The analogy to noise-cancelling headphones is helpful but in that case we clearly know which is signal and which is noise. Here, if we knew why would we even bother to the noise-cancelling work?
To make things worse, low attention values will have very low gradient, thus needing a lot of weight updates to undo that kind of mistakes. On the other hand, subtracting the output of two softmax allows the model to predict a weight of exactly zero for some of the values, while keeping a reasonable gradient flowing through.
So the model already knows what is noise, but a single softmax makes it harder to exclude it.
Moreover, with a single softmax the output of all heads is forced to stay in the convex hull of the value vectors, whereas with this variant each head can choose its own lambda, thus shifting the "range" of the outputs outside the convex hull pre-determined by the values. This makes the model as a whole more expressive.
Also, where is each softmax happening here? For each attention head?
In the context of standard transformer attention, each output lies in the convex hull ("somewhere between") the input values. With the modification of this paper, the input values can be scaled a little so that the output of different heads can be in different "regions" and thus do not interfere with each other (so yes to your third question, the two softmaxes are performed separately for each head).
The convex hull (https://en.wikipedia.org/wiki/Convex_hull) of a set is the smallest convex shape that includes that set. Geometrically, it's what you'd get if you "shrink wrapped" the thing you're looking at: edges still protrude, but any indentations get smoothed over.
In this context, the grandparent comment is pointing out that with a traditional transformer block, the resulting computed value for a token can never "stick out" past some weighted average of the values of attended-to tokens, but this differential attention formalism allows that result.
Then y is a convex combination of the v_i, and sits in the convex hull of the v_i.
https://chatgpt.com/share/67058973-ba94-8008-bed7-c7f9d08dc5...
I just don't see how you could answer these questions without trying it out. And chatgtp DEFINITELY isn't doing that.
Plus the obvious question I'd pose is not in there. What's the difference in performance between this trick and just "softmax() - 0.5 * 2" ? That seems very relevant.
Wouldn’t this be pretty unlikely, though?
Attention is really good at finding this smattering of words (ie assign most weight there). But it struggles to put exactly 0 on the other words.
SoftmaxA[n] - SoftmaxB[n] is exactly 0?
Even if 2 attention layers learn two different things, I would imagine the corresponding weights in each layer wouldn’t exactly cancel each other out.
Everything I am seeing in this paper is related to reduced size and noise, which implies a reduction in expressiveness.
The improvement in needle and a haystack, benchmarks on multi-hop questions of in corpus data and multishot in-context learning points to this.
This is a wonderful thing if robustness is more important than generality, but it doesn't address trimming away activations that may be spurious in the general use case but may improve an individual domain specificity.
Context would dramatically impact what tradeoffs and more desireble, and noise is probably never desirable. But the ability of this paper to enable bit size for inference points to a reduction in expressiveness.
Perhaps I am too focused on generalization?
Maybe expressiveness is not the right term, or not the main consequence. I could imagine that having different subspaces like that also introduces a degree of robustness to out-of-distribution inputs, as this would make it harder for the outputs of one attention head to shift towards the in-distribution outputs of another head, and thus for the following layer to confuse them.
I wonder if a different approach without that issue exists. For instance, using max(0, exp(x)-1) instead of exp(x) in the softmax attention formula. That way when the query is orthogonal to the key (or worse), it does not contribute.
Won't this cause the gradient to vanish on the left half, causing problems with training?
If we’re subtracting one attention matrix from another, we’d end up with attention scores between -1 and 1, with a probability of effectively 0 for any single entry to exactly equal 0.
What’s more, the learnable parameter \lambda allows for negative values. This would allow the model to learn to actually add the attention scores, making a score of exactly 0 impossible.
- Rectify the difference of the softmaxes. (min(0, s(A1) - lambda s(A2)))
- Apply the Heaviside function to the second softmax. (softmax(A1) - lambda H(s(A1) - lambda s(A2))
The second one being a bit more drastic and maybe harder to train.
The better example is the differential signalling used in professional audio and many digital signaling protocols like Ethernet, HDMI and USB.
Instead of using one wire, referencing to ground, they send the signal as the difference between both wires. Both wires end up carrying the same signal with inverted polarity. Because both wires are running next to each-other any external noice will be applied to both equally.
The voltage will change, but the difference in voltage between both wires is untouched. And when you subtract the two voltages at the receiver end, any noise simply gets subtracted out.
It sort of feels closer to heterodyning and "demodulating" the signal encoded in the softmax. Those tiny little errors we're trying to denoise with this technique are almost closer to carrier waves (when encoded to softmax) than noise imo. This wouldn't get rid of noise in the training data or noise in the dimensionality of the key / value space. It's really only removing noise introduced by the process itself.
The simple way of doing this would be to just remove the softmax or use a sigmoid instead, but in practice a softmax works better it seems.
to eli5:
rope is the modern strategy used to give information to the model about how far a query and a key are apart when doing attention. It's the best strategy we have now, but has a major downside, where it makes some connections between tokens that are far apart much stronger than you would like them to be. Xpos (https://arxiv.org/pdf/2212.10554) is another paper by microsoft tackling issues with rope and you can see figure 1 on page 4 to get a visual interpretation of the sinusoidal attention strength (you would like it to be smooth).
I think a big reason differential transformers is working so well, especially on long sequence stuff, because when both q1 and q2 don't match a token, the rope relative strength will still have the same value and the noise will cancel out. Leaving intended matches, but at the cost of somewhat dampening the original value rope brought.
Just a hypothesis though. It would be easy to test by running this experiment against a baseline where both use alibi attention (https://arxiv.org/pdf/2108.12409) which has a different set of tradeoffs this wouldn't mitigate, but still a really interesting result.
If that is the case, then the "signal" in this case would be the softmax that encodes the dimensions captured by the query / key space. Since the noise ideally is the same in both softmax encodings, subtracting them should "cancel out" the noise.
For example, if you are trying to send a +1V signal on one wire, and a -1V signal on the other and a +0.5V noise exists, one wire will have +1.5V and the other will have -0.5V,
Take the difference and divide by 2:
(+1.5V - -0.5V) / 2 = +1V or, if your setup is different (-0.5V - +1.5V) / 2 = -1V
The analogy I can think of is when you're paying attention to a variety of things and you actively avoid concentrating on something because it will distract you. You don't give it zero attention, you give it negative attention.