I'm considering doing a side-project to auto transcribe minutes from voice recognition for the meeting and index all that data so its searchable and also plugable to existing enterprise systems through REST API. Does that sound something useful?
Smart image CDN. Process images on-the-fly, store and serve with CDN of your choice. No need to setup any imaging infrastructure (storage, processing and CDN delivery) for web/mobile apps.
Developers get started within 5 mins to signup, point us to where your original images are and replace the <img src=""> from your domain to the new origin you just created.
One of the features is to serve the right image format which can reduce image bandwidth on a page anywhere between 40% to 400%, or more.
The SoC (card sized) has everything you'd need, not just the connectors to the several I/O and networking interfaces you may need. You will hook in connectors are you like with the 400-pin interface on the SoC and get going.
There's one serious problem with using 3rd party Imaging SDKs or utilities to build your service on: You're stuck with their flaws and you can't do much about it.
We have written our own image processing SDK, using only encoders which are 3rd party (libpng, libjpeg, libwebp etc), rest is all handwritten, GPU optimized code. The rationale behind it was to be able to tweak/optimize any feature.
Currently, only AWS offers GPU instances. There are other, smaller providers like Nimbix or Peer1 but they offer bare-metal instances. With AWS, they currently have Grid K520 GPUs which is targeted for public and private Clouds.
The gist of this paper is that non ECC memory does not zero the memory but it conveniently skips the fact that public or private Cloud uses either NVIDIA Tesla or Quadro series cards, both of which have ECC (one of the main reasons they're expensive than say GTX Titan which offers same single/double-precision performance but does not offer ECC). No one really uses GeForce series cards for anything serious and definitely not virtualization so even though this is a vulnerability, it may not effect more than 1e-10% of virtualized GPUs out there
The whole premise behind introducing Phi to compete with discrete GPUs from NVIDIA/AMD was to have a plug-in accelerator which supports x86 which meant no code porting needed, hence enabling companies with millions of man hours invested in their code to simply take benefit of the accelerator. However, this is not the case -- The price/performance ratio for code which is not optimized to make use of massively parallel processors would be mediocre at best.
Besides, Xeon Phi's are reincarnation of project Larrabe which never took off.
If you have to end up optimizing your code for accelerators in any case, x86 or not -- you are better off optimizing it for GPUs instead.
1. Object recognition (there's dog and frisbee in the photo)
2. Object localization (Dog and frisbee's ROI in the photo)
3. Relation estimation (Based on X factors, the dog might be chasing the frisbee).
Not sure what you meant by spatial relations (localization?) but recognizing (what) and localizing (where) would be key to drawing relationships between objects.
They used to. Now Cisco is also building their "enterprise" Cloud to target their financial sector clients (banks et al) because they're seeing business going to Amazon and decline of dedicated data centers to do something which can be done more efficiently by Amazon et al. Security is just going to become even more important and significant
With banks considering the switch to the Cloud, how would this specification address the issue of key management on the Cloud? That's the "key" to Cloud security. Currently, there are solutions available which are obnoxious for those who <i>really</i> want security of their data and not just a <i>tick mark</i> on some data security checklist because key managers CANNOT be hosted on the Cloud. The dominant approach (CipherCloud for example) is to have an encryption gateway in-house which encrypts/decrypts/tokenizes on-the-fly before data leaves for the Cloud. Their approach is to enable support for popular SaaS apps like SalesForce and archive storage like S3 etc which makes sense BUT then again, it's dependent on whether the vendor would add support for more apps or open the API for 3rd party devs to plug in support for other Cloud apps.
Facebook's algorithmic changes to the timeline are more frequent and more disruptive than Google's changes to search result rankings. With Google, search results are a show of explicit interest by a user to find something particular but with Social Media like Facebook, there's no indication of explicit user interest in whatever is shown in the timeline except what Facebook's psychological consultants believe a particular user may want to see based on a number of factors.
Display is definitely challenging since better, bigger displays do improve productivity.
It would really depend on what sort of programming one's trying to do. Perhaps sitting in a serene and quiet outdoor environment would help you formulate an algorithm and maybe quickly prototype it or something along these lines which has more to do with creativity than with writing complex code.
But yes, there's a reason why the new Apple campus will have almost all offices which will have outdoor view :)
Nice! Although training 100k+ samples may crash or be prohibitively slow. Maybe enabling Decaf for training on the GPU in the backend would be a killer feature!
Yet another parallel framework. Without the 3rd party eco-system of APIs for matrix math, any framework is doomed to just add noise, not value. Sure there's some benefit in getting marginal speedup on some algorithms but for real speedup, you need to know the parallel architecture of the processor (GPU, CPU or APU) which means a learning curve. The GPGPU industry has been trying since long to abstract away the fine details and offer a plug-and-play kind of easy-to-learn framework but then we suffer performance losses and it really doesn't make sense to invest in GPUs for the kind of performance gains you get with these high-level APIs.