chunk_uid
stringlengths
40
40
chunk_type
stringclasses
2 values
chunk_index
int64
0
6.71k
total_chunks
int64
1
6.71k
section_title
stringlengths
1
157
embed_text
stringlengths
1
83.3k
spans
dict
paper_doi
stringlengths
0
63
paper_id_arxiv
stringlengths
9
16
title
stringlengths
7
245
authors
listlengths
1
768
categories
listlengths
1
7
year
int64
2k
2.02k
language
stringclasses
2 values
discipline
stringclasses
8 values
dense_vector
listlengths
1.02k
1.02k
b19c43da8caf14349b719dfc5565b398ff717a6b
subsection
11
73
Prior on
Note that, by the independence of the components of the d-dimensional Brownian motion, the components \left\lbrace \left\lbrace {X}_{j}(t);\ 0 \le t \le T \right\rbrace ;\ j \in [d] \right\rbrace of \left\lbrace {X}(t);\ 0\le t\le T \right\rbrace are independent.More involved correlation models of \left\lbrace \left\lb...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1029, "openalex_id": "https://openalex.org/W2596317820", "raw": "S. Särkkä, Recursive Bayesian Inference on Stochastic Differential Equations, PhD thesis, Helsinki University of Technology, 2006.", "source_ref_id": "44dcdf47...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.014432395808398724, 0.01467649545520544, -0.040093377232551575, -0.001916564186103642, 0.0022674575448036194, 0.009168995544314384, -0.0083070183172822, 0.013753493316471577, 0.015973275527358055, 0.04296154901385307, -0.018444783985614777, 0.021267186850309372, 0.028071466833353043, 0....
43b82245566c2f385aba7a117cdad09a58239292
subsection
12
73
Prior on
In this case, the matrices A and Q from driftmatrix are given by{\normalcolor A(h)_{ij}} &= {\left\lbrace \begin{array}{ll} \mathbb {I}_{i\le j} \frac{h^{j-i}}{(j-i)!}, & \mbox{if }j\ne q, \\ {\normalcolor \frac{h^{q-i}}{(q-i)!} - \theta \sum _{k=q+1-i}^{\infty } \frac{(-\theta )^{k+i-q-1} h^k}{ k! }} , & \mbox{if }j=q...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.04849066957831383, -0.02459627389907837, -0.04073948413133621, -0.03707750886678696, -0.013900251127779484, 0.0234213899821043, -0.010123838670551777, 0.02069016546010971, -0.008224188350141048, 0.0149759566411376, -0.016250019893050194, 0.012305766344070435, -0.02587796561419964, 0.026...
75c1c0a76d5010ed3ecef4cd451e9773df379b98
subsection
13
73
The algorithm
To avoid the introduction of additional indices, we will define the algorithm {\Psi } for d=1; for statements on the general case of d\in \mathbb {N} we will use the same symbols from eq:C-predict–eq:CupdateMM as vectors over the whole dimension—see e.g. bound:r for a statement about a general r \in \mathbb {R}^d. By t...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.48550/arxiv.1605.03364", "end": 812, "openalex_id": "https://openalex.org/W2365455488", "raw": "H. Kersting and P. Hennig, Active uncertainty calibration in Bayesian ODE solvers, Uncertainty in Artificial Intelligence (UAI), (2016).", ...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.017108503729104996, 0.0025010311510413885, -0.06916657835245132, 0.004200816620141268, 0.01444531511515379, -0.0034892342519015074, 0.011255593970417976, -0.006921236868947744, 0.0133998803794384, 0.06434384733438492, -0.028951678425073624, 0.004292387515306473, 0.028478560969233513, 0....
bd64fa0e3c20f17357fe1d87fd9444dd896dc03b
subsection
14
73
The algorithm
In the subsequent step, the following quantities are computed first\beta ^{(i)}(t+h) &\frac{P^-(t+h)_{i1}}{(P^-(t+h))_{11} + R(t+h)} \in \mathbb {R}, \quad \text{(Kalman gain on $i$\textsuperscript {th} state)}, \\ y(t+h) &f\left( m^{-,(0)}(t+h) \right) \in \mathbb {R}, \qquad \text{(measurement/data on $\dot{x}$)}, ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s11222-017-9798-7", "end": 1650, "openalex_id": "https://openalex.org/W2534928453", "raw": "M. Schober, S. Särkkä, and P. Hennig, A probabilistic model for the numerical solution of initial value problems, Stat. Comp., 29 (2019), pp...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.023667512461543083, -0.002769602695479989, -0.06250298768281937, 0.021561698988080025, 0.011505677364766598, 0.01777733862400055, 0.0050165667198598385, 0.029603466391563416, 0.007568721193820238, 0.026963569223880768, -0.04272665083408356, 0.011490417644381523, -0.011017371900379658, 0...
8935572fcc954d8e16394bae7561944578acc3f3
subsection
15
73
The algorithm
Concerning the data generation mechanism for y eq:defyMAP, we only consider the maximum-a-posteriori point estimate of \dot{x}(t) given \mathcal {N}(m^{-,(0)}(t),P_{00}^-(t)); a discussion of more inclined statistical models for y can be found in . Next, for lack of such a discussion for R, we will examine different ch...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s11222-017-9798-7", "end": 248, "openalex_id": "https://openalex.org/W2534928453", "raw": "M. Schober, S. Särkkä, and P. Hennig, A probabilistic model for the numerical solution of initial value problems, Stat. Comp., 29 (2019), pp....
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.023470142856240273, -0.020189205184578896, -0.03131387010216713, 0.018037522211670876, -0.028032934293150902, -0.041904427111148834, -0.01954827830195427, -0.033572375774383545, 0.0200213436037302, 0.038150426000356674, -0.015626415610313416, 0.0029623806476593018, -0.0015613061841577291,...
f44297150a1ae880c6654270ac12961676537b88
subsection
16
73
Measurement noise
Two sources of uncertainty add to R(t): noise from imprecise knowledge of x(t) and f. Given f, previous integration steps of the filter (as well as an imprecise initial value) inject uncertainty about how close m^-(t) is to x(t) and how close y = f(m^-(t)) is to \dot{x}(t)) = f(x(t)). This uncertainty stems from the di...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s11222-016-9671-0", "end": 1302, "openalex_id": "https://openalex.org/W2467245180", "raw": "P. R. Conrad, M. Girolami, S. Särkkä, A. Stuart, and K. Zygalakis, Statistical analysis of differential equations: introducing probability m...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.036653898656368256, -0.02024967595934868, -0.035921432077884674, 0.013665098696947098, 0.01496980618685484, -0.0016862014308571815, 0.014115260913968086, 0.012818182818591595, 0.024324027821421623, 0.027452275156974792, -0.021226299926638603, -0.049411043524742126, -0.014229709282517433, ...
d11780c341f1b8971afa4227c911ecdc0ec46ef7
subsection
17
73
Regularity of flow
Before we proceed to the analysis of {\Psi }, we provide all regularity results necessary for arbitrary q,d \in \mathbb {N} in this section.Assumption 1 The vector field f \in C^{q} ( \mathbb {R}^d; \mathbb {R}^d ) and all its derivatives of order up to q are uniformly bounded and globally Lipschitz, i.e. there exists...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.09456505626440048, 0.030784009024500847, -0.028387831524014473, -0.007516670506447554, 0.008104268461465836, -0.02449595183134079, 0.0010263880249112844, -0.016391683369874954, 0.03174553066492081, 0.029105158522725105, -0.00493353046476841, 0.007276289630681276, 0.016727453097701073, 0...
a72c8a0f1f0eb02be3406c65a7d9125ec3e9249a
subsection
18
73
Regularity of flow
Now, applying \Vert \Phi ^{(q+1)} \Vert \le L to the term \Phi _{\tau }^{(q+1)}(a) (for some \tau \in (0,h)) in the Lagrange remainder of the (q-i)th-order Taylor expansion of \Phi _h^{(i)}(a) yields eq:Taylorexpansion. Under ass:fGlobalLipschitz and for all sufficiently small h>0,\sup _{a\ne b \in \mathbb {R}^d} \f...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1090/gsm/140", "end": 450, "openalex_id": "https://openalex.org/W1521738998", "raw": "G. Teschl, Ordinary Differential Equations and Dynamical Systems, American Mathematical Society, 2012.", "source_ref_id": "0edfeb8a5b39d3b0ea5866...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.07488027215003967, 0.014509197324514389, -0.03591445833444595, 0.0009359080577269197, 0.009733824990689754, -0.009001499973237514, -0.012365619651973248, 0.0005320801865309477, 0.015867050737142563, 0.002114970935508609, -0.04909631609916687, -0.006556600797921419, -0.028453893959522247, ...
b1d81ce1a7d536dcc8ee0417a62b84a35532cd93
subsection
19
73
Regularity of flow
The norm \cdot h is used to make rigorous the intuition that the estimates of the solution^{\prime }s time derivative are `one order of h worse per derivative^{\prime }.We bound the second summand of{\left|\hspace{-1.0625pt}\left|\hspace{-1.0625pt}\left|{\Phi }_h(a) - {\Phi }_h(b) \right|\hspace{-1.0625pt}\right|\hspa...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.07548839598894119, -0.000977359595708549, -0.006133050657808781, -0.0030226632952690125, -0.015195319429039955, -0.017590565606951714, 0.022304775193333626, 0.0008944032015278935, 0.021221576258540154, -0.009939508512616158, -0.10716056823730469, 0.02450168877840042, -0.03231293708086014,...
a2fdbd8283c5fb2315511ca318844a55f2a68b68
subsection
20
73
Auxiliary bounds on intermediate quantities
The ODE filter \Psi iteratively computes the filtering mean {m}(nh)=( m^{(0)}(nh),\dots , m^{(q)}(nh) )^{\intercal } \in \mathbb {R}^{(q+1)} as well as error covariance matrices P(nh) \in \mathbb {R} on the mesh \lbrace nh \rbrace _{n=0}^{T/h}. Ideally, the truncation error over all derivatives{\normalcolor {\varepsilo...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06715576350688934, -0.019673585891723633, -0.047161657363176346, -0.018422046676278114, 0.006635447032749653, 0.005998230539262295, 0.007997640408575535, -0.031929511576890945, 0.007047539111226797, 0.07258927077054977, -0.057631850242614746, -0.002262691268697381, 0.040873438119888306, ...
efc54671b106d8b75dc2870239813969d1a93c1e
subsection
21
73
Auxiliary bounds on intermediate quantities
To capture this difference, let us first, for all i \in [q+1], recursively define f^{(i)} \colon \mathbb {R}^d \rightarrow \mathbb {R}^d by f^{(0)}(a) = a, f^{(1)}(a) = f(a) and f^{(i)}(a) = \left( \nabla _x f^{(i-1)} \cdot f \right)(a). This implies{\normalcolor \Phi _0^{(i)} (m^{(0)}(nh) ) = f^{(i-1)}(m^{(0)}(nh)), }...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.0867280587553978, -0.05239693447947502, -0.02691560611128807, -0.024260664358735085, -0.0019969274289906025, 0.0255118440836668, -0.020552903413772583, 0.022155022248625755, 0.013625644147396088, 0.06261997669935226, -0.04492037743330002, -0.03494146093726158, -0.01895078271627426, 0.00...
0e9c86d4308c98fab7b120de910c8218c471aae6
subsection
22
73
Auxiliary bounds on intermediate quantities
We apply the triangle inequality to the definition of \Delta ^{-(i)}((n+1)h), as defined in def:Delta-, which, by Aprediction, yields\Delta ^{-(i)}((n+1)h) &\le \sum _{k=i}^q \frac{h^{k-i}}{(k-i)!} \delta ^{(k)}(nh) + {\normalcolor K \theta \left|m^{(q)}(nh) \right|h^{q+1-i}} \\ &\quad + \underbrace{\left|\sum _{l=i}^q...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1098/rspa.2015.0142", "end": 1691, "openalex_id": "https://openalex.org/W3099671035", "raw": "P. Hennig, M. A. Osborne, and M. Girolami, Probabilistic numerics and uncertainty in computations, Proc. A., 471 (2015), pp. 20150142, 17, http...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.041768044233322144, 0.009313144721090794, -0.06974561512470245, -0.00495023000985384, -0.0302963238209486, 0.045429229736328125, 0.058121345937252045, 0.04396475479006767, 0.002877464285120368, 0.035330455750226974, -0.0634605810046196, 0.016612635925412178, 0.004404865670949221, 0.0325...
5491bd12dc7367c117e93200e2e0f02019d4cbde
subsection
23
73
Auxiliary bounds on intermediate quantities
Application of the orders of A and Q from def:AIOUP,eq:QIOUP, the triangle inequality and ass:assumption2 to the definition of P^- in eq:C-predict yields\left|P^-(h)_{k,l} \right|&\stackrel{{eq:C^-_predict}}{\le } \left|\left[ A(h)P(0)A(h)^{\intercal } \right]_{k,l} \right|+ \left|Q{\normalcolor (h)}_{k,l} \right|\\ &\...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.024259045720100403, -0.011214086785912514, -0.06316506117582321, -0.02343515306711197, -0.00834572222083807, 0.005061596632003784, 0.029065081849694252, 0.051203370094299316, 0.011565004475414753, 0.046107444912195206, -0.05495665594935417, -0.003913487307727337, -0.0011862521059811115, ...
3e01df8a95abebe1ff742108ae814cc40b684598
subsection
24
73
Local convergence rates
With the above bounds on intermediate algorithmic quantities (involving state misalignments \delta ^{(i)}) in place, we only need an additional assumption to proceed—via a bound on \delta ^{(i)}(0)—to our first main result on local convergence orders of {\Psi }.Assumption 3 The initial errors on the initial estimate o...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.04049532860517502, 0.008361507207155228, -0.01412911619991064, 0.0005731370765715837, -0.008041084744036198, 0.014304585754871368, 0.027129124850034714, 0.00017153582302853465, -0.012328646145761013, 0.026350954547524452, -0.04308922961354256, -0.0228263046592474, -0.010085687041282654, ...
6dd5cfed85c4af90ab0a10bf7d3d9bd2a0682a14
subsection
25
73
Local convergence rates
Now, we can bound the local truncation error \varepsilon ^{(0)}(h) as defined in eq:defvarpepsilon.[Local Truncation Error] Under ass:fGlobalLipschitz,ass:assumption2,ass:eps0bound and for all sufficiently small h>0,\left\Vert \varepsilon ^{(0)}(h) \right\Vert \le {\left|\hspace{-1.0625pt}\left|\hspace{-1.0625pt}\left...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.037009093910455704, 0.013165230862796307, -0.005591027904301882, 0.0007022155332379043, 0.0008037579245865345, -0.03707011789083481, 0.028923843055963516, -0.02189120091497898, 0.005099047906696796, 0.010243861004710197, -0.05369827523827553, 0.010396413505077362, 0.0031082453206181526, ...
c54cd288fed600321dd844b4302845e9ee9cb88e
subsection
26
73
Local convergence rates
\end{align} The remaining bound on (i)(h), for all i [q+1] and sufficiently small h>0, is obtained by insertion of the bounds from {lemma:Delta^-i,lemma:r,lemma:Kalman_gain_local_order} (in the case of n=0), into {def:Delta^-}: \begin{align} \Delta ^{(i)}(h) &\le K \left[ 1 + {\normalcolor \theta \left\Vert m^{(q)}(0) ...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06482133269309998, 0.012100592255592346, -0.026261184364557266, -0.018463702872395515, -0.006633962504565716, 0.029618220403790474, 0.016327407211065292, 0.0034562221262604, 0.001705222181044519, 0.01728873886168003, -0.05032503232359886, 0.016922516748309135, -0.016251109540462494, 0.0...
988cccf7f198c378824e49cf74027fe64b4942e7
subsection
27
73
Local convergence rates
A global analysis for IOUP is therefore more complicated than for IBM: Recall from {A_prediction} that, for q=1, the mean prediction for x((n+1)h) is \begin{align} \begin{pmatrix} m^{-,(0)}((n+1)h) \\ m^{-,(1)}((n+1)h) \end{pmatrix} \stackrel{{A_prediction}}{=} \begin{pmatrix} m^{(0)}(nh) + h m^{(1)}(nh) - \theta \lef...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.05923935025930405, 0.011002466082572937, -0.041995540261268616, 0.01844934932887554, -0.005249442532658577, -0.03265641629695892, -0.003143561538308859, 0.022905271500349045, 0.04330790415406227, 0.0362882986664772, -0.042361781001091, 0.0003726780123542994, -0.0038951169699430466, 0.01...
05a8f9924889adf2c9281cbed0ca5218e6b75032
subsection
28
73
Global analysis
As explained above, we only consider the case of the IBM prior, i.e. \theta = 0, in this section. Moreover, as for q \ge 2 the proof of an analogue of proposition:steadystatesglobal would be very technically involved, we restrict our analysis to q=1 in order to maintain readability of this first global analysis. (See s...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s11222-017-9798-7", "end": 1319, "openalex_id": "https://openalex.org/W2534928453", "raw": "M. Schober, S. Särkkä, and P. Hennig, A probabilistic model for the numerical solution of initial value problems, Stat. Comp., 29 (2019), pp...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.0552818588912487, 0.013675468042492867, -0.026877788826823235, 0.02545834891498089, 0.00613182969391346, 0.015400164760649204, 0.0018019642448052764, 0.008600587025284767, 0.04575787112116814, 0.06028805673122406, -0.02266525663435459, 0.03014402836561203, 0.029548779129981995, 0.000565...
ea704bf89296fc91f2078be7ddcf9efb2af058fd
subsection
29
73
Outline of global convergence proof
The goal of the following sequence of proofs in subsection:Globaltruncationerror is theorem:GlobalTruncation. It is proved by a special version of the discrete Grönwall inequality whose prerequisite is provided in lemma:truncationerrorglobalbound. This lemma:truncationerrorglobalbound follows from lemma:flowmapregulari...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1016/0166-218x(87)90064-3", "end": 248, "openalex_id": "https://openalex.org/W2042816208", "raw": "D. S. Clark, Short proof of a discrete Gronwall inequality, Discrete Appl. Math., 16 (1987), pp. 279–281, https://doi.org/10.1016/0166-218...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.05964728072285652, 0.03453745320439339, -0.01603306643664837, -0.03148644044995308, -0.027077730745077133, 0.003428574651479721, -0.034506943076848984, -0.028389664366841316, -0.017467042431235313, 0.032737355679273605, -0.0574810616672039, 0.014179577119648457, 0.00346861919388175, 0.0...
0c17150105e1fdcced5e467e1f4281e9d9c551b7
subsection
30
73
Global bounds on Kalman gains
Since we will analyse the sequence of covariance matrices and Kalman gains using contractions in proposition:steadystatesglobal, we first introduce the following generalization of Banach fixed-point theorem (BFT).Let (\mathcal {X},d) be a non-empty complete metric space, T_n \colon \mathcal {X} \rightarrow \mathcal {X}...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.02035289630293846, 0.045618560165166855, -0.0549863837659359, 0.0010422467021271586, -0.004764011595398188, -0.004027859307825565, 0.050866980105638504, 0.01652337796986103, 0.009917077608406544, 0.025464003905653954, -0.02709650807082653, 0.01608092337846756, -0.005088224075734615, 0.0...
d38d6a007eee3d05fedf78c7f3562f14673aa253
subsection
31
73
Global bounds on Kalman gains
Then, for all x_0 \in \mathcal {X}, the recursive sequence x_n T_n(x_{n-1}) converges to u^{\ast } as n \rightarrow \infty .See Appendix:Banach.For constant R \equiv K h^p with p \in [0,\infty ], the unique (attractive) steady states for the following quantities areP_{11}^{-,\infty } &\lim _{n \rightarrow \infty } P_{1...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.03286968171596527, 0.03335799649357796, -0.028139133006334305, -0.027360880747437477, -0.0177471861243248, 0.024507291615009308, 0.033083319664001465, 0.014298464171588421, 0.020783893764019012, 0.0300923902541399, -0.04892302677035332, 0.009102490730583668, -0.006664732005447149, 0.000...
990885a3e00ba0583c573ff46a2678928c9b6ec8
subsection
32
73
Global bounds on Kalman gains
The recursions for P(nh) and P^-(nh) given by eq:C-predict,eq:CupdateMM follow a discrete algebraic Riccati equation (DARE)—a topic studied in many related settings . While the asymptotic behavior lemma:steadystatesvaluesi of the completely detectable state X^{(1)} can also be obtained using classical filtering theory ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 575, "openalex_id": "https://openalex.org/W2293807537", "raw": "B. Anderson and J. Moore, Optimal Filtering, Prentice-Hall, Englewood Cliffs, NJ, 1979.", "source_ref_id": "82b0c05c1d4a1f23abdcc4eb7de22a18ecccfcf1", "st...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.05076931044459343, 0.008420870639383793, -0.030434126034379005, 0.016307810321450233, 0.0008018515072762966, -0.023859133943915367, 0.04661989584565163, 0.044941823929548264, -0.0047825053334236145, 0.04518590494990349, -0.04250099137425423, -0.014446675777435303, 0.011777015402913094, ...
ed29769ac60eb329580fc9bafa5497ade5140168
subsection
33
73
Global bounds on state misalignments
For the following estimates, we restrict the choice of p to be larger than q=1.Assumption 4 The noise model is chosen to be R\equiv Kh^p, for p \in [q,\infty ] = [1,\infty ], where Kh^{\infty } : = 0.Before bounding the added deviation of {\Psi } from the flow {\Phi } per step, a global bound on the state misalignment...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.0697498768568039, -0.012616612017154694, -0.015286390669643879, -0.0025477318558841944, -0.021159904077649117, 0.0051603009924292564, 0.01662890799343586, 0.01099186111241579, 0.028909891843795776, 0.015538113191723824, -0.029901523143053055, 0.005412023048847914, 0.028848867863416672, ...
7d2574f1fc15e6d284c44756a1b98c453447826a
subsection
34
73
Prerequisite for discrete Grönwall inequality
Equipped with the above bounds, we can now prove a bound on the maximal increment of the numerical error stemming from local truncation errors which is needed to prove ineq:lemma:developmentepsilon, the prerequisite for the discrete Grönwall inequality.Under ass:fGlobalLipschitz,ass:assumption2,ass:eps0bound,ass:R and ...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.08651144802570343, 0.042156778275966644, 0.001879273448139429, -0.02483312040567398, -0.012813401408493519, 0.016774185001850128, -0.017369447275996208, -0.03537994623184204, -0.016529975458979607, 0.019628390669822693, -0.05195571109652519, 0.038493625819683075, -0.0433778278529644, 0....
d9d08d7332801c1e1792b2b686a2338a5e1b5528
subsection
35
73
Global convergence rates
With the above bounds in place, we can now prove global convergence rates. [Global truncation error] Under ass:fGlobalLipschitz,ass:assumption2,ass:eps0bound,ass:R and for all sufficiently small h>0,\max _{n \in [T/h + 1]} \left\Vert \varepsilon ^{(0)}(nh) \right\Vert \le \max _{n\in [T/h + 1]} {\left|\hspace{-1.0625p...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06275129318237305, 0.002676315139979124, 0.0051847645081579685, 0.009194468148052692, -0.026232847943902016, -0.030933091416954994, -0.00379796396009624, -0.03873122110962868, -0.01568782329559326, 0.021746253594756126, -0.049199942499399185, -0.007290717214345932, 0.030261628329753876, ...
30d9f799d8a096943970f536469540b981bfab94
subsection
36
73
Global convergence rates
The convergence rate of this theorem is sharp in the sense that it cannot be improved over all f satisfying ass:fGlobalLipschitz since it is one order worse than the local convergence rate implied by lemma:Psiwithdelta.Using \left\Vert \varepsilon ^{(0)}(nh) \right\Vert \le {\left|\hspace{-1.0625pt}\left|\hspace{-1.062...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1016/0166-218x(87)90064-3", "end": 2922, "openalex_id": "https://openalex.org/W2042816208", "raw": "D. S. Clark, Short proof of a discrete Gronwall inequality, Discrete Appl. Math., 16 (1987), pp. 279–281, https://doi.org/10.1016/0166-21...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.033883851021528244, 0.055313099175691605, 0.00931805931031704, -0.035715412348508835, -0.00984463281929493, -0.003044967772439122, -0.010989357717335224, -0.018239280208945274, -0.00913490355014801, 0.022863969206809998, -0.0676456019282341, 0.012134082615375519, -0.02935074269771576, -...
e9461145db123d7b07bdf375b17b822f07759b7a
subsection
37
73
Global convergence rates
Recalling that \left\Vert \varepsilon ^{(0)}(nh) \right\Vert \le {\left|\hspace{-1.0625pt}\left|\hspace{-1.0625pt}\left|{\varepsilon }(nh) \right|\hspace{-1.0625pt}\right|\hspace{-1.0625pt}\right|}_h, by def:hnorm, concludes the proof.
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.03168494999408722, 0.003268082858994603, 0.012538143433630466, 0.018391313031315804, -0.02420632541179657, -0.05369347333908081, 0.015293024480342865, -0.012690768577158451, 0.0065247188322246075, -0.02559521235525608, -0.019841250032186508, 0.02220693789422512, -0.001821006997488439, 0...
3f48d3768b609fb8f84d03cfe1d342cc25332467
subsection
38
73
Calibration of credible intervals
In PN, one way to judge calibration of a Gaussian output \mathcal {N}(m,V) is to check whether the implied 0.95 credible interval [m-2\sqrt{V},m+2\sqrt{V}] contracts at the same rate as the convergence rate of the posterior mean to the true quantity of interest. For the filter, this would mean that the rate of contract...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.05485507473349571, -0.0026866935659199953, -0.02463596872985363, 0.010212867520749569, -0.022469835355877876, -0.007734016515314579, -0.010784910060465336, -0.03438357636332512, -0.0032930588349699974, 0.06461793929338455, -0.05357369780540466, 0.011311189271509647, -0.0017161278519779444...
457660507f8ddd623e2df11a0a5ca53510d905b5
subsection
39
73
Calibration of credible intervals
By P_{01}^-(nh) \le P_{01}(nh) and \vert \beta ^{(1)} \vert \le 1 (due to defbetaMM), application of the triangle inequality to eq:recursionP00- yieldsP_{00}^-\left((n+1)h \right) \le & P_{00}^- ( n h ) + \left|\beta ^{(0)}(nh) \right|\left|P_{01} (nh) \right|+ 2hR \left|\beta ^{(0)}(nh) \right|+ h^2 R + \frac{\sigma ^...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06443578749895096, 0.01836664043366909, -0.06272726505994797, -0.016154710203409195, -0.006715875118970871, 0.020853154361248016, 0.0243007130920887, 0.08084982633590698, 0.017680179327726364, 0.024361731484532356, -0.07169701904058456, 0.04420807585120201, -0.00945027731359005, 0.01212...
4754bfa3c1ab57d4494af28ec0c82f71234b4bb9
subsection
40
73
Calibration of credible intervals
Hence, the sole negative summand - \beta ^{\infty ,(0)} P_{01}^{\infty } of eq:recursionP00- is in \Theta (h^{5 \wedge \frac{3p + 7}{2} }) and thereby of higher order than the remaining sum of positive summands:\underbrace{2hR}_{\in \Theta (h^{p+1})} \underbrace{\beta ^{\infty ,(0)}(nh)}_{\in \Theta (h), \text{ by {lem...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06152940168976784, 0.0070960246957838535, -0.05377718433737755, -0.0014602169394493103, 0.011223927140235901, 0.0021536052227020264, 0.037082452327013016, 0.07849881798028946, 0.007912448607385159, 0.004291950259357691, -0.06909849494695663, 0.016099583357572556, -0.022569935768842697, ...
a68c52de49ed0568ebe26fe0162de2b0bf174680
subsection
41
73
Numerical experiments
In this section, we empirically demonstrate thatthe worst-case convergence rates from theorem:GlobalTruncation hold not only for q=1 but also for q \in \lbrace 2,3,4\rbrace and are sometimes even outpaced in practice (see subsec:exporiginalWPD), the convergence rates of the credible intervals from theorem:contractiono...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.053243156522512436, 0.03182384744286537, 0.01204454805701971, -0.008222940377891064, -0.04857484623789787, 0.004542449954897165, -0.01494317315518856, -0.022060059010982513, -0.02225838601589203, 0.03203742951154709, -0.04158763587474823, 0.007631773594766855, 0.012784460559487343, 0.01...
96ad647e030fd12694d79b485ae4760c4444c907
subsection
42
73
Global Convergence Rates for
We consider two test IVPs with closed-form solutions which allow the computation of exact global errors: the logistic equation\dot{x}(t) = \lambda _0 x(t)\left( 1 - x(t)/ \lambda _1 \right) , \ \forall t \in [0,1.5], \qquad \textrm { with } (\lambda _0,\lambda _1) = (3,1) \textrm { and } x(0) = 0.1,which has the logist...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1606, "openalex_id": "", "raw": "E. Hairer, S. Nørsett, and G. Wanner, Solving Ordinary Differential Equations I – Nonstiff Problems, Springer, 1987.", "source_ref_id": "f65f59a9da7fd31acd639eaf5fc09d9ef90acc22", "star...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.07104521989822388, 0.015609806403517723, -0.03442397341132164, -0.008377110585570335, -0.00836948025971651, 0.01452642772346735, 0.0028534054290503263, -0.01760871522128582, 0.01149754598736763, 0.03884904086589813, -0.012428335845470428, -0.027694817632436752, 0.04660053551197052, 0.01...
dbf468c5955632bb996901ab8686f332a57fe3f2
subsection
43
73
Global Convergence Rates for
The (dash-)dotted gray lines visualize idealized convergence rates of orders one to five. The upper and lower rows employ the minimal R\equiv 0 and maximal measurement variance R \equiv K_R h^q (K_R = 1) which are permissible under ass:R.]In the case of R \equiv 0, even (q+1)th order convergence rates appear to hold tr...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.02492641657590866, -0.009519022889435291, -0.04048635810613632, -0.01348528265953064, -0.025262022390961647, 0.00012203875667182729, 0.01752781681716442, 0.002978508360683918, -0.017070170491933823, 0.059066757559776306, -0.041157569736242294, -0.01630742847919464, 0.04265254735946655, ...
cd77c3248dfbc6d57b1b860082d93904091a7d0f
subsection
44
73
Calibration of Credible Intervals
To demonstrate the convergence rates of the posterior credible intervals proved in theorem:contractionofcredibleintervals, we now restrict our attention to the case of q=1, that was considered therein. As in subsec:exporiginalWPD, we numerically solve the IVPs eq:exODElogistic,eq:exODElinear with the Gaussian ODE filte...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06807205826044083, 0.002289419062435627, -0.03296763449907303, 0.004765807185322046, -0.01680433563888073, -0.002407705644145608, 0.0011246771318838, -0.03895064815878868, -0.0024878354743123055, 0.05677759274840355, -0.032173968851566315, -0.027259349822998047, 0.03574546426534653, 0.0...
b33bd6a50852819b6afb2959d8f76bf52bef9a38
subsection
45
73
Necessity of ass:R
Having explored the asymptotic properties under ass:R in subsec:exporiginalWPD,subsec:exptheeffectofR, we now turn our attention to the question of whether this assumption is necessary to guarantee the convergence rates from theorem:GlobalTruncation,theorem:contractionofcredibleintervals. This question is of significan...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s11222-017-9798-7", "end": 515, "openalex_id": "https://openalex.org/W2534928453", "raw": "M. Schober, S. Särkkä, and P. Hennig, A probabilistic model for the numerical solution of initial value problems, Stat. Comp., 29 (2019), pp....
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.02669304981827736, 0.00318591995164752, -0.019809935241937637, 0.01215609721839428, -0.024480074644088745, -0.034034017473459244, -0.004147419240325689, -0.04334377497434616, 0.0014794496819376945, 0.0654429942369461, -0.04660981893539429, -0.029134951531887054, 0.022450242191553116, 0....
bfd75bafad632cf3060148216209f868a9242411
subsection
46
73
Discussion of experiments
Before proceeding to our overall conclusions, we close this section with a comprehensive discussion of the above experiments. First and foremost, the experiments in subsec:exporiginalWPD suggest that theorem:GlobalTruncation, the main result of this paper, may be generalizable to q \ge 2. Moreover, we demonstrated the ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s11222-017-9798-7", "end": 1094, "openalex_id": "https://openalex.org/W2534928453", "raw": "M. Schober, S. Särkkä, and P. Hennig, A probabilistic model for the numerical solution of initial value problems, Stat. Comp., 29 (2019), pp...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.0553765632212162, 0.04207398369908333, -0.020091166719794273, 0.007093692198395729, -0.022181136533617973, -0.023142216727137566, -0.015423059463500977, -0.042562153190374374, -0.0005978152039460838, 0.03603290393948555, -0.029747741296887398, -0.009328586980700493, 0.01128888688981533, ...
ba36ea7a7c430b023e898116dc0cf5deed7c2fe7
subsection
47
73
Discussion of experiments
As the observed convergence rates in practice sometimes outperform the proved worst-case convergence rates, we believe that an average-case analysis of the filter in the spirit of may shed more light upon the expected practical performance. Furthermore, it appears that the UQ becomes asymptotically accurate as well as ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/bfb0103934", "end": 241, "openalex_id": "https://openalex.org/W1598589417", "raw": "K. Ritter, Average-Case Analysis of Numerical Problems, vol. 1733 of Lecture Notes in Mathematics, Springer-Verlag, Berlin, 2000, https://doi.org/10...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06975873559713364, 0.023741161450743675, -0.025602614507079124, 0.012519038282334805, -0.01100851409137249, -0.036984946578741074, 0.028211701661348343, 0.009154689498245716, 0.013754921033978462, 0.055660512298345566, 0.0032422859221696854, 0.012869968079030514, 0.002818881534039974, 0...
68f743df521a361ecead6b6ee0f34d4345064d9b
subsection
48
73
Conclusions
We presented a worst-case convergence rate analysis of the Gaussian ODE filter, comprising both local and global convergence rates. While local convergence rates of h^{q+1} were shown to hold for all q \in \mathbb {N}, IBM and IOUP prior as well as any noise model R \ge 0, our global convergence results is restricted t...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1137/0704038", "end": 1323, "openalex_id": "https://openalex.org/W1977449329", "raw": "F. R. Loscalzo and T. D. Talbot, Spline function approximations for solutions of ordinary differential equations, SIAM J. Numer. Anal., 4 (1967), pp. ...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.05912185087800026, 0.008454700000584126, -0.03836663439869881, -0.0021880732383579016, 0.018053987994790077, -0.00853863637894392, -0.010789661668241024, -0.007664932869374752, 0.033025216311216354, 0.07368102669715881, -0.039465438574552536, 0.008286826312541962, 0.01690940000116825, -...
35b5db70f83f1beb4a98800ec3d1903fa6e32fff
subsection
49
73
Conclusions
To comprehend this idea, recall that the posterior filtering mean is the Bayes estimator with minimum mean squared error in linear dynamical systems with Gauss–Markov prior (as defined by the SDE SDE), i.e. when the data is not evaluations of f but real i.i.d. measurements, as well as in the special case of \dot{x}(t) ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 399, "openalex_id": "https://openalex.org/W2098626000", "raw": "E. Solak, R. Murray-smith, W. E. Leithead, D. J. Leith, and C. E. Rasmussen, Derivative observations in Gaussian process models of dynamic systems, in Advances in Neu...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.08167753368616104, -0.02391418069601059, -0.059762559831142426, -0.0023750122636556625, 0.025150330737233162, 0.00764582259580493, 0.033513426780700684, 0.018603309988975525, -0.023700524121522903, 0.05445168912410736, -0.056618768721818924, -0.004650827497243881, 0.015589237213134766, ...
4edc22614b6f831802478eafee26886ae027f2eb
subsection
50
73
Derivation of
As derived in the solution of the SDE SDE, i.e.\mathrm {d}{\normalcolor {X}(t)} = \begin{pmatrix} \mathrm {d}{\normalcolor X^{(0)}(t)} \\ \vdots \\ \mathrm {d}{\normalcolor X^{(q-1)}(t)} \\ \mathrm {d}{\normalcolor X^{(q)}(t)} \end{pmatrix} = \underbrace{\begin{pmatrix} 0 & 1 & 0 \dots & 0 \\ \vdots & \ddots & \ddots &...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1158, "openalex_id": "https://openalex.org/W2596317820", "raw": "S. Särkkä, Recursive Bayesian Inference on Stochastic Differential Equations, PhD thesis, Helsinki University of Technology, 2006.", "source_ref_id": "44dcdf47...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.02968604862689972, 0.01323363184928894, -0.05558887869119644, -0.03292008861899376, -0.0036192266270518303, 0.01589561253786087, 0.005480325315147638, 0.04799193516373634, 0.04024244472384453, 0.03938816860318184, 0.006818943191319704, 0.022028084844350815, 0.006201119627803564, 0.00948...
59dfb3e5fc49fd721083bef8db53bf9246741b27
subsection
51
73
Derivation of
By matrix:Q and{\normalcolor \begin{pmatrix} (Fh)^k \end{pmatrix}_{i,j} = h^k \left[ \mathbb {I}_{j-i=k} + (- \theta )^{k+i-q} \mathbb {I}_{\lbrace j=q,\ i+k\ge q\rbrace } \right], }it follows thatA(h)_{ij} &= \begin{pmatrix} \sum _{k=0}^{\infty } \frac{(hF)^k}{k!} \end{pmatrix}_{i,j} {\normalcolor = {\left\lbrace \b...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.04027235880494118, 0.009839269332587719, -0.01563604734838009, -0.056533847004175186, -0.015109761618077755, 0.006727314554154873, 0.00858838576823473, 0.014247871935367584, 0.006853165570646524, 0.02663467265665531, -0.02590244822204113, 0.001530235167592764, 0.006151450332254171, 0.01...
e47c50fb955dad91d3551450fad353cd3e67d532
subsection
52
73
Derivation of
\end{array}\right.} }Analogously, it follows that{\normalcolor \exp (F(t-\tau )) = {\left\lbrace \begin{array}{ll} \mathbb {I}_{i\le j} \frac{(t-\tau )^{j-i}}{(j-i)!}, & \mbox{if }j\ne q, \\ \frac{(t-\tau )^{q-i}}{(q-i)!} - \theta \sum _{k=q+1-i}^{\infty } \frac{(-\theta )^{k+i-q-1} (t-\tau )^k}{ k! } , & \mbox{if }j=...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06800337880849838, -0.0016930261626839638, -0.02855348400771618, -0.03726756200194359, -0.010545407421886921, 0.010430948808789253, -0.012384367175400257, 0.037664350122213364, 0.023608893156051636, 0.047034651041030884, -0.02727155201137066, 0.034276388585567474, -0.03858001530170441, ...
cd57c601eb0405ac11400f65f7258a287b57be0e
subsection
53
73
Extension to
The algorithm in Gaussian ODE filtering employs a prior X with independent dimensions {X}_j = \left( X_j^{(0)}, \dots , X_j^{(q)} \right)^ \intercal , j \in [d], by SDE. While this constitutes a loss of generality for our new theoretical results, which do not immediately carry over to the case of x with dependent dimen...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s11222-016-9671-0", "end": 2236, "openalex_id": "https://openalex.org/W2467245180", "raw": "P. R. Conrad, M. Girolami, S. Särkkä, A. Stuart, and K. Zygalakis, Statistical analysis of differential equations: introducing probability m...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.03505712375044823, 0.03643012046813965, -0.0673072412610054, 0.0032322576735168695, -0.0025724575389176607, 0.00017782184295356274, 0.013051074929535389, 0.008649865165352821, -0.007284498307853937, 0.04125085473060608, -0.045247793197631836, -0.017696373164653778, 0.026758136227726936, ...
8c4843c0a49831c0665a584ce31d6f47ea216e4a
subsection
54
73
Extension to
Even more flexible correlation models (over all modeled derivatives) can be employed by inserting arbitrary matrices (of the same dimensionality) for K_x \otimes F and {K_{\varepsilon }} \otimes L in eq:SDEwithdependentdimensions, but such models seem hard to interpret. For future research, it would be interesting to e...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.48550/arxiv.1801.04153", "end": 482, "openalex_id": "https://openalex.org/W2962978099", "raw": "X. Xiaoyue, F.-X. Briol, and M. Girolami, Bayesian quadrature for multiple related integrals, in International Conference on Machine Learning...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.04253145307302475, 0.03649040311574936, -0.019557146355509758, 0.021250471472740173, 0.0025323606096208096, -0.011593940667808056, 0.03453773632645607, -0.013050810433924198, 0.02919842302799225, -0.005652045831084251, -0.04067032411694527, -0.0016370719531551003, 0.017345145344734192, ...
c1b66a6cd2d9ec33f55c2ab83a90c93010e20cd8
subsection
55
73
Illustrative Example
To illustrate the algorithm defined in Gaussian ODE filtering, we apply it to a special case of the Riccati equation \frac{\mathrm {d}x}{\mathrm {d}t} (t) = f(x(t)) = - \frac{(x(t))^3}{2}, \quad x(0) = 1, \qquad \qquad \left(\textrm {solution: } x(t)= (t+1)^{-1/2} \right),with step size h=0.1, measurement noise R=0.0 (...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.2307/2311950", "end": 676, "openalex_id": "https://openalex.org/W119916070", "raw": "H. T. Davis, Introduction to Nonlinear Differential and Integral Equations, Dover Publications, New York, 1962.", "source_ref_id": "59df08ca3b5bfe...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.07049474120140076, 0.008232015185058117, -0.06378095597028732, 0.012496794573962688, 0.0002906286681536585, 0.0057334196753799915, 0.013068992644548416, 0.015281490050256252, -0.00873554963618517, 0.046294596046209335, -0.052245453000068665, -0.02601591683924198, 0.018905406817793846, 0...
7f212c44b11dbe6cca5682349014b4cd88bf99d9
subsection
56
73
Illustrative Example
In the subsequent update step defbetaMM,eq:defyMAP,def:r,eq:defpredictivemean, a Bayesian conditioning of the predictive distribution concreteexampleprediction:mean,concreteexampleprediction:covariance on this data is executed:{\beta } (h) &= \left( \beta ^{(0)}(h), \beta ^{(1)}(h) \right)^{\intercal } = \left( \frac{P...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.027153288945555687, -0.032327525317668915, -0.026313811540603638, -0.00039827494765631855, 0.004060019738972187, -0.007375956512987614, 0.036326490342617035, 0.048720233142375946, -0.017171135172247887, 0.026023808866739273, -0.06434978544712067, 0.023154322057962418, -0.03760860115289688...
9b8ed355dc7a9f950776d8b3b26f5ce4b5f9be9a
subsection
57
73
The role of the state misalignments
Recall that, as in concreteexampleprediction:mean, the next step h \rightarrow 2h starts with computing m^{-,(i)}(2h) by a (q-i)th-order Taylor expansion of the ith state m^{(i)}(h), for all i \in [q+1], and note that—unlike in the previous step—there is now a non-zero state misalignment (recall def:deltai):\delta ^{(1...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.05363832786679268, -0.018718495965003967, -0.04482065141201019, 0.0029595664236694574, 0.006064060144126415, 0.035301219671964645, -0.014416445046663284, 0.0359419509768486, -0.014538489282131195, 0.017986230552196503, -0.031395815312862396, 0.006704790983349085, -0.03173143416643143, 0...
786125a2af5dc7759087f7d189f6158cfb384c5c
subsection
58
73
The role of the state misalignments
Moreover, for the IOUP prior, this effect can (in the worst case) become more pronounced since its predictions additionally deviate from the Taylor expansion of the state by adding a drift on the qth derivative—which can be seen by inserting def:AIOUP with \theta > 0, instead of \theta = 0, into the mean prediction con...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/bfb0103934", "end": 534, "openalex_id": "https://openalex.org/W1598589417", "raw": "K. Ritter, Average-Case Analysis of Numerical Problems, vol. 1733 of Lecture Notes in Mathematics, Springer-Verlag, Berlin, 2000, https://doi.org/10...
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.06792523711919785, 0.010222351178526878, -0.03802104666829109, 0.019163094460964203, -0.01219053566455841, -0.02602885290980339, -0.028058066964149475, 0.07445533573627472, -0.023252036422491074, 0.06185285747051239, -0.039058536291122437, -0.010344409383833408, -0.05751980096101761, -0...
7ca07284efa9f42e68046d03a5ef564998e7ff4a
subsection
59
73
Proof of eq:Phii=fi-1
We prove the stronger statement\Phi _t^{(i+1)}(a) = f^{(i)} \left( \Phi _t^{(0)}(a) \right),from which eq:Phii=fi-1 follows by inserting t=0 and \Phi _0^{(0)}(a) = a. Hence, it remains to show eq:appendixmoregeneral.[Proof of eq:appendixmoregeneral] By induction over i \in \lbrace 0,\dots ,q\rbrace . The base case (i=0...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.020512551069259644, 0.02286294847726822, -0.024892836809158325, -0.010958341881632805, 0.02652590349316597, -0.017048005014657974, 0.032875027507543564, 0.017902696505188942, 0.04075038060545921, 0.005147215910255909, -0.045756421983242035, 0.0058912537060678005, -0.03394338861107826, 0...
9d94747bb1c0d366b5bf259c2cc1b69910e06876
subsection
60
73
Proof of lemma:r
Again, w.l.o.g. d=1. Recall that, by def:r, r is implied by the values of m^{-,(0)} and m^{-,(1)}.
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.041805751621723175, 0.04659663140773773, -0.02711270935833454, -0.010680302046239376, 0.0015486438060179353, -0.014090369455516338, 0.04287378489971161, 0.01945340633392334, -0.042446572333574295, -0.016539210453629494, -0.00039955772808752954, -0.05663611367344856, -0.009070627391338348,...
f752254af0dab1c24922ce62c7077a9748847aec
subsection
61
73
Proof of lemma:r
By insertion of m^{-,(i)}((n+1)h)= \sum _{k=i}^q \frac{h^{k-i}}{(k-i)!} m^{(k)}(nh) + K \theta \left|m^{(q)}(nh) \right|h^{q+1-i} (due to Aprediction,def:Psi) into the definition def:r of r((n+1)h), we obtain the following equality which we then bound by repeated application of the triangle inequality:\left|r((n+1)h) \...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.04575177654623985, 0.013391321524977684, -0.022311238572001457, -0.03909808024764061, -0.02580595389008522, 0.043126922100782394, 0.024356182664632797, 0.03915912285447121, 0.036747924983501434, 0.03216969221830368, -0.058448731899261475, 0.018358701840043068, -0.021273506805300713, 0.0...
068c51e552f7eb6e10ef2f7ce4fb806b67c3c8ce
subsection
62
73
Proof of lemma:sequenceofcontractions
Let \tilde{u}_0 = u^{\ast } and \tilde{u}_n = T_n(\tilde{u}_{n-1}), for n \in \mathbb {N}. Then,d(u^{\ast },x_n) \le \underbrace{d(u^{\ast },u_n)}_{\rightarrow 0} + \underbrace{d(u_n,\tilde{u}_n)}_{a_n} + \underbrace{d(\tilde{u}_n,x_n)}_{= d\left((T_n \circ \dots \circ T_1) (u^{\ast }), (T_n \circ \dots \circ T_1)(x_0)...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.0437714159488678, 0.03259963169693947, -0.036445654928684235, -0.01567407324910164, -0.0043344078585505486, 0.0460607148706913, 0.014155504293739796, 0.028387319296598434, 0.011560964398086071, 0.02174834907054901, -0.04441241919994354, 0.024678653106093407, 0.001443212851881981, 0.0013...
840a2800db61b07024913d87055508a1ab97e239
subsection
63
73
Proof of proposition:steadystatesglobal
Again, w.l.o.g. d=1. We prove the claims in the following order: lemma:steadystatesvaluesi, ineq:steadystatelemmai, lemma:steadystatesvaluesii, ineq:steadystatelemmaii, lemma:steadystatesvaluesiii, lemma:steadystatesvaluesv, lemma:steadystatesvaluesvi, lemma:steadystatesvaluesiv, ineq:ineq:steadystatelemmavi, ineq:ineq...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.008982251398265362, 0.0744832381606102, -0.028785940259695053, -0.03260168060660362, -0.005322956945747137, 0.009104354307055473, 0.017933975905179977, 0.012775096110999584, -0.009852239862084389, 0.0030793019104748964, -0.007837529294192791, 0.0003963599447160959, -0.004533098544925451, ...
4da5c34a6ee88b84773569dec0ab4b57f6cf80f7
subsection
64
73
Proof of proposition:steadystatesglobal
As a start, for lemma:steadystatesvaluesi, we show that P_{11}^{-,\infty } is indeed the unique fixed point of the recursion for \lbrace P_{11}^{-}(nh) \rbrace _n by checking that, if P_{11}^-(nh) = \frac{1}{2} \left(\sigma ^2 h + \sqrt{4\sigma ^2 R h + \sigma ^4 h^2} \right), then also P_{11}^-((n+1)h) = \frac{1}{2} \...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.020311802625656128, 0.027118010446429253, -0.007882078178226948, -0.02740796096622944, -0.011163097806274891, -0.0167560912668705, 0.05209954455494881, 0.05994346737861633, 0.029605481773614883, 0.04825388267636299, -0.04532385617494583, -0.000923740619327873, -0.015306338667869568, 0.0...
21a5eab608d5e860227eec9a48dac3e0f02af684
subsection
65
73
Proof of proposition:steadystatesglobal
Since, by eq:CupdateMM, def:AIOUP with \theta = 0 and ass:assumption2,P_{11}^-(h) = P_{11}(0) + \sigma ^2 h \le Kh,we can, using the reverse triangle inequality and the (by BFT) strictly decreasing sequence \lbrace \vert P_{11}^{-}(nh) - P_{11}^{-,\infty } \vert \rbrace _n , derive ineq:steadystatelemmai:\left|P_{11}^-...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.0035589609760791063, 0.01603630557656288, -0.022932983934879303, -0.0018986847717314959, -0.015540415421128273, -0.014037488959729671, 0.024214668199419975, 0.04836830124258995, 0.02679329365491867, 0.02456560544669628, -0.061154622584581375, 0.004455376882106066, -0.029265111312270164, ...
7c4ef3092c1242fc42385540721fbfdd6a927ff5
subsection
66
73
Proof of proposition:steadystatesglobal
Since P_{11}(nh) monotonically increases in P_{11}^-(nh) (because the derivative of P_{11}(nh) with respect to P_{11}^-(nh) is non-negative for all P_{11}^-(nh) due to R\ge 0; see eq:P11expressedbyP11-), we obtain ineq:steadystatelemmaii:P_{11}(nh) &\stackrel{{eq:P_11_expressed_by_P_11^-}}{\le } \frac{\left( \max _n P_...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.011118133552372456, 0.029867107048630714, -0.01880238950252533, -0.032293714582920074, -0.007203512825071812, -0.011392843909561634, 0.03108804300427437, 0.03748268634080887, 0.013735512271523476, 0.017581455409526825, -0.05997840315103531, -0.019260240718722343, -0.05512518808245659, 0...
8dd5ef95ecd0f0da3052d03ef9fb067020063aa5
subsection
67
73
Proof of proposition:steadystatesglobal
Hence, insertion of the limits lemma:steadystatesvaluesi,lemma:steadystatesvaluesii yield\lim _{n \rightarrow \infty } \left( 1 - \alpha (nh) \right)^{-1} &= \frac{\sigma ^2 h + \sqrt{4\sigma ^2 R h + \sigma ^4 h^2} + 2R}{\sigma ^2 h + \sqrt{4\sigma ^2 R h + \sigma ^4 h^2}}, \qquad \text{and} \\ \lim _{n \rightarrow \i...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.035282913595438004, 0.007344248704612255, -0.029499081894755363, -0.012078904546797276, 0.010484153404831886, 0.0015222624642774463, 0.01880127750337124, 0.05667852237820625, 0.028919171541929245, 0.029438037425279617, -0.06708637624979019, 0.022921686992049217, -0.011628711596131325, 0...
ea85e3abbc19e892fa075d492f31a6c0bea2ada4
subsection
68
73
Proof of proposition:steadystatesglobal
Next, recall thatP_{01}(nh) \stackrel{{eq:C_update_MM}}{=} \left( 1-\frac{P_{11}^{-}(nh)}{P_{11}^{-}(nh) + R} \right) P_{01}^{-}(nh) = R \frac{P_{01}^{-}(nh)}{P_{11}^{-}(nh) + R} \stackrel{{def_beta_MM}}{=} R \beta ^{(0)}(nh),which, since P_{01}(nh) depends continuously on \beta ^{(0)}(nh), implies the unique (attracti...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.032284997403621674, 0.016081467270851135, -0.016417134553194046, -0.014029325917363167, 0.011000700294971466, -0.024229004979133606, 0.025495382025837898, 0.04696276783943176, 0.030423574149608612, 0.019087206572294235, -0.05507979169487953, 0.012877380475401878, -0.02134532667696476, 0...
1e2515935812340a9f7ccc8041cb9402b28b80e8
subsection
69
73
Proof of proposition:steadystatesglobal
The basis (n=1) follows from\left|\beta ^{(0)}(h) \right|= \frac{\left|P_{01}^-(h) \right|}{P_{11}^-(h) + R} \stackrel{{eq:C^-_predict}}{\le } \frac{ \left|P_{01}(0) \right|+ h P_{11}(0) + \frac{\sigma ^2}{2} h^2 }{\sigma ^2 h} \stackrel{\text{Ass. }\ref {ass:assumption2}}{\le } \left( \frac{2K_0}{\sigma ^2} + \frac{1}...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.014726300723850727, 0.003277174197137356, -0.01092645712196827, 0.044926661998033524, -0.020296351984143257, -0.04910801723599434, 0.04159989207983017, 0.0318637490272522, 0.03323718532919884, 0.021974997594952583, -0.06610812246799469, 0.032382600009441376, -0.02542385272681713, 0.0116...
2c9d841eb0ad14847e4ae8acd79d4e4f526cff35
subsection
70
73
Proof of proposition:steadystatesglobal
P_{01}^- \le \hat{\beta }( P^-_{11} + R )) yields, after some rearrangements, that\left|\beta ^{(0)}(nh) \right|&\le \frac{\hat{\beta }\left( P^-_{11} + R \right) h \alpha (nh) + hP^-_{11} \alpha (nh) + \frac{\sigma ^2}{2} h^2}{P^-_{11}\alpha (nh) + \sigma ^2 h + R} \\ &= \frac{2 \hat{\beta }P^-_{11} R + \sigma ^2 h \l...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.009434930980205536, 0.026970021426677704, -0.017756281420588493, -0.002852598438039422, -0.0003806483873631805, -0.029654821380972862, 0.05247560888528824, 0.07676083594560623, 0.03935670480132103, 0.013492638245224953, -0.03484136238694191, 0.02971583977341652, -0.05159084498882294, 0....
90e04280db14c0c9bbfea965daf1d51105fc3419
subsection
71
73
Proof of lemma:deltaglobalbound
For all n \in [T/h + 1], we can estimate\delta ^{(1)}(nh) &= \left\Vert m^{(1)}\left(nh\right) - f\left( m^{(0)}\left(nh\right) \right) \right\Vert = \left\Vert \Psi _{\normalcolor h}^{(1)}({m}((n-1)h) - f\left( m^{(0)}\left(nh\right) \right) \right\Vert \\ &\le \underbrace{\left\Vert \Psi _{\normalcolor h}^{(1)}({m}((...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.03970131650567055, 0.024077123031020164, -0.04086092486977577, -0.000001072828126780223, -0.0047261654399335384, -0.01924033835530281, 0.0005178183782845736, 0.02957000397145748, 0.06750138849020004, 0.037626225501298904, -0.056363049894571304, 0.008201174437999725, 0.00915479939430952, ...
64c7d2ea57c69d9a2d68ef6cd5c1ecdd8acbecf7
subsection
72
73
Proof of lemma:deltaglobalbound
For the inductive step, we distinguish two cases. If \delta ^{(1)}((n-1)h) \le \delta ^{\infty }, then \bar{T}(\delta ^{(1)}((n-1)h)) < 2 \delta ^{\infty }, since\bar{T}(\delta ^{(1)}((n-1)h)) - \delta ^{\infty } \le \left|\delta ^{\infty } - \bar{T}(\delta ^{(1)}((n-1)h)) \right|< \delta ^{\infty } - \underbrace{\delt...
{ "cite_spans": [] }
1807.09737
Convergence Rates of Gaussian ODE Filters
[ "Hans Kersting", "T. J. Sullivan", "Philipp Hennig" ]
[ "math.NA", "cs.LG", "cs.NA", "math.ST", "stat.CO", "stat.ML", "stat.TH" ]
2,018
en
Mathematics
[ -0.042904943227767944, 0.02511434443295002, -0.019667308777570724, 0.020750612020492554, -0.008254471234977245, 0.0008901974651962519, 0.01718028634786606, 0.0449800044298172, 0.03460469841957092, 0.029142403975129128, -0.024809187278151512, 0.016188528388738632, -0.009894685819745064, -0....
f38402d8c2a180faa068b555b8d3c63d7b9e5ad7
abstract
0
60
Abstract
Raziel combines secure multi-party computation and proof-carrying code to provide privacy, correctness and verifiability guarantees for smart contracts on blockchains. Effectively solving DAO and Gyges attacks, this paper describes an implementation and presents examples to demonstrate its practical viability (e.g., pr...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.033898353576660156, 0.029474176466464996, 0.015004063956439495, 0.012906393967568874, -0.0227463748306036, 0.013661555014550686, 0.06419634073972702, -0.0045385961420834064, -0.030816685408353806, 0.015164249576628208, -0.01893242821097374, 0.014615042135119438, -0.028101155534386635, -...
e31ad91b4c0e8ee0fbd410a0bdfc9af72fc55e4d
subsection
1
60
Introduction
The growing demand for blockchain and smart contract, , , , , technologies sets the challenge of protecting them from intellectual property theft and other attacks: security, confidentiality and privacy are the key issues holding back their adoption. As shown in this paper, the solution must be inter-disciplinary (i.e...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 251, "openalex_id": "", "raw": "Nick Szabo. The Idea of Smart Contracts. Nick Szabo's Papers and Concise Tutorials, 1997. https://web.archive.org/web/20150812055200/http://szabo.best.vwh.net/idea.html.", "source_ref_id": "99...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.03311898931860924, 0.021489188075065613, -0.011774793267250061, -0.046671830117702484, -0.013903871178627014, -0.015445354394614697, -0.02258806675672531, -0.0025983909144997597, 0.0000969984830589965, 0.009905172511935234, 0.009966221638023853, -0.018879350274801254, 0.006753527093678713...
c2685398b38e0e214e58f6b7ff4d4f4c49ab61d3
subsection
2
60
Contributions
We propose a system for securely computing smart contracts guaranteeing their privacy, correctness and verifiability. Our main and novel contributions are:Practical formal verification of smart contracts: the proofs accompanying the smart contracts can be used to prove functional correctness of a computation as well as...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 620, "openalex_id": "", "raw": "Phil Daian. Analysis of the DAO exploit. Hacking Distributed, 2016. http://hackingdistributed.com/2016/06/18/analysis-of-the-dao-exploit.", "source_ref_id": "0930a560fee456dc4fc2f98e02a8360d01...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.05167783051729202, 0.04600029066205025, -0.00632618460804224, -0.008295009844005108, 0.003933833912014961, -0.007989765144884586, 0.07014510035514832, 0.007379277143627405, -0.004964032676070929, 0.024907922372221947, -0.01578112132847309, 0.00015047582564875484, -0.01817728765308857, -...
7077cd128cf266a3027f707038b9956e1d58f0cb
subsection
3
60
Minimal set of functionalities
We argue that the combined features here described are indeed the minimal set of functionalities that must be offered by a secure solution to protect computations on blockchains:A pervasive goal of blockchain technologies is the removal of trusted third parties: towards this end, the preferred solution in cryptography ...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.029854467138648033, 0.022623492404818535, -0.015522187575697899, 0.0001620867260498926, 5.959070676908595e-7, 0.01989280804991722, 0.023462530225515366, 0.006361427251249552, 0.008268330246210098, 0.01855034939944744, -0.028145883232355118, 0.006220316514372826, 0.00016590053564868867, ...
d1e552d4c62d2940ea88b017e30ce8c8c9911011
subsection
4
60
Background
This section provides a brief introduction to the main technologies that underpin Raziel: blockchains, secure multi-party computation and formal verification.
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 158, "openalex_id": "", "raw": "Shafi Goldwasser. Multi-Party Computations: Past and Present, 1997. https://groups.csail.mit.edu/cis/pubs/shafi/1997-podc.pdf.", "source_ref_id": "73f6f8f773df4d8e2c054cdcd845c21c7a323c70", ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.00900097656995058, 0.04833066463470459, -0.010564705356955528, 0.020427638664841652, -0.055134791880846024, -0.004302161745727062, 0.03762102872133255, 0.032098397612571716, -0.017315438017249107, 0.007826272398233414, -0.018276559188961983, -0.011663129553198814, -0.013966768980026245, ...
ebfe9afd75b6917c94b2f8eee0cd3f09561222e6
subsection
5
60
Blockchains
A blockchain is a distributed ledger that stores a growing list of unmodifiable records called blocks that are linked to previous blocks. Blockchains can be used to make online secure transactions, authenticated by the collaboration of the P2P nodes allowing participants to verify and audit transactions. Blockchains ca...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.2139/ssrn.3440802", "end": 576, "openalex_id": "https://openalex.org/W4248175462", "raw": "Satoshi Nakamoto. Bitcoin: A Peer-to-Peer Electronic Cash System, 2008. https://bitcoin.org/bitcoin.pdf.", "source_ref_id": "9f02459b5e414ba...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.004008457064628601, -0.0036385043058544397, -0.03646131977438927, 0.021175026893615723, 0.00022907511447556317, -0.01585075818002224, 0.011724068783223629, 0.005099245347082615, 0.006148080341517925, -0.004939059726893902, -0.040366802364587784, 0.004874222446233034, -0.01340220496058464,...
ea94aa1d6135ba7948f00759dc25da5e09e883b0
subsection
6
60
Secure Multi-Party Computation
Protocols for secure multi-party computation (MPC) enable multiple parties to jointly compute a function over inputs without disclosing said inputs (i.e., secure distributed computation). MPC protocols usually aim to at least satisfy the conditions of inputs privacy (i.e., the only information that can be inferred abou...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/359168.359176", "end": 1185, "openalex_id": "https://openalex.org/W2141420453", "raw": "Adi Shamir. How to Share a Secret, 1979. https://cs.jhu.edu/~sdoshi/crypto/papers/shamirturing.pdf.", "source_ref_id": "8f95a4706a947c85b6...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.05424545705318451, 0.020471710711717606, -0.04384180158376694, 0.02202768251299858, -0.004408587701618671, -0.00004957754208589904, 0.05940152332186699, -0.02930414117872715, 0.0275346040725708, 0.00941210612654686, -0.04893684759736061, -0.008222244679927826, -0.0023968834429979324, -0...
cf3b55b7ee453ddff899098190a26f859e9dddc2
subsection
7
60
Formal verification
Formal verification uses formal methods of mathematics on software to prove or disprove its correctness with respect to certain formal specifications. Deductive verification is the preferred approach: smart contracts are annotated to generate proof obligations that are proved using theorem provers or satisfiability mod...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 705, "openalex_id": "", "raw": "Phil Daian. Analysis of the DAO exploit. Hacking Distributed, 2016. http://hackingdistributed.com/2016/06/18/analysis-of-the-dao-exploit.", "source_ref_id": "0930a560fee456dc4fc2f98e02a8360d01...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.0592990480363369, -0.000768228608649224, -0.010491018183529377, -0.00927024520933628, -0.020585283637046814, -0.009743294678628445, 0.036104362457990646, 0.02195865474641323, 0.014008370228111744, 0.007568792439997196, -0.0232252050191164, -0.025971945375204086, 0.012207729741930962, 0....
a503446cf85c1f9632d7f4f374ad9461c0751048
subsection
8
60
Model and Goals
Parties executing smart contracts need to protect their private financial information and obtain formal guarantees regarding their execution. The central goal for Raziel is to offer a programming framework to facilitate the development of formally-verified, privacy-preserving smart contracts with secure computation.
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.05136777088046074, 0.029178602620959282, -0.004780437797307968, -0.015962772071361542, -0.021502433344721794, -0.0016662859125062823, 0.07349589467048645, -0.01367365475744009, -0.016100117936730385, -0.0028098903130739927, 0.012429901398718357, -0.012613031081855297, -0.06806305795907974...
cc4b9a9e3cf078ab2cba72491d029cebefd00764
subsection
9
60
Threat Model and Assumptions
A conservative threat model assumes that parties wish to execute smart contracts but mutually distrust each another. Each party is potentially malicious and the smart contract is developed by one of the parties or externally developed. We assume that each party trusts its own environment and the blockchain; the rest of...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.06206044182181358, 0.007772818673402071, -0.01964382454752922, -0.0027244931552559137, -0.015492215752601624, -0.027443353086709976, 0.03721184656023979, 0.012126665562391281, -0.007528606336563826, -0.009135065600275993, -0.008333743549883366, 0.009516647085547447, -0.05000246316194534, ...
f48e5d00e28afcccda27c2a13a633383e0a1e496
subsection
10
60
Goals
A secure smart contract system should operate as follows: a restricted set of parties willing to execute a smart contract check the accompanying proofs/certificates to verify it before its execution. Then these parties send their private inputs to the nodes at the start of the execution of the smart contract; after tha...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.017046906054019928, 0.030446352437138557, -0.028599731624126434, 0.011659655719995499, 0.0026230409275740385, 0.021442167460918427, 0.01936662755906582, -0.0076306648552417755, 0.0182067658752203, 0.010743976570665836, -0.003140018554404378, 0.02136586233973503, -0.012514290399849415, 0...
492f67a5d176d87f47667df730c287af7812018d
subsection
11
60
Private Smart Contracts
Enabling smart contracts with secure computation techniques (e.g., secure multi-party computation, homomorphic encryption, indistinguishability obfuscation) is a key-step to the global adoption of blockchain technologies: encrypted transactions could be stored on the blockchain; secure computations could be carried out...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 415, "openalex_id": "https://openalex.org/W1568804695", "raw": "R L Rivest, L Adleman, and M L Dertouzos. On Data Banks and Privacy Homomorphisms. Foundations of Secure Computation, Academia Press, pages 169–179, 1978. people.csai...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.04637638479471207, 0.036918044090270996, -0.021555867046117783, 0.028756409883499146, -0.008733711205422878, -0.02294410765171051, 0.004969443194568157, -0.014706196263432503, 0.033531349152326584, 0.03237193822860718, -0.014759589917957783, 0.005572031252086163, 0.00397021509706974, 0....
1bbf0ae7ac74f0d76611050bebe16292127cb977
subsection
12
60
Off-chain computation
Even when using the fastest available secure multi-party computation techniques, the overhead would be very significant if secure computations would have to be executed on every full node of a blockchain, as have been previously proposed. In Ethereum, at current prices (380 $/ETH, 21 Gwei/gas, 30/August/2017), multiply...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 238, "openalex_id": "", "raw": "Vitalik Buterin. Secret Sharing DAOs: The Other Crypto 2.0, 2014. https://blog.ethereum.org/2014/12/26/secret-sharing-daos-crypto-2-0/.", "source_ref_id": "19c994aef0f5b6811d3e9dc06925317ccef0...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ 0.02144249714910984, 0.020755726844072342, -0.0007788166403770447, 0.03836756944656372, 0.010728878900408745, -0.026463551446795464, 0.0022911422420293093, -0.038153909146785736, -0.009691093116998672, 0.009683461859822273, -0.018680153414607048, 0.012338974513113499, -0.01313257496803999, ...
73570ba9867e74ea158b37190c0954df7ce88a79
subsection
13
60
Latency and its impact
As previously mentioned, MPC protocols can be roughly divided into two classes: constant-round protocols, ideal for high latency settings; and protocols with rounds dependent on the depth of the evaluated circuit, usually faster but only on very low latency settings. The present paper proposes the use of two protocol s...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2976749.2978331", "end": 581, "openalex_id": "https://openalex.org/W2536058570", "raw": "Toshinori Araki, Jun Furukawa, Yehuda Lindell, Ariel Nof, and Kazuma Ohara. High-throughput semi-honest secure three-party computation with an ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.05803157389163971, 0.010701669380068779, -0.04021325707435608, 0.02983957901597023, 0.006800098344683647, 0.027810608968138695, 0.024866316467523575, -0.009115106426179409, 0.020472757518291473, 0.031639717519283295, -0.033500876277685165, 0.00401980010792613, 0.012753521092236042, -0.0...
a901c3b2766659cd0ae378f4b879eaba318a4732
subsection
14
60
Blockchain Solutions
The guarantees of privacy, correctness and verifiability are designed for the more threatening setting of public permissionless blockchains, although the smart contracts can also be used on private permissioned blockchains: it's also preferred that private blockchains keep their communications open to public blockchain...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 658, "openalex_id": "", "raw": "DEDIS. Cothority, 2019. https://github.com/dedis/cothority/.", "source_ref_id": "5468a5654f0b1dfeb963b190ed1b4e71a3b1668a", "start": 376 }, { "arxiv_id": "", "doi": "...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ 0.0037768702022731304, 0.025835318490862846, -0.04745885357260704, 0.010140324011445045, 0.012108874507248402, -0.013276270590722561, 0.026369420811533928, -0.03656315430998802, 0.005100682377815247, 0.03607483208179474, -0.01012506429105997, 0.02111995220184326, 0.013558582402765751, -0.0...
863cce7d916744e738b8b3bda5f9f927c238ded5
subsection
15
60
Alternative Protocols and Standards
Although this paper is focused on blockchains, it could be adopted to other financial standards such as:Financial Information eXchange: messaging standard for trade communication in the equity markets, with presence in the foreign exchange, fixed income and derivatives market. Financial Products Markup Language (FpML)...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 277, "openalex_id": "", "raw": "FIX Trading Community. FIX Protocol Application Layer, 2016. http://www.fixtradingcommunity.org/pg/structure/tech-specs/fix-protocol.", "source_ref_id": "7f79df3e32ed2d2c84bc11e76421528694d00d...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.040467701852321625, -0.0034028273075819016, -0.0209510400891304, 0.007343545090407133, -0.005676465108990669, -0.05459782853722572, -0.017090432345867157, 0.018829993903636932, -0.025071054697036743, -0.015549241565167904, -0.024277569726109505, 0.023331493139266968, -0.021942894905805588...
120813bcbe2d7b587c90f54c6c0d773c83b1364e
subsection
16
60
Functionality and Protocol
The functionalities and protocols of this sub-section and section constitute an open framework on which to instantiate different secure multi-party computation protocols, thus benefiting from upcoming research advances.In this sub-section we present our secure protocol for private smart contracts, consisting of the fo...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.019332533702254295, 0.033995963633060455, -0.032378558069467545, 0.026031019166111946, 0.01879848539829254, 0.0010776323033496737, 0.03552181273698807, -0.013686884194612503, 0.01259590033441782, 0.01867641694843769, -0.012252584099769592, 0.045928116887807846, 0.013793693855404854, 0.0...
d5cd0633a474b7f3af6c5314b9ac45e1afd72bac
subsection
17
60
Functionality and Protocol
Output: results from the secure computation E_{1}:\overrightarrow{r_{1}},E_{2}:\overrightarrow{r_{2}},...,E_{N}:\overrightarrow{r_{N}}.The security of the protocol is proved on REF .
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ 0.0019181109964847565, 0.028950028121471405, -0.002926287706941366, 0.015245691873133183, -0.000859382445923984, 0.012376632541418076, 0.01642078533768654, 0.04270014539361, -0.0006576516898348927, 0.014307143166661263, -0.02542474865913391, 0.01829788275063038, -0.017702706158161163, -0.0...
e3d8d5d91b4737d1d02a9ffc7b3c43b5b2bd7781
subsection
18
60
Experimental Results
Table REF summarizes the execution cost for several example applications, chosen for their economic significance:Millionaire's Problem (i.e., determining who's got the bigger number without revealing anything else) Second-price auction: sealed-bid auction not revealing the bids between the participants and without an ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1111/j.1540-6261.1961.tb02789.x", "end": 403, "openalex_id": "https://openalex.org/W2015007620", "raw": "William Vickrey. Counterspeculation, auctions, and competitive sealed tenders. The Journal of Finance, 16(1):8–37, 1961. http://dx.d...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.011284006759524345, 0.0027237257454544306, -0.0210726335644722, 0.06646806001663208, -0.00724038016051054, 0.052643436938524246, 0.03689618408679962, 0.017563071101903915, 0.017303669825196266, 0.026459049433469772, -0.004066514782607555, 0.020004505291581154, -0.015022452920675278, -0....
b870e44dca6e84c6a9f630bccc79787b032a67b5
subsection
19
60
Experimental Results
Suppose two risky currencies with different interest rates but constant exchange rate: we calculate the calls and puts with the following equations, Call & = & S_{0}e^{-\rho t}N\left(d_{1}\right)-X^{-rt}N\left(d_{2}\right)\\ Put & = & Xe^{-rt}N\left(-d_{2}\right)-S_{0}e^{-\rho t}N\left(-d_{1}\right)\\ d_{1} & = & \fra...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1949, "openalex_id": "https://openalex.org/W2400524050", "raw": "Charanjit S. Jutla. Upending stock market structure using secure multi-party computation. Cryptology ePrint Archive, Report 2015/550, 2015. http://eprint.iacr.org/20...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.017888538539409637, 0.015858525410294533, -0.0194453913718462, 0.0388602577149868, -0.003191929077729583, 0.01570589281618595, 0.03422022610902786, 0.011928237043321133, 0.0031633104663342237, 0.04078342765569687, -0.0005747571121901274, 0.02620701491832733, -0.0011180336587131023, -0.0...
f177892b5c006971eb7acf623ae98285bb9e1f78
subsection
20
60
Modes of Interaction
Secure computation can be carried on the nodes of the blockchain or on the parties themselves. Additionally, said secure computations can be outsourced to the cloud (see REF ) or use mined pre-processing data for secure multi-party computation (see REF ).
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.013787997886538506, 0.015245390124619007, -0.043798066675662994, 0.030048221349716187, -0.012254302389919758, 0.011025819927453995, -0.003145601600408554, 0.03152850642800331, 0.025485286489129066, -0.000046139623009366915, -0.03357343375682831, 0.024768035858869553, -0.02023562416434288,...
c96fdef5009d1f93b56760bcb695577abe7d1d3b
subsection
21
60
Outsourcing Secure Computations for Cloud-based Blockchains
The following scheme makes use of a multi-party non-interactive key exchange, to establish a shared secret between the computing parties and the executing nodes of the blockchain: replacing the NIKE protocol would require a PKI infrastructure between the computing parties and the executing nodes that will be used to es...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/978-3-662-44371-2_27", "end": 447, "openalex_id": "https://openalex.org/W288529411", "raw": "Dan Boneh and Mark Zhandry. Multiparty key exchange, efficient traitor tracing, and more from indistinguishability obfuscation. Cryptology ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.03341497480869293, 0.01096285879611969, -0.02873077616095543, 0.04925275966525078, -0.0006074582925066352, 0.02052198350429535, 0.021742621436715126, 0.010802649892866611, 0.021803652867674828, 0.02239871397614479, 0.0102762496098876, 0.023146355524659157, -0.042691826820373535, 0.03991...
dba1980fa3c26b919dfa201e4ae1ee129afbdc14
subsection
22
60
Outsourcing Secure Computations for Cloud-based Blockchains
The following protocol uses a pivot table during the secure computation, allowing the evaluator node to obliviously map the encoded inputs by the parties to the encoding expected by the circuit created by the garbling node. To implement Functionality B.1 (Outsourcing for Cloud-based Blockchains), the following protocol...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.05538938567042351, 0.029190052300691605, -0.018661491572856903, 0.039337143301963806, 0.004428862128406763, 0.015113824978470802, 0.033630359917879105, -0.012565608136355877, -0.0029678333085030317, 0.046020492911338806, -0.007369992323219776, 0.008644101209938526, 0.013114924542605877, ...
e6d5651108b8c164ba4f8a43d5da0e1d510293cd
subsection
23
60
Outsourcing Secure Computations for Cloud-based Blockchains
Compute and save garbled inputs: using the pivot keys, encrypt Enc_{s_{jl}^{0}}\left(w_{jl}^{0}\right) and Enc_{s_{jl}^{1}}\left(w_{jl}^{1}\right); then save them into pivot table P_{q}\left[j,l\right] in random order. Secure computation at node N_{E}: for every bit of x_{j} and using the encoding X_{jl}^{x_{j}\left[...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.03316422179341316, 0.04893781989812851, -0.012036141008138657, 0.020167019218206406, -0.00569771695882082, -0.019953450188040733, 0.0148430485278368, 0.008191898465156555, -0.023340046405792236, 0.0632164403796196, -0.014484558254480362, 0.03307269141077995, -0.005568050313740969, -0.01...
b61529a7a7fe25e07acf08f09d7ceeb398aa7922
subsection
24
60
Outsourcing Secure Computations for Cloud-based Blockchains
Then, the simulator computes one entry of the pivot tables by encrypting random values and the other entry by encrypting each garbled value with the random values generated in the SendPrivateParameters phase: A simulator of garbled circuits generates a simulated garbled circuit GC_{SC} and garbled values w_{jl} for ea...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.06671183556318283, 0.02355971559882164, 0.010780400596559048, 0.022186415269970894, 0.006847423501312733, 0.000849729054607451, 0.033325400203466415, 0.01614389754831791, -0.004779845010489225, 0.049438778311014175, -0.04293849691748619, 0.009262141771614552, -0.013893213123083115, 0.00...
69070450b2eb7816189ce93e716f1d11b206134f
subsection
25
60
Outsourcing Secure Computations for Cloud-based Blockchains
In this hybrid, the call SECCOMP by party E_{i} computes the pivot table P_{q}\left[j,l\right]=Enc_{s_{jl}^{x_{j}[l]}}\left(0\right),Enc_{s_{jl}^{x_{j}[l]}}\left(w_{jl}^{x_{j}[l]}\right), that is, only one garbled value per wire is encrypted: therefore, a contradiction will be reached since an adversary distinguishing ...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.049387335777282715, 0.011026684194803238, 0.014842146076261997, 0.02008458971977234, -0.0038250002544373274, -0.020389826968312263, 0.011690574698150158, 0.02283172309398651, -0.035102248191833496, 0.06306195259094238, -0.027501847594976425, 0.01836000196635723, -0.001877207076177001, -...
4817c352e44cea4c1057787eff21e9076fe79bba
subsection
26
60
Outsourcing Secure Computations for Cloud-based Blockchains
Therefore, the indistinguishability of the view of E_{i} is due to the secure channels: the case for more corrupted parties trivially follows from this argument.
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.010024523362517357, -0.0036047089379280806, -0.0009111902909353375, 0.013503354042768478, 0.00025819442817009985, -0.012145389802753925, 0.0007891260902397335, 0.041379768401384354, 0.0014437908539548516, 0.032774243503808975, 0.016844861209392548, -0.024870583787560463, -0.02438232675194...
79fa277acf6fc29d63e0d08243abeaeb1ee38580
subsection
27
60
Mining pre-processing data for Secure Multi-Party Computation
The Proof-of-Work of crypto-currencies consumes great amounts of computational power and electricity: 14TWh for Bitcoin and 4.25TWh for Ethereum just calculating hash functions. Miners could create pre-processing data for secure multi-party computation and be incentivised with crypto-tokens: 50-80% of total execution t...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 177, "openalex_id": "", "raw": "Digiconomist. Bitcoin Energy Consumption Index, 2017. http://digiconomist.net/bitcoin-energy-consumption.", "source_ref_id": "c52e0f71e28d17c0f93745e5c4efd52bcfd43372", "start": 0 },...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.010177689604461193, 0.023712338879704475, -0.04998817294836044, 0.054993096739053726, -0.010360796935856342, 0.004131348803639412, 0.03417994827032089, -0.007049614563584328, -0.006588032469153404, 0.030105819925665855, -0.023513972759246826, -0.0014944190625101328, 0.009681775234639645, ...
1047707c13e2b515c4775b124cf15a385cad342d
subsection
28
60
Mining pre-processing data for Secure Multi-Party Computation
Specifically, a value x\in \mathbb {\mathbb {F}}_{q} is secret shared among parties P by sampling \left(x_{i}\right)_{i\in P}\leftarrow \mathbb {F}_{q}^{\left|P\right|} subject to x=\sum _{i\in P}x_{i} with a party i holding the value x_{i}; the MAC is obtained by sampling \left(\gamma \left(x\right)_{i}\right)_{i\in P...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1224, "openalex_id": "", "raw": "Peter Scholl, Nigel P. Smart, and Tim Wood. When It's All Just Too Much: Outsourcing MPC-Preprocessing. Cryptology ePrint Archive, Report 2017/262, 2017. http://eprint.iacr.org/2017/262.pdf.", ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.02863943949341774, -0.0012072966201230884, -0.057065267115831375, 0.014190025627613068, -0.008834435604512691, 0.02552678808569908, 0.05926243215799332, 0.01693648286163807, -0.010787471197545528, 0.018370743840932846, -0.033140577375888824, 0.015654802322387695, 0.01003219559788704, 0....
1502df430f5aac001da4ffb977602c99f22af8c5
subsection
29
60
Security setting
Let E denote the set of n_{E} parties who are to run the online phase and O the set of n_{O} outsourcing parties that run the pre-preprocessing for the executing parties E (respectively Q and R in ). Adversaries can corrupt a majority of parties in E and in O, but not all parties in E nor all parties in O: that is, eac...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 199, "openalex_id": "", "raw": "Peter Scholl, Nigel P. Smart, and Tim Wood. When It's All Just Too Much: Outsourcing MPC-Preprocessing. Cryptology ePrint Archive, Report 2017/262, 2017. http://eprint.iacr.org/2017/262.pdf.", ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.05864162743091583, 0.014690917916595936, -0.005316495429724455, 0.01598762348294258, -0.006685664411634207, 0.013455233536660671, 0.007757354062050581, 0.019954020157456398, -0.01931329444050789, 0.051227521151304245, -0.008756580762565136, 0.019145485013723373, 0.008909134194254875, 0....
741a38b300b543402de4237fa2f2dbdd1c2d172d
subsection
30
60
Preventing Sybil attacks
In the context of blockchains, Sybil attacks in which an attacker creates a large number of pseudonymous identities can easily be prevented: before running secure computations, any party could be required to deposit some arbitrarily high amount of money on a smart contract that would be confiscated in case any abnormal...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.021860580891370773, 0.024118337780237198, -0.010198044590651989, 0.05320984497666359, -0.03261543810367584, -0.008908987045288086, 0.05925086885690689, -0.0013395898276939988, 0.005724482238292694, 0.01760440319776535, 0.0024045875761657953, 0.0009100629831664264, -0.0271693617105484, -...
98bf936b2de688ab7ac2ac36f984948f13639e0f
subsection
31
60
Other Applications
Applications of this technology can be found on many commercial/financial settings. Some of the most noteworthy are as follows:the most immediate application of secure multi-party computation is the removal of third parties and the financial industry has plenty of them: market makers, escrows, custodians, brokers, even...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1257/aer.102.3.65", "end": 1853, "openalex_id": "https://openalex.org/W3123665471", "raw": "Andrew W. Lo Emmanuel A. Abbe, Amir Khandani. Privacy-Preserving Methods for Sharing Financial Risk Exposures, 2011. https://papers.ssrn.com/sol3...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.025890355929732323, 0.028377173468470573, -0.03258797898888588, 0.0526961050927639, -0.031184377148747444, 0.010977083817124367, 0.024395214393734932, 0.006346722599118948, 0.05416073277592659, 0.016705917194485664, -0.016385529190301895, 0.028346659615635872, -0.005435144063085318, -0....
15b1270babd2b4a1e4a83959a089a1245d2b4073
subsection
32
60
Verifiable Smart Contracts
After the DAO attack, there has been some work (, , , , ) to include formal methods in the development of smart contracts: unfortunately, current solutions are very complex and cumbersome, , , , almost equivalent to formally verifying assembly code. Only very high-level languages should be used to write smart contracts...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 249, "openalex_id": "", "raw": "Yoichi Hirai. A Next-Generation Smart Contract and Decentralized Application Platform, 2017. https://yoichihirai.com/malta-paper.pdf.", "source_ref_id": "de5ec62b60803351a6b963b7ee04056fce042b...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.045505862683057785, 0.021074309945106506, -0.0371127612888813, -0.006916677113622427, 0.010430334135890007, -0.00943842250853777, 0.059575747698545456, 0.009499463252723217, -0.01727452501654625, 0.026690058410167694, 0.007122689858078957, -0.015191511251032352, 0.008271018043160439, -0...
da71c29cac7c4517907ea64f675bbfd540ae14b3
subsection
33
60
Verifiable Smart Contracts
Smart contracts must be written with the specific purpose of verification in mind, otherwise it becomes extremely complex to generate complete proofs; regarding the size, effort and duration of the verification process, there is a strong linear relationship between effort and proof size and a quadratic relationship bet...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1007/s00165-007-0060-5", "end": 397, "openalex_id": "https://openalex.org/W2009752377", "raw": "Jim Woodcock, Susan Stepney, David Cooper, John Clark, and Jeremy Jacob. The Certification of the Mondex Electronic Purse to ITSEC Level E6. ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.03955772519111633, 0.012491511180996895, 0.006963044870644808, 0.0020240521989762783, 0.01573457382619381, 0.014170273207128048, 0.004120596218854189, 0.007180520799010992, 0.028523683547973633, 0.011110348626971245, -0.008309869095683098, 0.019488895311951637, 0.005635296925902367, 0.0...
29b21efb91bd60ab463d72bae1dc891fc5c83302
subsection
34
60
Case Study
[H] class Crowdfunding {int minimum = 1000;// requires 0 < n// ensures \result >= minimumpublic int crowdfund(int n, int[] inputs) {int sum = 0;// invariant 0 <= i && i <= nfor (int i = 0; i < n; i++) {sum += inputs[i];}return sum;}}Crowdfund example processed with PCC toolchainProofs are written in Coq: the following ...
{ "cite_spans": [] }
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.05558697134256363, 0.011707738973200321, -0.031134195625782013, 0.0018410133197903633, 0.019754426553845406, -0.010906883515417576, 0.024788372218608856, -0.004118683282285929, -0.030401986092329025, 0.019922224804759026, 0.011684857308864594, 0.008450928144156933, 0.041552938520908356, ...
0a7dca9175b00b56452dd853b208dbf306a57ea8
subsection
35
60
On the Size of Certifiable Certificates
Since the invention of Proof Carrying Code, the question of how to efficiently represent and validate proofs has been a focus of much research. Different techniques are considered in this work to reduce the size of certificates:The reflection technique. This technique decreases the size of the proof using the reduction...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/263699.263712", "end": 143, "openalex_id": "https://openalex.org/W2034711041", "raw": "George C. Necula. Proof-carrying Code. In Proceedings of the 24th ACM SIGPLAN-SIGACT Symposium on Principles of Programming Languages, POPL '97, ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.024872956797480583, 0.004970776382833719, -0.013657236471772194, 0.005001295357942581, -0.010735046118497849, -0.04178046062588692, -0.0074122934602200985, 0.03799610957503319, 0.023636937141418457, 0.031098823994398117, -0.024644063785672188, -0.0005660314345732331, -0.009842365980148315...
c6af913a09cee8648cceddde9e3adbea4f7fd975
subsection
36
60
Zero-Knowledge Proofs of Proofs
Although it's possible to obfuscate certifiable certificates in such a way that de-obfuscation wouldn't be any easier while keeping the certifiable certificate sound and complete, this level of security isn't acceptable in a formal cryptographic model. When smart contracts are fully encrypted with homomorphic encryptio...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 252, "openalex_id": "", "raw": "François Dupressoir. Code and Proof Obfuscation, 2008. http://fdupress.net/files/m2-material/report.pdf.", "source_ref_id": "9d08e5a8dca2b6d3b473ba4f478c34d7d5a26987", "start": 0 }, ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.045255597680807114, 0.032011546194553375, 0.0031183792743831873, 0.0010690223425626755, 0.0027769787702709436, -0.04772741720080376, 0.028670016676187515, -0.01224464550614357, -0.026503363624215126, 0.054257892072200775, -0.018523363396525383, -0.007560401223599911, 0.002759813331067562,...
f27764ddd1cae8a5b9674c6f1ba4dde9e9ae57e9
subsection
37
60
Proof-Carrying Data
A conceptually related technique to Proof-Carrying Code is Proof-Carrying Data: in a distributed computation setting, PCD allows messages to be accompanied by proofs that said messages and the history leading to them follow a compliance predicate, in such a way that verifiers can be convinced that the compliance predic...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 399, "openalex_id": "https://openalex.org/W1504441476", "raw": "Alessandro Chiesa and Eran Tromer. Proof-Carrying Data and Hearsay Arguments from Signature Cards, 2010. https://people.eecs.berkeley.edu/~alexch/docs/CT10.pdf.", ...
1807.09484
Raziel: Private and Verifiable Smart Contracts on Blockchains
[ "David Cerezo Sánchez" ]
[ "cs.CR" ]
2,018
en
Computer Science
[ -0.043723512440919876, -0.019008895382285118, -0.053304240107536316, 0.0064685167744755745, -0.016369618475437164, -0.03353254497051239, 0.03170183300971985, 0.03685833886265755, -0.012128468602895737, 0.03902468457818031, -0.04314378648996353, -0.022502504289150238, -0.01437109149992466, ...