I’m not sure if you follow tech conferences, but earlier this week Google held its annual Google I/O event. This is a conference targeting developers focused on Google ecosystem, during which there’s usually a lot of interesting announcements concerning Google’s plans and research. I feel that last year conference was much more exciting (“Now on tap”, Machine learning based search in pictures application and so on), still I found one pretty interesting news from the field of AI and Machine Learning.
I’m sure that you heard that earlier this year Google open-sourced their Tensor Flow engine for Machine Learning applications. So in this article [1] Google unveils a custom made ASIC chip called Tensor Processing Unit (TPU). TPU is tailored to Machine Learning computations. Google is not (yet?) giving away the exact numbers but they are stating that TPU is an order of magnitude better optimised in terms of performance per watt.
So Google is obviously spearheading the Machine Learning and AI research in general.
- Do you feel that they have any reasonable competition right now?
- Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
- Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
Publication of articles for discussion on the weekend just before closing deadline is completely unacceptable and unprofessional. I am not going to participate in such behavior and I will ignore your post.
ReplyDeleteI suggest everybody else follow my steps.
Sorry... :)
DeleteUnfortunately I had a really busy week at work and I couldn't find enough time to prepare the article for discussion earlier. Taking into account that usually most people participate in the discussion during weekends, I've decided to post it. I guess it's better later than never ;), and I think that the topic might be of interest to many people on the course.
DeleteOf course I understand your disapproval. I wish you a nice weekend.
@Kinga sorry, but I understand @Krzysztof ... overloaded human resources are nowadays standard additionally I thought that the week 5 will start tomorrow ;), but yours comment has won "the best konwersatorium comment award" signed by me.
Delete@Krzysztof - but the way: very interesting topic ;)
Hi, I remember a comparison of Google's and Stanford's Machine Learning Centers I saw some time ago. Google used CPUs and Stanford GPUs. I cannot remember exact numbers but with pretty comparable computing abilities CPUs used a way lot more energy.
ReplyDeleteUnfortunately the article does't describe TPUs n detail, I wonder what are its computing capabilities.
I think that you can call Baidu a Google's Chinese competition. Andrew Ng (lecturer at Stanford and founder of Coursera) is Chief Scientist at Baidu's Research Lab.
Companies like Google hire many academic researcher. Ng as I already mentioned works for Baidu, Hinton works part-time for Google. I think that only cooperation between companies and researchers can lead to big improvements.
I did a Coursers class on Machine Learning lectured by Andrew Ng. I wasn't aware that he's the founder of Courser and also works for Baidu. Pretty nice info, thanks for that.
DeleteI actually feel that privately held companies are sucking the talent from academia and motivating scientists to focus on bringing in more revenue with their research. Usually this research is conducted behind closed door. But I agree with you that cooperation between Academia and private companies can be really beneficial, provided it's done in the open.
I have also took Andrew Ng ML classes at Coursera :) I enjoyed them very much. I think the guy is real expert in the field of ML and pretty good teacher as well. I recommend his course to anyone interested in the topic.
Delete1. Do you feel that they have any reasonable competition right now?
ReplyDeleteI know for fact that they do. For instance, IBM is well known for working on brain-like chip.
2. Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
I am not at all exited by Google activities. I lost my faith with Google long time ago; nowadays I must say I distrust advertising companies.
3. Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
Actually, I don't think that private sector research will ever be possible without academic research. Especially, because their research is usually hard to replicate - their papers are quite often not "open" enough.
I understand your concerns about Google. I have a big problem with their approach to user privacy. They provide many amazing services which I'm using daily. Still advertising is their main revenue source and I'm afraid that by using their services I'm helping them to sell more ads.
DeleteI partially agree with Kinga that publication of article without discipline of time is unacceptable however this subject is a
ReplyDeletein the interest of my research and I try shortly answer questions.
There are. :) However not so long ago I read an article where Google figured as a monopoly of the Polish internet advertisement provider. You should be very well aware that Google is a private company so the decisions would be made behind the closed doors. The situation is similar in our bellowed Academy.
I'm glad that you like the topic. Thanks for taking the time to go through the article.
DeleteYes, Google is extremely strong in internet advertising, still I feel that Facebook is competing with them successfully. Nevertheless, I'm not that concerned about the decision making process but with keeping the research not available to the public.
1.Do you feel that they have any reasonable competition right now?
ReplyDeleteIn my opinion Google don't have any reasonable competition
or maybe I don't hear about it:)
2.Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
I'm not excited about their research , because I looking for new technologies but now in my opinion companies race for research new things and show to other company we have "stronger" but 50 % this things never publicate.
3.Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
This is good question which I don't know what to say;) But in my opinion not all want to use Google so make researcher alone.
Machine learning is spreading more and more in all types of applications, from voice recognition to translation and data analysis. A chip can accelerate workloads is essential. Especially the effect of Moore's Law is slow and that the gains in performance of a processor generation to another are reduced. The use of accelerators to perform key tasks is very important. Google says that the gains provided by its processing unit advance Moore's Law of three generations, about seven years. The TPU is already used in production on the Google cloud. It also feeds the results of sorting system RankBrain and Google voice recognition services. When developers pay to use the voice recognition service from Google, they use the TPU units. At a press conference held during Google I / O, Urs Hölzle, senior vice president of Google for technical infrastructure, said that "the TPU can speed machine learning process, but some functions still require processors and GPUs. " Adding that Google had begun to develop the processing unit about two years ago
ReplyDeleteWow, It's actually pretty impressive that regular developers can already use this in their applications - I'm thinking about Google voice recognition services here. Thanks for pointing this out.
DeleteStill for the whole research community it would be really beneficial if they would open source TPU hardware design. But I guess this is their competitive advantage, as Michail pointed out couple of comments below - it is in their interest not to give everything away.
Hello,
ReplyDeleteI do not think so is any company, which can right now be competitor for Google (they have unlimited resources and what is most important access to data). They research the market and are able to buy each interesting startup/solution, which developes product similar or better to existing in google solutions pallet. Deep Learning if I good remember was an idea bought by google and picasa (or something similar too) image recognition too. If I am wrong feel free to correct my sentence.
Yes I like their solutions, but tons of data about human behaviours might be powerful weapon against network users.
And regarding to yours last question: google is checking researches and tries to absorb each interesting project, but they do not have exclusiveness on ideas and science so everyone - not only academies can try to do something better :) (I will ;)). Otherwise, if I am wrong - why are we in PhD studies ?
So you were pretty close here, in 2014 Google bought DeepMind Technologies, an artificial intelligence startup. These are the guys behind the most impressive (or publicly visible) AI breakthroughs like AlphaGo beating a human professional Go player. They are in fact heavily using Deep Learning among other techniques.
DeleteI like your answer to the last question, and I was actually hoping to see more answers like that. I agree with you on this, I’m glad that you think this way :)
Hey, thanks for bringing us this topic. It is true that Google I/O event is one of the most popular and prominent conferences. Unfortunately I missed it this year, will need to catch up during next week. Now I'll try to answer your questions:
ReplyDelete1. In my research I don't really use any Machine Learning tools and therefore I am not completely up to date with any alternatives to Tensor Flow engine. A quick google search indicate there isn't much competition in this department.
2. Google is one of the few companies that "open" their projects to the public. However active or emerging competitors might come up with an alternative based on Google's idea. It is in the companies interest to not give everything away, especially on projects that are upcoming or related to updates to existing ones.
3. The Google team usually focus on general topics, in Academia you can find several niche subjects which are supplemented by Machine Learning tools. Therefore it is still a prominent problem area involving a great deal of work for the future.
Hi Michail, it's great that you also follow Google I/O events. I hope that you found this article useful and you'll have less catching up to do next week ;).
ReplyDeleteI agree that every company needs to take care of its competitive advantage and not give everything away. Google is really good at this and it's really hard to compete with them. I really like your last remark, you are right to say that there is still a lot of work in niche areas.
Do you feel that they have any reasonable competition right now?
ReplyDeleteInteresting topic. I am not deep enought in the subject to discuss or compare Machine Learning applications, but I think that Google is trying very hard (and seems that successfull) to be a big fish on a market.
Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
We might be anxious about Google politics, but we cannot forget that it is a company, so their main aim is their future well-being. But despite that it is hard to deny that their researches are interesting and exciting.
Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
In my opinion it is always a chance, there will always be the subjects not covered by the big companies.
1. Do you feel that they have any reasonable competition right now?
ReplyDeleteI am not aware of any competition in this field.
2. Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
I am excited about most of Google's work. I am glad that companies like Google see the value in conducting such research. I like the idea of "research that matters" and I don't think that being motivated by future revenue is any bad.
3. Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
As far as I know Google is constantly cooperating with Academia. American universities receive a lot of grant funding so that shouldn't be an issue
1. Do you feel that they have any reasonable competition right now?
ReplyDeleteI am not aware of any competition in this field.
2. Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
I am excited about most of Google's work. I am glad that companies like Google see the value in conducting such research. I like the idea of "research that matters" and I don't think that being motivated by future revenue is any bad.
3. Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
As far as I know Google is constantly cooperating with Academia. American universities receive a lot of grant funding so that shouldn't be an issue
I agree with Kinga. Publication of articles for discussion on the weekend just before closing deadline is completely unacceptable and unprofessional. I am not going to participate in such behavior and I will ignore your post. Have a nice week :-)
ReplyDelete1. Do you feel that they have any reasonable competition right now?
ReplyDeleteI am afraid I am not up to date with this topic. However, I think that they have so engaged and so brilliant team so it is really hard to become a competition for them right now.
2. Are you excited about their research?
I am excited, however, I am quite anxious about it, too. Gathering all available knowledge frightens me.
3. Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
I think that the private sector has to involve academic research. The academic research and knowledge is not always available for everyone so there is a need of a cooperation.
ReplyDelete1. Do you feel that they have any reasonable competition right now?
Until I read the answers of my predecessors I was not aware of any competition.
2. Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
For me it is quite obvious that future revenues is the strongest motivation for a private company. Anyway what they do they do it well. It would be a pity if many of the good ideas which were developed and implemented or bought by Google did not have a chance to be brought to daylight, because of lack of money, access to the market or motivation.
3. Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
I believe there is always a chance to find a fields non covered by private companies. Do we have to perceive both directions of research as a competition ? Apparently, they are cooperating already.
> Do you feel that they have any reasonable competition right now?
ReplyDeleteI had watch Microsoft Build 2016 and amazed how AI could enrich UX. FB and Amazon are also very active there.
> Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
There you defined the word `Capitalism`.
> Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
Don't know much about this.
They do not have significant competition, and this will not change for several years. I am excited about these studies but also whose fear as process and use data. Regarding the third question - cooperation between science and business is the largest development of base
ReplyDeleteI must say Tensorflow it's really great I've made some tutorials from it and saw some online tutorials from Google and this is very impressive. It' so simple, you can use it with C++ or Python. API looks also very nice with examples which is big plus.
ReplyDelete1. Do you feel that they have any reasonable competition right now?
Not long ago I thought that IBM with his Watson (expert system) can beat Google, but after Go match (DeepMind) I must say that I really don't know. Maybe because of Tensorflow they don't have any real competitors.
2. Are you excited about their research? I’m personally a bit anxious that most of their activities is done behind closed doors and is motivated by future revenues.
Yes I must say that Tensorflow is really great and I will you use it in my future researches.
3. Does research on Machine Learning and AI conducted in Academia stand any chance with almost limitless resources of a private company like Google?
When it comes to Google and academia this second is without any chance. Because Google has the best academia scientists in AI field. Academia don't have any chance when scientist can do the same things in Google but for much bigger money.