ewin
@ewintang
Followers
3,435
Following
380
Media
33
Statuses
314
postdoc in theory, UC Berkeley EECS & Miller Institute
Joined November 2018
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
ORM BKK INTER CATWALK
• 160332 Tweets
裏金議員
• 123665 Tweets
True Way Of Worship
• 76710 Tweets
SB19 LIVE
• 56762 Tweets
JACK AND JOKER FLEXTALK
• 42134 Tweets
#ドラゴンボール超スーパーヒーロー
• 33693 Tweets
タリーズ
• 22556 Tweets
Mauritius
• 17866 Tweets
スタートライン
• 17316 Tweets
Chagos Islands
• 17206 Tweets
ピッコロさん
• 15017 Tweets
コータロー
• 13982 Tweets
チカッパ
• 11735 Tweets
イグナイター
• 11132 Tweets
quantum computers are basically one big kickstarter project for physics departments bc they can't afford gpus
9
45
506
Given access to a quantum system of interacting particles at thermal equilibrium, can we estimate the interaction strengths (i.e. learn the Hamiltonian) in polynomial time?
We show that the answer is yes!
6
42
494
dating a complexity theorist must be an adventure. one week she asks you to fold her laundry, the next week you find out you were actually filing her taxes
12
36
408
silly personal note: makise kurisu from Steins;Gate published a solo-author paper in Science at 17. this paper that I wrote at 18 is now appearing at PRL! i'm calling it close enough :)
8
67
282
Consider a system of locally interacting qubits in thermal equilibrium. We show: above a fixed constant temperature, these states are *fully unentangled* and *easy to prepare*! Joint work with
@aineshbakshi
, Allen Liu, and Ankur Moitra.
I find the entanglement result surprising;
4
30
272
A new approach to learning a Hamiltonian H from applications of e^(-iHt), its real-time evolution! Joint with
@aineshbakshi
, Allen Liu, and Ankur Moitra:
3
23
193
I wrote a survey about quantum-inspired classical algorithms on my website; check it out and let me know what you think!
6
56
177
New preprints giving dequantized, low-rank versions of Gilyen et al's lovely quantum singular value transformation algorithm framework:
SVT is powerful. From this result we recover all quantum-inspired ML that I know of, (1/3)
1
44
172
it's the best feeling when your research problem can finally fit in your head so you can think about it whenever.. instant buff to productivity
3
3
171
New paper! A classical algorithm for “quantum singular value transformation”: computing v ≈ p(A)b in O(d⁹r²/ε²) time + linear-time pre-processing, where p is a bounded polynomial of degree d; A is a matrix with stable rank r; b is a vector; ε is error.
1
14
145
Ever wanted to know what your d-dimensional quantum gate does, but it's stuck in a pesky black box? We give an algorithm that applies it O(d²/ε) times to output an estimate ε-close in diamond norm. Our algo has optimal query complexity, zero space overhead, and is pretty simple!
1
8
141
FYI: I updated my recommendation systems paper ()! Now featuring simpler proofs and better runtime.
5
29
136
help, someone DMed me 2TB of random quantum circuit samples and now my computer won't stop training neural nets and stealing government secrets
2
19
132
who are "break" and "lunch" and how do i co-author with them
1
14
121
NOBODY is discussing the most important criterion for evaluating a paper:
does it spark joy?
4
16
113
Thanks to Nature Reviews Physics for inviting me to write this Comment; really happy with how it turned out!
"Dequantizing algorithms to understand
#quantumadvantage
in
#machinelearning
"
@ewintang
explains how dequantizing algorithms can uncover when there is no quantum speedup and perhaps help explore analogies between quantum and classical linear algebra.
0
12
45
0
12
107
If you're interested in a jargon-less explanation of "Learning quantum Hamiltonians at any temperature in polynomial time":
@science_eye
wrote a great piece on it. Thanks, Lakshmi!
An innovative new approach for quickly determining quantum particle dynamics has thrilled the theoretical computer science community.
@science_eye
reports:
1
45
136
1
7
106
.
@kevinjtian
and I tried to write a blog post and idk what happened but next thing you know we had half a monograph. For those intimidated by proofs of Quantum Singular Value Transformation: try our versions!
3
11
93
No, I didn't respond to your email, but I did star it and then for the next week feel bad about not responding, which if you think about it is almost as good
1
6
92
Today I got a paper rejection and a paper acceptance. My life is in perfect balance.
1
1
87
Thanks for the kind words! See also my lecture notes & problem sets at
The great Ewin Tang just finished her grad summer school lectures on Quantum Linear Algebra here at
@the_IAS
Park City Math Institute. Lecture vids at this link:
(I was on the edge of my seat for the dequantization stuff!)
Stay tuned for further updates...
0
30
164
0
14
88
why can't electric kettles also make water colder? like.. just run the circuit backwards lol
14
2
80
you fools. for every vote against me i only grow stronger
Quantum computer scientists, what do you fear the most?
9
18
44
4
11
76
A cool result from this paper: for nxn Hermitian matrices A, B, consider the matrices |A| and |B| where |*| takes the absolute value of all the eigenvalues. Then
|| |A| - |B| || = O(log(n) || A - B ||),
and the log factor is necessary: this is *tight up to constant factors*!
I got answered! turns out I can't get what I want; if you're curious, see the intro of
1
1
15
1
8
76
me seeing literally any classical algorithm: they should make a quantum one of these
3
4
76
you: complex conjugate of the state (|00...0⟩ + |11...1⟩)/√2
me, an intellectual: ket cat bar
1
6
74
practicing my talk in a frantic whisper so as to avoid disturbing others at the airport
2
0
76
omg, this was the other problem scott suggested to me in undergrad (alongside quantum recommendation systems)!
from abstract: "... this means that no graph property testing problems can have super-polynomial quantum speedups"
How symmetric is too symmetric for large quantum speedups?
0
2
17
1
6
75
pros defeat connes :)
Connes disproven!!
Congratulations to Zhengfeng Ji, Anand Natarajan, Thomas Vidick, John Wright,
@henryquantum
!!!!!
3
30
169
0
3
75
New preprint!
And here's a 30-minute talk on this work that I gave at Simons, if you're into that:
Just uploaded a new preprint with
@ewintang
and Jeongwan Haah. We give a sample-optimal and time-optimal algorithm for learning a Hamiltonian H from copies of its Gibbs state at high temperature. As a bonus, we also show how to learn H from the ability to implement exp(-iHt).
0
2
95
3
9
69
Put out a new version of this SVT paper for STOC (). It's totally different now: faster algorithms, simpler backend, stronger results.
It's also now 79 pages long, so here's my less scary 20-minute talk on it:
New preprints giving dequantized, low-rank versions of Gilyen et al's lovely quantum singular value transformation algorithm framework:
SVT is powerful. From this result we recover all quantum-inspired ML that I know of, (1/3)
1
44
172
2
13
67
recently learned about Maarten Van den Nest's '09 paper "Simulating quantum computers with probabilistic methods" from Matt Coudron (and I think someone else, I forget who):
there are strong parallels with my "quantum-inspired" techniques. (1/5)
2
15
65
Somehow I have a wikipedia page?!
thanks to
@OptoLia
and
@fgrosshans
for doing much of the writing; now I have to keep my nose clean and try to avoid a controversy section :p
1
4
60
wow, it turns out designing quantum machine learning algorithms is hard. who knew
4
4
56
bus doesn't show up? have a long wait for an errand? no worries, that just means it's time to Enter The Mind Palace
0
0
55
Super honored to have helped as a
#QIP2020
PC member! I've, um, never gone to QIP before, so this was super new to me. I'm happy with how the program turned out, though, and hopefully I'll see y'all there.
The list of accepted talks for
#QIP2020
is now available. Lots of impressive results this year!
0
37
73
1
2
55
losing all my goodwill with the math gods by repeatedly attempting to bash out proofs without trying any examples first
5
2
54
i'm tired of being stuck on simple problems. time to think about something else! *gets stuck on a complicated problem*
0
1
55
New twin papers on quantum-inspired algorithms for low-rank matrix inversion: and
1
23
51
I recently changed my gmail to put my starred emails at the top of my inbox, but all it did was conveniently group all the difficult-to-respond-to emails together for me to totally ignore
3
0
52
I'm talking at this conference!
Hear from these incredible speakers and many more at our upcoming Diversity in Quantum Computing Conference! Join us for an important conversation about how we can shape a more inclusive and equitable quantum computing ecosystem. Register today:
4
28
104
0
2
48
i don't get this newton stuff. calculus already exists, dummy! what am i supposed to do, invent it again??
3
1
48
since quantum-like matrix inversion (without the low-rank assumption) is BQP-complete, we're getting close to finding the dividing line between classical and quantum capabilities with respect to machine learning.
2
11
45
scary truths like this keep me up at night
1
6
44
intentionally losing a couple log factors in the analysis, to leave evidence of a struggle
1
0
44
have y'all heard about this really obscure algorithm "stochastic gradient descent"? it's wild
3
3
42
babbage also created a lesser-known indifference engine. it was highly successful but quickly forgotten
2
3
39
including classical versions of quantum
* recommendation systems
* principal component analysis
* supervised clustering
* low-rank matrix inversion
* low-rank semidefinite programming
* support vector machines
* low-rank Hamiltonian simulation
* discriminant analysis (2/3)
2
5
39
had some fun with this new resource!
Pretty extraordinary - Tom Lehrer, aged 92, has decided to put all his lyrics and sheet music into the public domain, before he passes away.
386
5K
14K
2
0
38
when i heard about Bernstein's lethargy theorem i got excited bc it sounded really helpful. turns out it's just some math thing about polynomials
3
0
39
i like the phrase "throwing an exception" because yeah sometimes seeing an exception does feel like getting smacked in the face
3
0
37
@QuantumMemeing
she probably just didn't understand you, you should try repeating the information a few more times until it gets through to her
1
0
35
what I've been thinking about this summer:
* neon genesis evangelion
* I'll be honest, it's mostly been eva
4
1
32
does an apple falling on your head help with any research problem or just gravity stuff? *eyeing the honeycrisps in the fridge*
1
1
33
(Those last two are new to our work.) The two preprints are independent work from groups {Nai-Hui Chia, András Gilyén,
@tongyang93
, Han-Hsuan Lin, me, Chunhao Wang} and {Dhawal Jethwani, François Le Gall, Sanjay K. Singh}. I hope y'all find these results helpful! (3/3)
2
5
33
Have QML researchers here seen this paper from
@hayatayamasaki
, Subramanian, Sonoda, and Koashi claiming an exponential quantum speedup for a ML problem?
The speedup passes my sanity checks, but I don't understand the ML context. Thoughts?
3
3
31
I'm very happy that we finally got a resolution to this problem. Check it out!
Joint work with
@AineshBakshi
, Allen Liu, and Ankur Moitra.
0
3
28
I can tell today was productive because instead of no progress, I made negative progress
1
0
27
I definitely felt some twinges of betrayal when I learned that the QRAM I read about in all these QML papers could, as far as we know, not exist
Fig. 1.52: A meme [1].
[1] Quantum Computing Memes for QMA-Complete Teens, American Journal of Political Memeing (2019).
Disclaimer: we do not endorse any political ideology.
3
9
69
1
1
30
I was on
@twimlai
! Feel free to listen to me ramble about my work and then yell at me about the things I get wrong :p
1
8
28
also: if you're my reviewer, you must thank me and my paper before throwing us away
0
1
28
just realized that calculus was invented multiple times. this really takes the wind out of my sails. expect my arxiv manuscript by Q1 2021
1
1
27
Enjoyed this collection of quotes from textbooks showing polynomial interpolation's decades-long bad reputation in numerical analysis.
"A great deal of confusion underlies remarks like these. Some of them are literally correct, but they are all misleading."―Trefethen, ATAP Ch.14
2
3
25
bad time to have a cough in seattle lol. i'm now a vector, which is not as fun as it sounds
3
1
26
have you ever seen a visualization in a paper and thought, wow, this is begging to be made animated or interactive? [Shadrin, Twelve proofs of the Markov inequality]
2
1
24
I'm sorry y'all, I can't keep up the facade... I'm not actually a cat, just a regular person. I understand if you unfollow me after this unconscionable betrayal
3
0
23
nice new paper from
@nadiia_chepurko
,
@ken_clarkson
, Lior Horesh, and David Woodruff:
using sketching tech to give better classical data structures for the "quantum recommendation systems" problem (maybe a sixth power slower than quantum?) and lots else
2
2
25
0
0
24
As we do CS research, our guide is naturally CS themed. Featuring:
⬩ the Cosine-Sine decomposition, used to lift QSP to QSVT
⬩ Chebyshev Series, used to find polynomial approximations to piecewise smooth functions
⬩ and much more (hopefully) aCCeSSible mathematiCS!
2
1
25
Sometimes a union bound is natural and prudent but the one I just did was definitely neither
0
0
24
weird thought: the lo-fi hip-hop girl might be the most visible southpaw right now, since she's everywhere on youtube and always shown writing with her left hand
3
0
23
This article covers a great result, as well as the cool backstory of the technical contributions it builds on!
An outsider has answered the geometry question that legendary problem-solver Jean Bourgain was stuck on for more than three decades -- my latest for
@QuantaMagazine
11
143
524
1
2
24
The big open problem of this paper is (relatively) easy to state; it's about bounding a derivative of the log-partition function.
We solve this problem when β is small. Generalizing to all β would get you halfway towards removing the high-temperature assumption from our paper.
1
1
22
I need a place to get work done so I open up discord and join an empty voice call. "ah, nice and quiet," I think to myself
0
0
21
This was a big open question. Prior work focused on special cases (high-temp, 1D, commuting), but all of the algorithmic techniques provably didn’t extend to the low-temperature regime. The situation was so dire that a recent survey conjectured it was computationally intractable.
1
2
18
ok maybe it's not "good for science" but i think it'd be funny to click submit on your preprint at 3AM and suddenly have a big spinning roulette wheel pop up. if you're lucky you win an anime drawing of a scientist wearing a hair clip with the arxiv logo on it or something
1
1
19
The proof of this fact is left as an exercise for the girl reading this :)
0
2
19
The 1/ε² was a surprise! Lots of fun math from combining sketching and iterative methods: improving the stability analysis of the Clenshaw recurrence (d³ to d²); bounding arithmetic progressions of Cheb coefficients; and sparsifying matrices beyond n/ε². Joint with
@AineshBakshi
.
2
0
18
Anyways, I'm pretty happy with how this project turned out. I learned some new math for working with Gibbs states. Also, I got distracted by a constant in a lemma and ended up spending a week reading Analytic Combinatorics to optimize it, which was fun and definitely worth it
1
0
18
Key idea: we approximate exp(-βH) A exp(βH) (a global object) by a polynomial p(βH, A) (a local object). A and H almost commute, so their eigenstates align; from this mild bound we conclude we only need p ≈ exp in an interval independent of system size, so p can be low-degree!
2
1
15
you wouldn't believe the amount of time I've spent in the past two months just tweaking definitions
1
1
16