Desktop:
GPU: GTX 1080 Founders Edition
Mobo: ASRock Z170 Extreme7+
CPU: i5 6600K OCed to 4ghz
Cooler: Nzxt Kraken x61
SSD: Samsung 950 Pro
HDD: 5tb Toshiba x300
Mem: 32gb Ripjaws V DDR4
PSU: 1600w BFG
Sound Card: Creative Titanium X-Fi
Case: Custom Lego build
Fan Control: Built in Arduino with a 3.5in touch screen with custom program for temp monitoring and fan control
Work Computer:
2013 Macbook Pro
- CPU upgrade to i7 2.8ghz
- GPU upgrade to Gt 650M 1gb
- SSD upgrade to 756gb PCIe
Peripherals:
Keyboard: Razer Blackwidow Ultimate - Mac
Mouse: Razer Deathadder Chroma, Mad Catz Rat9
Monitors: 27in 2k 144hz Asus MG279Q, 28in 4k 10-bit Asus PB287Q, 4k 55in 21ms response time Samsung UN55JU7500
Sound: Harmon Kardon Sound Sticks II, Onkyo 7.2, 2x Polk T50 floor standing speakers, 2x Polk Monitor 40s bookshelf, Polk Monitor CS2 center, Polk PSW505 12in sub, Dayton 12in sub
Home Server:
Mobo: ASUS P8Z77-I Deluxe
CPU: i3-3220 3.3ghz dual
GPU: GTX 430 1gb
Mem: 8Gb Crucial Ballistix Tracer
SSD: Samsung 850
Case: Custom Lego build
Media Center:
Mobo: Intel DH61AG Thin Mini-ITX
CPU: G630T 3.2Ghz dual
Mem: 4gb
SSD: Samsung 850
Case: Custom Lego build
For software my main daily use OS is Mac. I use Ubuntu for my servers and on my desktop for development. And for games the necessary evil of Windows 10 on the desktop. For an IDE I like Eclipse but have been transitioning to a custom build of Jupyter which I can't recommend enough. Always Chrome for browsing.
Much of machine learning (artificial intelligence) research is already openly shared. Almost by definition this research is not directly related to short-term profits. Even the big companies that spend billions every year on machine learning, Google and Facebook most prominently, have been sharing not only all of their breakthrough algorithms but also the tools they develop to implement them.
One main reason machine learning models are openly shared is they don't just pertain to solving one particular problem. They are more general tools that can be applied to a wide range of data- and problem-sets. They are sophisticated tools that require a skilled practitioner with both a deep understanding of the particular model's strengths and weaknesses as well some amount of domain specific knowledge in order to effectively apply them to solve a problem. Further much of the research is not in solving new problems but rather solving well-defined problems better than previous methods, things like Character Recognition, Sentiment Analysis, Machine Translation...
Machine learning models, being very general tools, are quickly achieving state-of-the-art in a wide range of fields. Much of the new models, what I would consider actual research, already comes from non-profit universities, often indirectly from companies working closely with and recruiting faculty and students. The profit is usually in applying well understood models to new datasets and problems, not so much in developing new models. Despite being non-profit, the competition in these arenas of optimizing solutions to well-defined problems is so fierce that if you can get +.5% accuracy over the current best you publish a paper immediately, then shortly after get funded by Google.
Slashdot Mirror
Desktop: GPU: GTX 1080 Founders Edition Mobo: ASRock Z170 Extreme7+ CPU: i5 6600K OCed to 4ghz Cooler: Nzxt Kraken x61 SSD: Samsung 950 Pro HDD: 5tb Toshiba x300 Mem: 32gb Ripjaws V DDR4 PSU: 1600w BFG Sound Card: Creative Titanium X-Fi Case: Custom Lego build Fan Control: Built in Arduino with a 3.5in touch screen with custom program for temp monitoring and fan control Work Computer: 2013 Macbook Pro - CPU upgrade to i7 2.8ghz - GPU upgrade to Gt 650M 1gb - SSD upgrade to 756gb PCIe Peripherals: Keyboard: Razer Blackwidow Ultimate - Mac Mouse: Razer Deathadder Chroma, Mad Catz Rat9 Monitors: 27in 2k 144hz Asus MG279Q, 28in 4k 10-bit Asus PB287Q, 4k 55in 21ms response time Samsung UN55JU7500 Sound: Harmon Kardon Sound Sticks II, Onkyo 7.2, 2x Polk T50 floor standing speakers, 2x Polk Monitor 40s bookshelf, Polk Monitor CS2 center, Polk PSW505 12in sub, Dayton 12in sub Home Server: Mobo: ASUS P8Z77-I Deluxe CPU: i3-3220 3.3ghz dual GPU: GTX 430 1gb Mem: 8Gb Crucial Ballistix Tracer SSD: Samsung 850 Case: Custom Lego build Media Center: Mobo: Intel DH61AG Thin Mini-ITX CPU: G630T 3.2Ghz dual Mem: 4gb SSD: Samsung 850 Case: Custom Lego build For software my main daily use OS is Mac. I use Ubuntu for my servers and on my desktop for development. And for games the necessary evil of Windows 10 on the desktop. For an IDE I like Eclipse but have been transitioning to a custom build of Jupyter which I can't recommend enough. Always Chrome for browsing.
Much of machine learning (artificial intelligence) research is already openly shared. Almost by definition this research is not directly related to short-term profits. Even the big companies that spend billions every year on machine learning, Google and Facebook most prominently, have been sharing not only all of their breakthrough algorithms but also the tools they develop to implement them. One main reason machine learning models are openly shared is they don't just pertain to solving one particular problem. They are more general tools that can be applied to a wide range of data- and problem-sets. They are sophisticated tools that require a skilled practitioner with both a deep understanding of the particular model's strengths and weaknesses as well some amount of domain specific knowledge in order to effectively apply them to solve a problem. Further much of the research is not in solving new problems but rather solving well-defined problems better than previous methods, things like Character Recognition, Sentiment Analysis, Machine Translation... Machine learning models, being very general tools, are quickly achieving state-of-the-art in a wide range of fields. Much of the new models, what I would consider actual research, already comes from non-profit universities, often indirectly from companies working closely with and recruiting faculty and students. The profit is usually in applying well understood models to new datasets and problems, not so much in developing new models. Despite being non-profit, the competition in these arenas of optimizing solutions to well-defined problems is so fierce that if you can get +.5% accuracy over the current best you publish a paper immediately, then shortly after get funded by Google.