultimate application is going to be. And you probably can't resell technology you create, because no one who buys it is going to want that restriction, either.

Re: Google launches an end-to-end AI platform

#44

Post by holoduke »

pplonski86 wrote: post #21
petard wrote: post #14

This will make a bunch of startup's life really hard. I think it makes it harder to justify investing in your own ML pipeline or even building your own models for many use cases.

I'm running a startup which offers the same solution as Google AuotML Tables. Recently I decided to go open source. I will need to compare my solution with Google AutoML Tables (compare in terms of final model accuracy). But anyway I think many times the best model accuracy is not the most important in ML solutions. Any ideas what can I do with such a situation with my solution? Can I compete with Google?

Usually google is good in the initial release of a product, but they lack good customer care and support. They have non transparant pricings and they disrespect privacy. Those are something's you can differentiate on

Re: Google launches an end-to-end AI platform

#45

Post by _ptgt »

I work in building and deploying production ML/AI models but I'm having a lot of trouble cutting through the marketing jargon in this article and on Google's website as well.

Can someone explain what this does in engineering terms? How does this differ from something like AWS Sagemaker?

Re: Google launches an end-to-end AI platform

#46

Post by jedberg »

This is where Google really has a big lead on AWS -- the AI space. AWS has AI tools, but Google's are better and easier to use.

The big question is: If all your data sits in AWS, because your app that generates the data is there, do you reach across and try to use the Google AI tools, or are their tools compelling enough to get you to move your app and all your data to GCS?

Re: Google launches an end-to-end AI platform

#47

Post by filoleg »

pplonski86 wrote: post #21
petard wrote: post #14

This will make a bunch of startup's life really hard. I think it makes it harder to justify investing in your own ML pipeline or even building your own models for many use cases.

I'm running a startup which offers the same solution as Google AuotML Tables. Recently I decided to go open source. I will need to compare my solution with Google AutoML Tables (compare in terms of final model accuracy). But anyway I think many times the best model accuracy is not the most important in ML solutions. Any ideas what can I do with such a situation with my solution? Can I compete with Google?

Check this comment in the thread https://news.ycombinator.com/item?id=19627515 . If your solution allows for embedded devices and other use cases that the solution from Google prohibits (which is a lot of scenarios, actually), then your offering has a chance. Considering that it isn't a matter of choice at this point, with many companies legally not being able to use their offering, you will be doing quite fine.

Add on top of that better customer support, like other replies suggested, and your product won't be dying due to competition from Google any time soon.

Re: Google launches an end-to-end AI platform

#48

Post by mlthoughts2018 »

It’s so disingenuous for Google to brand these efforts as “democratizing AI.” It is precisely the opposite.

This is classic commoditization of your complement. On one hand, Google is pushing to centralize the integration, data management and computing platforms for machine learning, so that these things become as much of a commodity as possible.

On the other side, they are offering massive compensation packages or acqui-hiring as much AI talent as they can, not really because they have useful work for these folks to do, but to artificially reduce the supply of statistical algorithm talent, making their consulting and pre-packaged AI solutions go up in value in a manner that is pretty much the same as De Beers pushing silos full of diamonds to keep diamond prices artificially high.

This is very much the opposite of democratization, and my advice to anyone considering services like this or like Amazon’s out of the box models, don’t do it!

If you think it’s going to save you money over paying competitively to get your own in-house machine learning staff, you’re wrong, and you’re going to waste probably hundreds of thousands of dollars before you learn you’re wrong.

Re: Google launches an end-to-end AI platform

#49

Post by zeroxfe »

jimrandomh wrote: post #43
siavash wrote: post #35

Make sure you've read the service terms[1] if you plan on building apps for speakers, cars, TVs or smart watches... 12.1 The following terms apply only to current and future Google Cloud Platform Machine Learning Services specifically listed in the "Google Cloud Platform Machine Learning Services Group" category on the Google Cloud Platform Services Summary page: Customer will not, and will not allow third parties to…

I have a hard time imagining who this wouldn't be a dealbreaker for. These terms also mean you can't use it for open source, and you can't use it if you don't know what the ultimate application is going to be. And you probably can't resell technology you create, because no one who buys it is going to want that restriction, either.

It's not a dealbreaker for in-house ML applications particularly in the enterprise (banks, telcos, etc.), which is a huge market for cloud providers.

Re: Google launches an end-to-end AI platform

#50

Post by eanzenberg »

petard wrote: post #14

This will make a bunch of startup's life really hard. I think it makes it harder to justify investing in your own ML pipeline or even building your own models for many use cases.

Most people including ones here don't understand the value prop. In ML, the model (and modeling) is an afterthought that takes 3 lines of code. The value-add isn't the model or modeling, it's the feature generation and methodology. When (and if) this gets automated then yes, it will make life hard on many data scientists. But I don't see it happening within 20 years and most likely not within my lifetime.
Post reply on HN