back to article Alibaba crafts AI chips, Facebook uses Bayesian magic to tweak code performance, and more

Good morning – here's some machine-learning bits and bytes to kick start your week. If this sort of tech is right up your street, then perhaps check out our AI conference, M3, in London, England. OK, on with the show... Alibaba to make its own AI chips: Jack Ma, CEO and chairman of Alibaba, the giant Chinese e-commerce …

  1. Joe W Silver badge

    Gah!

    Bayesian inference is neat etc., but it is not magic - whatever FB thinks, and I would not really call it machine learning any more than I would call linear regression machine learning. It's just that the number crunching for Bayesian inference has become feasible for complex-ish models with current hardware levels. After all, much of what happens in your favourite algorithm[1] (Gibbs sampler or whatever) is linear algebra, and graphic cards are good in that.

    [1] assuming you have a model where you cannot write down the posterior PDFs analytically, or you are just a lazy sod (like me...)

    1. Michael Wojcik Silver badge

      Re: Gah!

      I take it you didn't read the blog post, much less the paper.

      This isn't about using Bayesian inference for machine learning.1 It's about using Bayesian optimization for tuning ML system parameters, typically the parameters used in the interconnects between component models (NNs, etc).

      Bayesian optimization is an application of the Bayesian interpretation of probability and Bayes's Theorem, inasmuch as it computes posterior probabilities based on priors. But it is not Bayesian inference, at least not in the sense in which that term is normally used.

      1And, frankly, it doesn't matter what you'd call it. Research and industry use the term "machine learning". That ship has sailed. I think it's completely appropriate (unlike the unfortunately largely meaningless "artificial intelligence", which depends on the cipher "intelligence" and raises sticky questions about artifice as well), to be honest. ML systems develop progressively more accurate models. That's a reasonable, if very constrained, definition for "learning", and it's being done by machines.

  2. David Roberts
    FAIL

    Smoke and mirrors?

    Tried the "tell me everything" service.

    First thing you have to do is register. I got an email back almost immediately wanting me to talk to them soon as to aid my recruitment process.

    Then it starts the "analysis", you get some cutesy "working on it" messages followed by an "unable to contact our server" result.

    Not selling themselves well at all.

    1. BitCoward

      Re: Smoke and mirrors? It's an MNP

      Minimum Non-viable Product

      Fail whale ahoy!

    2. Michael Wojcik Silver badge

      Re: Smoke and mirrors?

      Seems it's already slashdotted - the server (after allowing scripts from a zillion domains) now says that due to load they've had to pull the free service.

      Given the description in the article, I suspect any CS grad student who cared could have put something similar together. Take the inputs, use various web APIs to scrape a bunch of data, run it through some stock sentiment-analysis and LSA models and the like, and summarize the results mapped through some sort of pop-psych bullshit like Myers–Briggs.

      Maybe the site does something more sophisticated, but really, the claims as represented by the Reg could be satisfied (in a minimal sense) by some toy code.

  3. Anonymous Coward
    Anonymous Coward

    'What does your social media profile say about you?: Deepsense is an online tool that allows users'

    'You can try it here'...

    Oh for fck sake, more profiling scraping surveillance.... The following 'mission-creep' was in the news over the weekend before it was hidden away / buried. How long before the two are merged rightly or wrongly:

    ______________

    https://www.bbc.co.uk/news/technology-45590293

    ______________

    "fitness tracking to all policies - One of the largest life insurance providers will no longer offer policies that do not include digital fitness tracking. John Hancock will now sell only "interactive" policies that collect health data through wearable devices such as a smartwatch. Policyholders can earn discounts and rewards such as gift cards for hitting exercise targets.

    "But critics said the announcement was "creepy" and "dystopian". - Activity-tracking devices such as the Apple Watch and Fitbit can record how much exercise somebody is doing and can be used to log dietary choices.

    "But privacy advocates have warned that insurers could use tracking data to punish customers who fail to meet targets. "Naturally the American dystopian surveillance state will combine insurance with fat-shaming. Welcome to hell."

    1. Anonymous Coward
      Anonymous Coward

      In Soviet America, Hancock screws with you

      Yes. JH is a customer of my employer, and I sometimes work with some of their people directly, so I've been reluctant to comment on this story. But the new policy is vile. And I say that as someone who could almost certainly benefit personally - at least in a financial sense - from one of these "interactive" policies.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like

Biting the hand that feeds IT © 1998–2022