emcq

21 ms·

151.
by emcq 10y ago
Allen Institute for AI and many other smaller companies depending on how left field you want to get.
152.
by emcq 10y ago
This seems slightly overblown. I suspect this was a metric useful for engineering that found its way to external usage. That's because in many recommendation systems in practice you want to filter out the spurious views, and a simple w
153.
by emcq 10y ago
Meh gamma is a simplistic nonlinearity to model the world; if you care about perception use a CIE colorspace, if you care about gaming they have developed more sophisticated nonlinearities for HDR.
154.
by emcq 10y ago
"In fact, the only machine learning program in the world, according to Thrun, is at Carnegie Mellon, which still isn’t churning out talent fast enough to meet the industry’s demand" While there is a separate department for ML at C
155.
Privacy Preserving Queries: Surveying Without Revealing Individuals' Identities(emmettmcquinn.com)
2 points by emcq 10y ago|0 comments
156.
by emcq 10y ago
I was fortunate to see him talk at UCSD recognizing his Nobel prize in 2008. Standing room only, isles filled with folks sitting down likely against fire code. Despite the vast number of people there to see him talk, he was astoundingly hum
157.
by emcq 10y ago
What is being shared between GPUs? Training data is easy to duplicate and share nothing. Large models with shared weights get tricky but less frequent asynchronous updates with schemes like hogwild seem to work with SGD. I believe TF has su
158.
by emcq 10y ago
What kind of gains do you see in practice with tax lots harvesting?
159.
by emcq 10y ago
Cuda did not share many similarities to the graphics pipeline back then. Even now it's a stretch to compare the two. x86 was and is still a weak selling point. Developers from all backgrounds are deploying their apps on varying archite
160.
by emcq 10y ago
Given a fixed area which fits a fixed number of transistors at the same cost, you allocate some portion of those transistors to compute and memory cells. If you want to maintain your number of memory cells without decreasing the number of c
161.
by emcq 10y ago
When there isn't much data, you shouldn't be segmenting your population. Instead it is common in data analysis to collapse variables to have meaningful observations. Why would that be any different here?
162.
by emcq 10y ago
I loved living in San Diego but nothing in the mountains compares to the lush spring, deep winter, hot summers with water temperatures warm enough to not wear a wetsuit, or crisp fall you get in other parts of the country you've undoub
163.
by emcq 10y ago
This feels like a hardware play to me. Start off getting developers to use your inference library, determine the common cases and killer apps, then build an ASIC that does it all using far less power. Perhaps the front facing camera might g
164.
by emcq 10y ago
There are more and more people with credit cards per centralized infrastructure to support payment transactions?
165.
by emcq 10y ago
Nice link! It's not that the algorithms are slow in some Big-O notation; it's that the utilization of hardware resources is typically poor. It's the general problem of balancing a general purpose solution versus an optimized
166.
by emcq 10y ago
Spot on. It seems completely against some of the higher level philosophies of optimization presented in the article. For example, the 90/10 "rule". You should focus your optimizations on things that matter. If applying the pr
167.
by emcq 10y ago
Just this week I was exploring C decompilers and stumbled upon the open source Snowman[0], which worked well for my purposes and can run in a self contained mode with a dependency on Qt5. [0] https://github.com/yegord/s
168.
by emcq 10y ago
Some folks force every line to be Jason serializable. That can be nice for not needing to rewrite a parser for structured log lines. Another case is if you primarily send data via messages between processes, you could save those streams as
169.
by emcq 10y ago
Do you have a reference to the work from DeepMind?
170.
by emcq 10y ago
If they balance compute to memory better than GPUs, you could definitely see a 10x. GPUs have large off chip memory and small caches (like 256kb). Cost to going to off chip memory can be 1-2 orders of magnitude more than on chip memory. You
171.
by emcq 10y ago
External memory kills your power budget. Efficient processors, whether they be GPUs, cpus, or asics all gain efficiencies by keeping data local to computation. Having your chip focused on off chip data is difficult to optimize power for if
172.
by emcq 10y ago
It often makes sense to just use spark for event processing to create a shaker set of features that are used as input to more robust ML libraries like scikit learn.
173.
by emcq 10y ago
That's because he has had tremendous impacts on computer science and robotics for many years now as a pioneering professor at CMU and Stanford. He was a leading force in the DARPA grand challenge which showcased self driving car techno
174.
by emcq 11y ago
$1 doesn't make any sense. Perhaps you're leaving out some form of equity swap. Bankrupt companies get purchased all the time for millions of dollars and I can't imagine Microsoft's business team would allow such a shitt
175.
by emcq 11y ago
It's confusing but automatic differentiation, symbolic differentiation, and finite differences are all different approaches. "Automatic differentiation" typically refers to reverse order chain rule. Higher level libraries do
176.
by emcq 11y ago
I've done at least one race where they gave you those twisty pipe cleaners, but I would be extremely surprised if this was used at a big event. Those things can be a bit easier to yank off. With most wrecks it's really hard to hav
177.
by emcq 11y ago
The original point was that there is risk. Its very unreasonable to claim that the risk is zero. Perhaps you don't believe this is the primary risk to Tesla, although it seems more like you think Tesla is unfailable, but your counterpo
178.
by emcq 11y ago
The grandparents post is an alarm clock. The blue light things you use during early morning like with breakfast to prevent SAD. I found the alarm clock to be amazing for about a week and then not much different than other alarms after I adj
179.
by emcq 11y ago
This if exactly the perspective at funding large projects at IBM Research. Interesting that the industry R&D labs fund more like top VCs in that regard.
180.
by emcq 11y ago
Furthermore it's not a useful trend anymore if it's due for an update. What would that new constant be that people feel confident will hold over the next 10 years? Seems like quite the Frankenstein "law" to be hacking pi

More ›