Weights should not be 0 (at least not frequently) but in a ReLU-based neural network, activations are 0 pretty often. You're absolutely right about die area though.
Hardware engineer here. Special casing the multiply by 0 and multiply by 1 paths is harder than it sounds. In software, the cost of adding special cases is simply performance. You're adding more instructions that execute in sequence on a CPU that already physically exists. Doing this for your multiply case is worthwhile because the speedup is large for 0 and 1 while the cost is not that large (relative to the time taken for the whole multiplication operation) for other values.
Hardware is different. Every operation that can be performed in hardware by a chip needs dedicated circuitry. Special casing 0 and 1 means adding at least OR reduction on each operand and a dedicated multiplexer for every bit of the output. Those transistors use power even when they're not in use (leakage power is a huge issue on modern semiconductor processes). They also degrade timing by adding more gates on critical paths through the multipliers. (The timing issue here is that all operations that happen between one flip-flop and another flip-flop need to finish within one clock cycle.) And unless there are whole blocks of 0's and 1's (this does happen in certain neural networks), you typically won't see a direct speedup anyway. In software terms, the matrix multiply is scheduled as many parallel operations that cannot be accelerated much overall by skipping a few operations in some "threads."
All of this makes zero skipping a nontrivial topic. People do still try to do it but it needs serious consideration as, depending on the application, the case is rarely one-sided.
> Take a look at the structure of something like CMOS and you’ll see why running transistors in anything other than “on” or “off” is definitely not energy efficient. In fact, the transitions are where the energy usage largely goes. We try to get through that transition period as rapidly as possible because minimal current flows when the transistors reach the on or off state.
Sounds like you might be thinking of power electronic circuits rather than CMOS. In a CMOS logic circuit, current does not flow from Vdd to ground as long as either the p-type or the n-type transistor is fully switched off. The circuit under discussion was operated in subthreshold mode, in which one transistor in a complementary pair is partially switched on and the other is fully switched off. So it still only uses power during transitions, and the energy consumed in each transition is lower than in the normal mode because less voltage is switched at the transistor gate.
The main issue is that when objects spin really, really fast, they tend to explode due to centrifugal force.
The tensile stress on a spinning round, homogeneous object is p * r^2 * w^2, where p is density, r is radius, and w is angular velocity. Using your numbers for a steel cylinder with density 8 g/cm^3 gives a tensile stress of (8 g/cm^3) * (5 mm)^2 * (2pi*1 MHz)^2 = about 8 TPa which vastly exceeds the tensile strength of steel or any other known material. Using cows connected by ropes would be even worse because the enormous centrifugal force would be borne by only a small rope.
You're ignoring the units of G. The gravitational constant has units N * m^2 / kg^2 = m^3 / (kg * s^2). That contains a length unit raised to the third power, which exactly balances the factor of 8 that you identified.
Landauer's limit assumes that computation uses a thermodynamically irreversible process and erases bits of information. This is not necessarily true for all useful computers [0]. So theoretically speaking, it's not impossible to create an adiabatic microprocessor. Skepticism is definitely warranted though.
To prove that it's possible to buy ads targeted on microphone audio. If this kind of targeting was really as widely available as the article implies, a journalist would be able to do it fairly easily and write an article about it.
In my experience, there are a lot of applications that are "trivial" enough that PID works fine. I'm reasonably comfortable with state space modeling for systems with more variables. But
OK, I'll bite - always open to picking up something new. What resources would you suggest for someone who wants to learn more about nonlinear controls?
Nuclear energy is hardly the only power source that has made an entire city uninhabitable. The Banqiao dam failure flooded 30 cities and affected over 10 million people.
https://en.m.wikipedia.org/wiki/1975_Banqiao_Dam_failure
And we're not even considering the effects of global warming here. Sure, nuclear has risks, but our civilization needs energy and nuclear has proven to be one of the least risky ways to produce it.
The most frequently discussed perovskite absorbers for solar cells are methylammonium lead trihalides, which are (at least partially) organic compounds.
Yes, it's thought that neutron stars are about 5-10% composed of protons and electrons. In fact, that's why a neutron star can have a magnetic field – charged particles are required to carry an electric current.
I agree that the comment you're responding to was off-topic, but calling it "Facebook-style misinformation" goes too far. It's well documented that the Obama administration sent multiple plane loads of cash to Iran in 2016. [0][1][2]
When nitrogen compounds combust, they tend to produce nitrogen gas (N2). Since nitrogen molecules contain a very strong nitrogen-nitrogen triple bond, a lot of energy is released when they form.
You'd need a lot of shielding. A planet is about the right order of magnitude, though it's still not clear that a human could survive. You wouldn't die of acute radiation poisoning; a few thousand km of rock would be enough to absorb the gamma rays. At 400 AU, you wouldn't immediately boil to death, either. Part of the collision-facing surface of the planet would be converted into plasma and ablated into space, which would insulate the rest of the planet from the worst of the heat.
My guess is that you'd still die from shock waves propagating through either the atmosphere or the planet itself, but it's difficult to predict the exact effects.
The question is currently closed, so I can't post a new answer. I voted to reopen it, and I've edited the title to focus on the aspect of the question that is not adequately answered by the duplicate.
Hi Ben, as the person who initially flagged your question as a duplicate, I'm sorry. I can see why this was frustrating. I'll try to provide a more detailed answer here.
According to a 1948 paper by Bekesy [0], a human ear can detect sounds through several mechanisms. In a gravitational wave, since the whole head would vibrate, sound would reach the ear by bone conduction through the skull. Fortunately, this also happens to be the most sensitive mechanism. The smallest vibration amplitude detectable by bone conduction is about 4 x 10^-9 cm at around 3 kHz. If a typical skull is about 17.5 cm from front to back, that gives a minimum detectable wave amplitude h (the fractional expansion and contraction of spacetime) of about 2 x 10^-10.
The frequency of the gravitational waves produced in a collision between a neutron star and a small black hole turns out to be pretty close to an optimal match for the human ear. To estimate the amplitude, the first black hole collision detected at LIGO produced a strain of about 10^-21 at a distance of 410 Mpc. (The events in the article were smaller, but within the same order of magnitude.) Thus, since the amplitude scales as 1/r, the gravitational waves from the collision in the article would have been theoretically audible at a distance of about 400 AU or 40 billion miles.
Alas, attempting to test this calculation experimentally would incur certain practical difficulties. Notably, you would be vaporized by gamma radiation from the collision several seconds before the gravitational waves became audible.