highd

10 ms·

61.
by highd 9y ago
That's what I'm thinking - some sort of configurable FPGA like fabric around a bulk of TPUv1 style cores, maybe for routing outputs around so you can do some nice pipe-lining like you might want with CV on video.
62.
by highd 9y ago
I don't think space is an issue, but an ASIC designed exactly for a workload will always beat an FPGA on power. But if you don't know the workload exactly or don't have the money to fab an ASIC then an FPGA will be superior i
63.
by highd 9y ago
This is probably going to be a hyper parallel fixed point / integer engine like TPU gen1. Doing fast matrix multiply over really small fields is very subpar on CPUs and GPUs. That was the initial reasoning behind TPU gen1 - improving r
64.
by highd 9y ago
It's sort of a weird definition for HLMI since it involves cost. I.e., if you can replace an engineer but leasing the hardware, power etc. costs more than the engineer's salary it doesn't count? This is an important point, be
65.
by highd 9y ago
Research labs have been trying this for years with every spectroscopy known to man - the signal is just too weak. I don't see any reason to think Apple has figured something out that labs of optics experts can't do on the benchtop
66.
by highd 9y ago
Sort of off-topic, but for me Jupyter is 90% of the way there to dominating my data analysis / ML work-flow, but that last 10% kills it for me: - Weird behavior when disconnecting / reconnecting to sessions, especially from multip
67.
by highd 9y ago
Exactly. Not to mention the inevitable "No true Scotsman!". This is the fundamental root of the degradation of political discourse over the past few decades, IMO. Conversations without labels I've found almost inevitably resu
68.
by highd 9y ago
It's unproductive to discuss political labels. It's sufficient to say this is a case where regulation was helpful - as are most cases involving protecting common resources. Issues are independent from each other. Continuing to dis
69.
by highd 9y ago
Sort of strange to describe it as a "discovery" - I'm sure a team of engineers with a variety of fields of expertise spent 1-3 years solving problems that led up to this. A "discovery" would seem to describe somethi
70.
by highd 9y ago
The Volta is 16-bit floating point, so closer to second-gen TPU. The first-gen TPU was integer only and used for prediction, not training.
71.
by highd 9y ago
Could you share what sort of model you're using? Something WaveNet inspired?
72.
by highd 9y ago
I've got a client that's wrangling their data into a place where I can do some targeting for upselling, end-to-end - I'm really looking forward to that. Lots of value on the table, IMO, and really interesting and unusual data
73.
by highd 9y ago
-Customer analysis / clustering / behavior visualization -Financials analysis / predictions -Image classification
74.
by highd 9y ago
It's often a fools errand, but on real-world datasets people make scatter plots after using t-sne to project to 2 or 3 dimensions - in many cases it works out alright.
75.
by highd 9y ago
FFT and PCA are totally different. FFT is a fixed linear transform - PCA finds the linear projection to k-space that retains the most variance from the original dataset. If you projected to the top k frequencies you'd just get a couple
76.
by highd 9y ago
This is really, really awesome. Some other ideas that would be really cool: 1) Do key detection and pitch-shift all the loops to a common key before processing. That might make more of the melodies come through the eigenvectors. 2) Visualiz
77.
by highd 9y ago
Srsly: https://fred.stlouisfed.org/series/LNS12300002 Men employment rate is down from 80% to 65%, but women's employment rate is up from 30% to 55% from the 50's to now. In total the average employment rate
78.
by highd 9y ago
Report: https://www.bls.gov/news.release/pdf/empsit.pdf My comment was going to be about how unemployment doesn't count multiple very large and important groups, including those looking for work for 12 months
79.
by highd 9y ago
Make sure to really spend a lot of time explaining your setup. It's really easy for interviewees to nod and say sure even if you're going too fast for them. Remember it's stuff you've seen everyday for 14 years, and thes
80.
by highd 9y ago
You'd prefer a $10,000 lidar over a $8 mobile GPU?
81.
by highd 9y ago
It seems almost willfully ignorant to miss this point. No one making >$50,000 a year is going to see a net gain in their post-tax income.
82.
by highd 9y ago
You basically just need to reduce the P-value you need to claim significance (0.1 to 0.001 or less) to account for the probability of finding those correlations even in noise. This is part of why particle physics has such high standards - y
83.
by highd 9y ago
The paper says "Runtime" is 1-2 ms. I'm under the impression that this is before any compression, as well.
84.
by highd 9y ago
> most of the experts in the Bitcoin system do not think eth has any chance of achieving decentralized scalability Most of the experts in the natural gas industry do not think solar power has any chance of achieving scalability, either.
85.
by highd 9y ago
Yeah, as am I. GNO, AUG and Golem all have some pretty compelling upside. EDIT: Also remember that these ICOs are not equity, they're generally pre-orders for functionality on the system. So it's sort of like comparing revenue vs
86.
by highd 9y ago
Are these valuations really that outrageous compared to startups? Yes they're really early, but the ETH platform is a huge force multiplier in terms of decentralized scaling engineering that you don't have to do, not to mention th
87.
by highd 9y ago
SEEKING WORK | Boston | Remote or local Algorithms / Machine Learning / Optimization Problems / Statistics as well as the full stack infrastructure to build a product around it (generally Flask + HTML/CSS/JS/D3
88.
by highd 9y ago
It's an issue of comprehensibility, not any current performance metric.
89.
by highd 9y ago
I don't think this will have anything to do with training, since it's the TPU team and trying to beat nVidia at floating-point matrix multiplication doesn't seem like a great idea. Run-time can generally be done with integer
90.
by highd 10y ago
The pesticide probably doesn't look very different in the visible spectrum at that low of a concentration - otherwise its presence would be pretty easily detectable to the human eye. If you require detection capabilities exceeding the

More ›