Hacker Newsnew | past | comments | ask | show | jobs | submit | congerous's commentslogin

TensorFlow: everything to all people.

Eager is actually not as innocent as "open-source projects borrowing the best parts from each other", as some commenters here suggest.

Google is attempting to dominate the machine-learning API and the Python ecosystem for scientific computing.

The company that controls the API influences which apps are built on it and how. Think about how Google bundled Android services on top of Android, and how that posed an existential threat to other companies. That's what's coming for TensorFlow. Many developers are too naive to realize it, or too short-sighted to care.


Huh? They're attempting to dominate the machine learning ecosystem by writing a bunch of free and high quality machine learning libraries? What exactly are they doing wrong?

I wouldn't compare a permissively licensed library to Android services at all.


I'm surprised I have to write this, but Google is not a charity. They are pouring commercial resources into Tensorflow for a reason. That reason is Google Cloud. Tensorflow is a Trojan horse to get people to use Google Cloud and other paid Google products. How do I know this? Because Tensorflow works better on Google Cloud than anywhere else, and Google is making a concerted effort to catch up with AWS in cloud, mostly through machine learning.

I didn't compare Tensorflow to Android services. I said that Tensorflow would serve as the basis of a service bundle, much like Android did. Let's come back in a couple years and I'll tell you I told you so.


> I'm surprised I have to write this,

Insulting the reader

> but Google is not a charity

truism

> They are pouring commercial resources...

As opposed to "non-commercial resources"?

> ... for a reason.

Everything happens for a reason.

> That reason is Google Cloud.

> How do I know this?

Pray tell!

> Because Tensorflow works better on Google Cloud than anywhere else.

This is the only real argument in this conspiracy. And if "anywhere" includes the users' hardware, it's wrong: tensorflow runs flawlessly on any Linux/NVIDIA hardware. Maybe it works better with GCE than AWS, but that would once again fall into that "rather unsurprising" category of factoids.

> Google is making a concerted effort to catch up with AWS in cloud, mostly through machine learning.

This can be re-written as "Google has a cloud offering, which it tries to sell. And right now, machine learning is pretty hot". Throwing a "concerted effort" in there is just trying to jazz it up to something ominous. Which it isn't.

> I didn't compare Tensorflow to Android services. I said that Tensorflow would serve as the basis of a service bundle, much like Android did.

"The basis of a service bundle" actually doesn't sound that scary. Nobody is disputing that Google offers services build on tensorflow. It just isn't any sort of "Trojan horse" conspiracy, and it is somewhat limited by the fact the tensorflow is OSS licensed and could be forked by anybody people suddenly find out it's full of geek soldiers.


> truism

Maybe, be people in this thread treat Tensorflow's creation as an act of simple altruism.

> And if "anywhere" includes the users' hardware, it's wrong: tensorflow runs flawlessly on any Linux/NVIDIA hardware. Maybe it works better with GCE than AWS, but that would once again fall into that "rather unsurprising" category of factoids.

Sorry, Tensorflow is slow on GPUs compared to other frameworks. This is not just an early blip, its a consistent pattern that has been repeatedly demonstrated. Why is Tensorflow slow on commodity hardware? Why isn't Google with it's infinite resources making Tensorflow run as fast as other frameworks on GPUs? Because it needs to demonstrate an advantage on the Google Cloud with TPUs.

On that cloud, it surrounds Tensorflow with other functionality that makes it easy to build AI, which aren't part of the Tensorflow project. Tensorflow is hard and inefficient to serve for inference, for example.

Machine learning is Google cloud's only hope to salvage Diane Greene's efforts and extend their dominance to a new sector. They're running a distant fourth.

> actually doesn't sound that scary.

It sounds scary to a lot of companies that don't want to be controlled or destroyed by Google. But by all means, lend them a hand, geek soldier.


In what way is Tensorflow working better on Google Cloud? Are they tuning the ML code for specifics of their infrastructure or does Google Cloud just have more tooling for Tensorflow?


Disclaimer: I am notoriously anti google and have tons of reasons to post these links. We push on hybrid cloud/on prem deep learning with our own deep learning framework that competes with the commerical sides of tensorflow, mxnet,..

Sample of search results: https://cloud.google.com/ml-engine/ https://cloud.google.com/tpu/

Even their docs: https://cloud.google.com/tpu/

Marketing content/training: https://www.coursera.org/learn/serverless-machine-learning-g...

vs (1 link I found with googling) for AWS: https://aws.amazon.com/tensorflow/

If we push the amazon equivalent though, run this: site:amazon.com aws mxnet

Every cloud vendor has their own framework. Microsoft has CNTK on azure as well.

Google doesn't want a repeat of what happened with map reduce and hadoop: https://www.quora.com/What-is-the-relationship-between-MapRe...

That being said, as a user: Just take it. You benefit from vendors competing. Google would love to pay you to use their tools.


This article adds nothing new to the discussion about AI.

* We knew about Tencent's AI Lab (Saying you need your own capabilities is another way of saying: We have NIH and we will waste 3 years building our own framework from the bottom up).

* We knew about China's AI plan, which was announced in July. That was huge news at the time.

* We knew about Baidu, but Baidu is actually slipping. They lost Andrew Ng and Adam Coates, and frankly they have executed poorly on social media and cloud computing compared to Tencent and Ali. It looks like they'll lose their lead in AI.

And the article actually contradicts its own sub-headline: "The West shouldn’t fear China’s artificial-intelligence revolution. It should copy it." The West is leading in AI research, and China is copying the West. So what does it mean that the West should copy China? China is playing catch up in AI just like it has been playing catch up in GDP growth. Catching up always moves faster than blazing a trail. The West is blazing a trail now. They don't have much to learn from players trying to catch up.


>so what does it mean that the West should copy China?

I don't think you're addressing OPs point at all. They pointed out that the US currently does have the lead in the field due to its larger prosperity for the most part, but that China is rapidly developing more talent.

What we should copy is thus not the technical status quo, but our attitude towards education and talent acquisition in the relevant fields.Technical advantages are not self-perpetuating. If the West grows complacent and does not develop the talent necessary to advance the research, then that lead will reverse.


No. The US has the lead because its competitive advantage in the world economy is exporting innovation. That's what it's good at. The Chinese economy is outperforming the American one in many ways, but the US enjoys a more open society and economy, which encourage innovation. No one would ever say that the West should be complacent, but people are making a big deal out of the Chinese AI push, exaggerating it in a way, and that aligns with more widespread fear-mongering about China. In reality, corruption, chaos and censorship continue to hold China back more than is being reported.


Data scientists arguably have too much choice. 10 data scientists will have 50 different tools, can't share work or build on another's experiments or even remember what the result of an experiment were. those are some of the reasons why most data science projects fail. that and integrations. standardization has real benefits.


Of course standardization has benefits but how do you choose? Standardization only works if choice is eliminated so choice is a barrier to achieving standardization.


It often just comes down to project requirements. Eg, what kind of model is required? How hard would it be to build with tool x?

For example, a big reason why a lot of computer vision research was built (and sorta still is because of momentum) on caffe was pre existing model zoos.

A big reason why people choose TF (despite lacking dynamic graphs) is just because of existing community.

Requirements for both papers as well as industry will continue to evolve. Each framework will have their own trade offs.


Curious: Does anyone use GANs for anything of value, or are they just gimmick NNs?


google used them to train their go AI AFAIK.


KPI = amount of funds raised.


I've been there. A ton of good advice is on this thread. You need to try to influence your internal state chemically if you want to stay on track. Exercise, medication, meditation, eating and sleeping healthy are all helpful. It's also good to reconnect with people that matter to you: family, friends, etc. If you're closing the deal, and the money is significant, just think what you could do for them, or for other people who need help. And find ways to spend more time with them, because our lives are defined by our relationships. You could say your self exists to the extent that your in conversation with people who understand you. So go find with and be with them. It'll help you refuel.


The "leading" DL framework is playing catch-up to Chainer, PyTorch and DyNet. Another Google product development bungle.


The way I see it, TF is about to pull _way_ ahead thanks to XLA JIT/AOT. All of a sudden you get the ability to fuse things at a much more granular level, which could reduce memory bandwidth requirements by a lot. Frameworks like Torch can't do any fusing at all, since their computation is fully imperative. Tactical win for imperative frameworks, I suppose, but strategically functional graph is the way to go. DB people realized this in the 70s, ML people are realizing this now.


TF is way behind on UI, which is why it's making Keras its front-end. It's fairly slow on multi-GPUs compared to Torch and neon. It might pull ahead in performance on GCE, but that's just for lockin.


TF is in a fortunate position of having several UIs at this point. It's a lower level framework with a lot of power. If you don't need all that power, Keras or TFLearn or Slim are pretty great. If you do, it's there for you. I see no evidence that Google's goal with TF is to lock you into anything, and especially GCE. I'm a former Google employee, and I can tell you unequivocally — that's not how Google actually works.


I have to say I find this piece pretty disappointing. Slightly smug throughout, which is easy in retrospect, with approximately zero to add to the conversation. The thesis seems to be: The future is hard to predict, and even if you can, it's hard to get the timeline right. No shit. Is this what VCs do with their spare time? Remember granddad and contemplate uncertainty?


The assumptions we make, the more performance goes down in tasks like machine translation, and that includes the assumption that parts of speech matter to machines.


There's some baseless speculation in this thread about what's Singhal's departure meant for Google search. It's important to remember that Singhal was committed Google's knowledge graph, a painstaking, manually constructed graph of entities and relationships that surfaces the occasional answer (ask Google what the capital of Azerbaijan is and you will see a knowledge graph answer). Knowledge graphs are brittle, hard to extend, and hard to modify in the face of changing data. Google has tilted towards machine learning and neural nets for a lot of functions, and Singhal was not part of that tilt. But you could argue that machine learning and neural nets are producing better results, which counters the argument that Singhal leaving means that Google doesn't care about organic results any more. That's BS, frankly.


The person who replaced Amit as head of search (John Giannandrea) is the founder / CEO of Metaweb and creator of Freebase -- aka what became Google's Knowledge Graph. Google using more ML in search doesn't mean a KG isn't also useful for query understanding.

And KGs aren't mutually exclusive with ML -- e.g. https://arxiv.org/abs/1612.04844


>Google doesn't care about organic results any more. That's BS, frankly.

Do they care less than they used to, though? Their placement on the screen as opposed to KG, widgets, and ads..and how that's progressed over time, seems to imply it.


Google seems to have largely solved its biggest problem: spammy results. It's very hard to find sites in the top results for strong keywords with poor backlink profiles.

It has done this by a) Actually improving its spammy backlink detection, and b) Scary webmasters away from spammy SEO tactics through sudden updates like Panda/Penguin and straight up propaganda.

Once you solve the spam issue, Google's search results organically become better. Ergo, it doesn't need to spend that much time monitoring and improving them


That's interesting, and would explain why no replacement was named for Matt.

I wonder how much they actually improved the webspam thing.

The search results you see today, versus years past, suggest a sort of "cheat" to do that...something like a seed set of trusted sites (big brands, universities, government sites, etc) , and weighting based on distance from those seeds.

I say that because I'll often get results that are heavy on authority, but low on relevance. What I'm looking for is on pages 2 and 3, on more niche, less prominent sites. It didn't used to be that way.


> Do they care less than they used to, though?

That's impossible to measure, but the arguments you have employed regarding personnel changes do not demonstrate your point.


Okay...the top two publicly visible leaders in the space leaving in the same timeframe could mean something. Perhaps in this case it does not. I don't think either of us knows for sure. It does, however, pique my interest in what the group direction is now.

What of the decision to not replace Matt with a specific role or person? Is that not interesting?

And, again, I mentioned many other observations, not just Amit and Matt.

Edit: If the primary driver was lack of expertise in the ML direction, Uber seems an odd place to land, doesn't it?


Another explanation for their leaving is that the overall strategy around Search is changing and now outside of those two individuals' realm of expertise.

I work for Google but this is pure speculation on my part.


> the top two publicly visible leaders in the space leaving in the same timeframe could mean something

Again, Amit may not have been aligned with the direction they're taking search. That doesn't mean Google doesn't care. In fact, it could mean just the opposite.


Singhal's title at Uber says that he is in charge of marketplaces and maps, not the self-driving stuff. So this would be stuff like pricing, dispatch, supply optimization, and route optimization which rely on traditional algorithms and not ML. But I agree with your larger point and find his departure from Google curious. Even if they were moving search in another direction or he was completely bored with what he was working on and wanted change, I am shocked that they would let someone of his caliber leave Google.


Supposedly, they leverage machine learning across those areas as well...perhaps not as the primary foundation, but used there nonetheless.

http://venturebeat.com/2016/11/02/uber-is-rolling-out-a-big-...

https://newsroom.uber.com/inferring-uber-rider-destinations/

There's other references, but you get the idea.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: