emcq

10 ms·

31.
by emcq 6y ago
Exciting to see new graphics research! This technique reminds me of old work using motion vectors to not upsample in pixel space but instead upsample in time, performing cheap interpolation. This technique got close to shipping with Force U
32.
by emcq 6y ago
For small networks it's often a win to stay on chip at least on the power side. But if you do need to go off chip for memory it's hard to beat the memory bandwidth you have on a GPU.
33.
by emcq 6y ago
It's no accident Facebook got a deal because they had great data; there is public documentation that Facebook acquried WhatsApp (and likely IG) using data tracking users from a VPN service they bought for much less: https://
34.
by emcq 7y ago
When should one use the Constant Q transform over a Mel or Bark spaced STFT?
35.
by emcq 7y ago
Tubeless bike tires. Lower tire pressures improve comfort, reliability, and traction even on the roads.
36.
by emcq 7y ago
The small energy released by small movements doesn't add up to a large earthquake. For example, magnitude 8 earthquake is a billion times more energy than a magnitude 3 earthquake. This was the sobering perspective of a geophysicist wh
37.
by emcq 7y ago
16 bits per sample can provide 96dB dynamic range [0]. If you were listening to music at that volume, it would generally be considered unsafe, and you would potentially have hearing loss beyond 37 minutes [1] 16 bits is plenty. [0] https:&
38.
by emcq 7y ago
The folks at Skydio are building something different. It's the first autonomous thing (car, boat, plane, drone, etc) that's widely available to consumers. They're releasing a new product soon and if it's better than what
39.
by emcq 7y ago
Imagine being asked to deploy highly complex firmware with non obvious failure cases without any software updates. I sure hope that the models with Tensil are not fixed! Unfortunately a big problem with ASICs is economics/operational i
40.
by emcq 7y ago
Partial checkouts is the main thing that comes to mind that is better with svn than git.
41.
by emcq 8y ago
For articles like this there always seems to be a few comments suggesting statistically Tesla's are the safest cars. Are the cars safe or is it the drivers? The population of Tesla drivers is far from average; the average driver certai
42.
by emcq 8y ago
The NXP i.MX 8M SOC has a Cortex-a53 and M4F.
43.
by emcq 8y ago
What motivates people to invent phrases like "perceptual binarization" when googling "audio binary mask" literally gives you citations in the field that have been doing this for years? For example, 2009 musical sound sep
44.
by emcq 8y ago
It's more than I made when I moved to the bay area. I remember meeting wait staff that earned more than me with more flexible schedules. That said it's easy to look over the skills of some of these folks. Being able to be gracious
45.
by emcq 8y ago
Not much. Even a tiny arm cortex M4, which could live in a hearing aid for a week on battery life, are often 64MHz with single cycle MACs. Z80 I believe was something like 4 cycles per instruction and only a megahertz, so we would be talkin
46.
by emcq 8y ago
How does this compare to approaches like presented by MSR at NeurIPS this year where also do keyword detection on the Pi Zero or impressively the M4F? https://dkdennis.xyz/
47.
by emcq 8y ago
If at first you don't succeed, try try again? This is kind of rediculous. He's pitching Otto + more magic deep learning juice. Is there sufficient evidence deep learning will solve these higher order planning tasks? Unless he has
48.
by emcq 8y ago
Yes there are certainly datasets you can train on a dataset effectively without repetition; this is more common in areas like reinforcement learning where you are simulating every training point. While theoretically possible you have the sa
49.
by emcq 8y ago
This article makes a lot of apples to oranges comparisons that are confusing: 1. The article makes the claim this is the first SNN processor, but then states they will first make an FPGA. They already reference existing processors, that hav
50.
by emcq 8y ago
Assuming the study is taking adults on a 2500 calorie diet down to 1600, an extra 300 calories from metabolic effects is not insignificant but the majority of variance is explained from a simple model of calories where fat calories ~ carb c
51.
by emcq 8y ago
Rajit Manohar (also from Caltech) is an expert in design tools for asynchronous hardware, and moved from Cornell to Yale where he runs an asynchronous VLSI design group [0]. I've worked in groups that have shipped a few chips with his
52.
by emcq 8y ago
There are many researchers not bound by NDAs making novel chip architectures, some of which are on HN. Secondly, GPUs don't really have very much hardware specialized for graphics anymore. If we called it a TPU I'm not sure you&#x
53.
by emcq 8y ago
At a high level there is a design tradeoff where you put your transistors for a given chip. For a dense linear algebra/tensor processor, it basically comes down to using your transistors for memory or compute. GPUs (and DSPs) historica
54.
by emcq 8y ago
There is certainly a lot of hype around AI chips, but I'm very skeptical of the reward. There are several technical concerns I have with any "AI" chip that ultimately leave you with something more general purpose (and not rea
55.
by emcq 9y ago
As a glimmer of hope, consider this: it may be easier to decode a much larger system with redundancies than the highly evolved and tightly optimized 302 neuron system. The C. elegans has a small number of heterogenous neurons. Some are quit
56.
by emcq 10y ago
There are thousands of people invested in the next CNN, GPU, CPU, TPU, or FPGA architecture. Having a few smart people with demonstrated experience try an idea that doesn't fit the current biases in the field is perfectly reasonable fu
57.
by emcq 10y ago
It's interesting you used a Feynman reference here, as he was actually involved spawning the neuromorphic field at Caltech with Carver Mead [0]. You're thinking about this as if it's purely an algorithm problem and that compu
58.
by emcq 10y ago
There have been successes of this line of research. For example, touchpads and Synaptics came from thinking about how to build analog capacitor networks [0]. There are spiking cameras that can produce significantly higher range and temporal
59.
by emcq 10y ago
There's nothing in this article about Turing. The claims are that Moore's law is dying.
60.
by emcq 10y ago
So an ensemble beats a monolith? Would be cool to see if the ensemble with distillation could fit to a more computationally friendly model!

More ›