moconnor

19 ms·

151.
by moconnor 6y ago
I used to do a lot of this with an iPad and bluetooth keyboard. I found you can work pretty much anywhere (in the park, against a tree, in a hammock) for an hour or so but after that you have to move on. The pain points after an hour are fi
152.
by moconnor 7y ago
Unfortunately Facebook’s approach sidesteps this complexity by ensuring each player uses the same random seed and searches policy based on information they can all see. It’s not really solving the problem as intended in my opinion.
153.
by moconnor 7y ago
Original human music song: https://m.youtube.com/watch?v=S1jWdeRKvvk With repeated notes/chords and drums this becomes... music! https://m.youtube.com/watch?v=OB2ekdVqPnk
154.
by moconnor 7y ago
My favourite part of this paper is the 1:1 scale image of the posited black hole in the appendix...
155.
The AGI problem we already have(yieldthought.com)
1 points by moconnor 7y ago|0 comments
156.
by moconnor 7y ago
You can get an idea of the content from their GitHub https://github.com/materialsintelligence/mat2vec/blob/master... The author emails are at the end of README.md if you still want to ask for a preprint.
157.
by moconnor 7y ago
I moved from the UK to Germany over ten years ago. I’ve lived in the South and North, city and countryside. None of this article really rings true. Germans are just as happy to tell polite lies and make small talk as the British. Maybe some
158.
by moconnor 8y ago
Actually this shows why OpenAI matters. Google have been training and refining Transformer architectures for years; how unlikely is it nobody tried training a language model at this scale or larger with similar results? Yet from Google we h
159.
by moconnor 8y ago
You can run TensorFlow Lite on a Pi with no problems at all. You can even train and run basic gesture recognition with full TensorFlow on a Pi. Source: wrote a tutorial doing this for Arm ( https://github.com/ARM-software&#x2
160.
by moconnor 8y ago
I worked in HPC when NVIDIA started taking serious market share from Intel. My memory of Intel’s performance comparisons were that they were often technically unsupportable once you scratched the surface. In one case a third party who were
161.
by moconnor 8y ago
Mozilla’s DeepSpeech is available as an offline pre-trained system for English. The quality is far below Google’s speech API as the model is somewhat out of date and more importantly the training data set is much smaller and less general. T
162.
by moconnor 8y ago
I worked in HPC for over a decade. Most of the codes are not hyper-efficient on Intel - particularly not current-generation Intels that need efficient use of very wide vectors to reach peak performance. Many codes are memory-bound, and the
163.
Arm announces 3 TOPs/W ML processor(developer.arm.com)
3 points by moconnor 9y ago|0 comments
164.
by moconnor 9y ago
Overkill is a point of view here. Training and deploying neural networks is becoming easier than ever. In my group at Arm there's a solid expectation that we'll see neural networks integrated into every part of a running applicati
165.
by moconnor 9y ago
I think this is correct - binary cross-entropy on the sigmoid outputs should at least make the network easier to train and may as a consequence improve test performance.
166.
by moconnor 9y ago
An exaflop of mixed-precision compute for $250M over 3 years. That’s ballpark what the HPC community is paying for their exaflop-class machines. You’d still build your own for that money, I think, but it’s an interesting datapoint.
167.
by moconnor 9y ago
Men are on average a couple of years older in a relationship ( https://fivethirtyeight.com/datalab/whats-the-average-age-di... ). That extra few years of career growth can provide a strong financial incentive for the wom
168.
by moconnor 10y ago
On device inference has two important properties: lower latency and lower power. A radio is expensive compared to a DSP (or even a CPU).
169.
by moconnor 10y ago
This should give a moment's pause to anyone who thinks we could keep a superhuman AI in a box.
170.
by moconnor 10y ago
With reluctance, since you ask TAU is a great tool for performance experts but few scientists have the expertise to use it. MAP is designed for them. Plus the multi-core support is damn cool.
171.
by moconnor 10y ago
Thanks! It's been an incredible journey ;-) One of the less obvious benefits of joining ARM is having extra sites to recruit developers at. Adding Manchester and potentially Cambridge to the existing two sites is cool.
172.
by moconnor 10y ago
Allinea product manager and ex debugger dev here, ARM is very serious about the server market and this is a small part of that larger plan. I was surprised and pleased to see us on Hacker News - happy to answer any questions!
173.
by moconnor 10y ago
Sure, just define a good score function...
174.
by moconnor 10y ago
If they're all paying $100/hr you don't need an actual job, you can just perform a dozen interview questions a week and down the offers.
175.
Supercomputer learns to defeat Pong from pixels in under 5 minutes(allinea.com)
3 points by moconnor 10y ago|0 comments
176.
by moconnor 10y ago
"between 29 percent and 38 percent of predictions about whether someone is low-risk end up being wrong" Wouldn't win a Kaggle contest with that error rate. What's not disclosed is the percent of predictions about whether
177.
by moconnor 10y ago
For this simple example just choosing the largest possible fully-connected network with ReLU and L2 regularization to prevent overfit quickly converges to a nice spiral (test loss of 0.001 for me): http://playground.tensorflow.or
178.
Time for Tea and Destiny(yieldthought.com)
2 points by moconnor 11y ago|0 comments
179.
by moconnor 11y ago
"DRAM chips are a little like people: Their faults are not so much in their stars as in themselves. And like so many people, they can function perfectly well once they compensate for a few small flaws." A beautifully whimsical way
180.
by moconnor 11y ago
One of the criticisms of "top" is that it clears the screen and you cannot turn this off. "top -b" works fairly well for this.

More ›