- Wittek (2014) P. Wittek, Quantum Machine Learning: What Quantum Computing Means to Data Mining (Academic Press, New York, NY, USA, 2014).
- Schuld et al. (2015) M. Schuld, I. Sinayskiy, and F. Petruccione, Contemp. Phys. 56, 172 (2015).
- Adcock et al. (2015) J. Adcock, E. Allen, M. Day, S. Frick, J. Hinchliff, M. Johnson, S. Morley-Short, S. Pallister, A. Price, and S. Stanisic, arXiv:1512.02900 (2015).
- Biamonte et al. (2016) J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, arXiv:1611.09347 (2016).
- Arunachalam and de Wolf (2017) S. Arunachalam and R. de Wolf, arXiv:1701.06806 (2017).
- Aïmeur et al. (2013) E. Aïmeur, G. Brassard, and S. Gambs, Machine Learning 90, 261 (2013).
- Rebentrost et al. (2014) P. Rebentrost, M. Mohseni, and S. Lloyd, Phys. Rev. Lett. 113, 130503 (2014).
- Cai et al. (2015) X.-D. Cai, D. Wu, Z.-E. Su, M.-C. Chen, X.-L. Wang, L. Li, N.-L. Liu, C.-Y. Lu, and J.-W. Pan, Phys. Rev. Lett. 114, 110504 (2015).
- Lloyd et al. (2016) S. Lloyd, S. Garnerone, and P. Zanardi, Nat. Commun. 7, 10138 (2016).
- Bennett et al. (1999) C. H. Bennett, D. P. DiVincenzo, C. a. Fuchs, T. Mor, E. Rains, P. W. Shor, J. a. Smolin, and W. K. Wootters, Phys. Rev. A 59, 1070 (1999).
- Giovannetti et al. (2001) V. Giovannetti, S. Lloyd, and L. Maccone, Nature 412, 417 (2001).
- Giovannetti et al. (2004) V. Giovannetti, S. Lloyd, and L. Maccone, Science 306, 1330 (2004).
- Niset et al. (2007) J. Niset, A. Acín, U. L. Andersen, N. J. Cerf, R. García-Patrón, M. Navascués, and M. Sabuncu, Phys. Rev. Lett. 98, 260404 (2007).
- Sentís et al. (2015) G. Sentís, M. Guţă, and G. Adesso, EPJ Quantum Technology 2, 17 (2015).
- Sentís et al. (2016) G. Sentís, E. Bagan, J. Calsamiglia, G. Chiribella, and R. Muñoz-Tapia, Phys. Rev. Lett. 117, 150502 (2016).
Devroye et al. (1996)
L. Devroye, L. Györfi, and G. Lugosi,
A Probabilistic Theory of Pattern Recognition(Springer, 1996).
- Hastie et al. (2008) T. Hastie, R. Tibshirani, and J. Friedman, The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd ed. (Springer, 2008).
Gammerman et al. (1998)
A. Gammerman, V. Vovk, and V. Vapnik, in
Proceedings of UAI-98, 14th Conference on Uncertainty in Artificial Intelligence(1998) pp. 148–155.
- Dunjko et al. (2016) V. Dunjko, J. M. Taylor, and H. J. Briegel, Phys. Rev. Lett. 117, 130501 (2016).
- Vapnik (1995) V. Vapnik, The Nature of Statistical Learning Theory (Springer, New York, NY, USA, 1995).
- (21) See Supplemental Material, which includes Refs. Bengtsson and Życzkowski (2006); Christandl et al. (2007); Christandl and Toner (2009).
- Bae and Acín (2006) J. Bae and A. Acín, Phys. Rev. Lett. 97, 030402 (2006).
- Sentís et al. (2010) G. Sentís, E. Bagan, J. Calsamiglia, and R. Muñoz-Tapia, Phys. Rev. A 82, 042312 (2010).
- Sentís et al. (2012) G. Sentís, J. Calsamiglia, R. Muñoz-Tapia, and E. Bagan, Sci. Rep. 2, 708 (2012).
- (25) All these tasks are meant within the framework described in Fig. 1, and therefore their performance is assessed only through registers and . In the fully quantum case of generating bipartite states from , this means that a risk observable would measure the dissimilarity between the states and , instead of comparing (perhaps more naturally in other contexts) the produced bipartite state with a copy of the target bipartite state saved as reference.
- Dieks (1982) D. Dieks, Phys. Lett. A 92, 271 (1982).
- Wootters and Zurek (1982) W. K. Wootters and W. H. Zurek, Nature 299, 802 (1982).
- (28) We note that a de Finetti theorem for fully symmetric quantum channels can be found in the literature Fawzi and Renner (2015).
- Christandl et al. (2007) M. Christandl, R. König, G. Mitchison, and R. Renner, Commun. Math. Phys. 273, 473 (2007).
- Sasaki and Carlini (2002) M. Sasaki and A. Carlini, Phys. Rev. A 66, 022303 (2002).
- Bisio et al. (2010) A. Bisio, G. Chiribella, G. M. D’Ariano, S. Facchini, and P. Perinotti, Phys. Rev. A 81, 032324 (2010).
- Nakahira et al. (2015) K. Nakahira, K. Kato, and T. S. Usuda, Phys. Rev. A 91, 052304 (2015).
- Chiribella et al. (2004) G. Chiribella, G. M. D’Ariano, P. Perinotti, and M. F. Sacchi, Phys. Rev. A 70, 062105 (2004).
- Chiribella et al. (2008) G. Chiribella, G. M. D’Ariano, and P. Perinotti, Phys. Rev. Lett. 101, 060401 (2008).
- Bengtsson and Życzkowski (2006) I. Bengtsson and K. Życzkowski, Geometry of quantum states: an introduction to quantum entanglement (Cambridge University Press, 2006).
- Christandl and Toner (2009) M. Christandl and B. Toner, Journal of Mathematical Physics 50, 042104 (2009).
- Fawzi and Renner (2015) O. Fawzi and R. Renner, Commun. Math. Phys. 340, 575 (2015).
Appendix A Supplemental material: Proof of our main result
Our main result, Theorem 3 (Theorem 1 in the main text), consists in showing that for every non-signalling CPTP map there is a symmetric one-way LOCC map that approximately reproduces all local expectation values, and is non-signalling by construction. The backbone of our result is the quantum de Finetti theorem, specifically in its form as it appears in Christandl et al. (2007), which we restate here:
Theorem 1 (Quantum de Finetti theorem Christandl et al. (2007)).
Let and be quantum systems and let be a symmetric quantum state under exchange of the systems. If admits a symmetric extension then there is a set , a POVM over on , and a map such that
where , , , only depends on the -extension and, in particular, is independent of . denotes the trace-norm of operator . In general, one can take . G and the accuracy of the approximation is independent of the dimension of .
Theorem 2 (Choi).
Every CP map can be represented by a positive semidefinite operator , such that
where , and . In addition, for any we have
The adjoint map is given by (we use the customary identification between and induced by the Hilbert-Schmidt product)
In addition, if is trace-preserving, then .
This allows us to characterize properties of channels by referring to properties of their respective Choi matrices. The non-signalling property of a quantum channel has a direct relation with the reduced states of its Choi matrix:
Let be a non-signalling quantum channel, and let be its Choi matrix. Then
and is the Choi matrix of the induced channel .
Lemma 1 is proved by straightforward evaluation.
Applying Theorem 1 to the Choi matrix of the CPTP map , , we get an approximation to as described by the Choi matrix
For the approximation is exact, so , therefore is a POVM. The positive semidefinite quantum states describe a family of completely positive maps .
The state does not, however, represent a quantum operation which is deterministically realizable, in the first place because may not be , as is required for a trace-preserving channel. Furthermore, a quantum channel can be implemented by 1-way LOCC iff its Choi matrix is of the form
where , for all . This would ensure that all corresponding CP maps are trace-preserving, and thus the channel described by can be implemented by first performing measurement on and then applying on each of the systems .
Although one does not expect that each in Eq. (9) satisfies
on average they approximately do. More importantly, we now show that the outcomes are concentrated with high probability on those which almost satisfy the condition. Let be the trace-norm and be the operator norm.
Let be a non-signalling CPTP map with Choi matrix , and let and be such that
is a separable approximation of such that
Define for all and for any subset ,
Then, the following holds
For any , let , . Then
Consider the measurement is performed on the state yielding outcome , and is to be applied on each of the test instances. Of course, for this to be deterministically implementable, one needs that , which amounts to . If this condition is met approximately, one can implement a suitably modified map at the expense of actually implementing a slightly worse approximation to . However, if the condition is not met even approximately, the implementation cannot be expected to approximate . Lemma 2 shows that this case is unlikely to occur, since
Hence, one can slightly modify the operators into in order to satisfy Eq. (11) and ensure that in all cases, either and are close enough, or is unlikely enough so that the approximation still converges in to the actual channel given by . We call this a 1-way LOCC approximation.
Lemma 3 (1-way LOCC approximation).
Let be a symmetric, non-signalling CPTP map with Choi matrix . Then there is a POVM and there are states such that and the quantum state
is a separable approximation to ,
where is a constant depending on and .
Proof of Lemma 3.
Therefore, for we have
Thus, we can ensure that all satisfy . We can define
where is the Choi matrix of any CPTP map . By definition every has , and using we can write
Thus, Lemma 6 shows that for all ,
and the subadditivity of the trace distance () leads to
Combining this with for all ,
and the triangle inequality we get
Taking and using the triangle inequality we get
Chosing and expanding around up to leading order we get
which using leads to
the desired result. ∎
Having established a 1-way LOCC approximation bound for any symmetric non-signalling channel, we can now proceed to prove our main result (Theorem 1 in the main text):
Theorem 3 (Main result).
Let be a non-signalling quantum channel, and let be a local operator. Then, there exists a POVM on and a set of quantum channels such that the quantum channel ,
Proof of Theorem 3.
We want to obtain approximation bounds for
The specific form of is irrelevant for our purposes, besides symmetry among the parties. Expressing in terms of the symmetrized local channel , and in turn, in terms of its Choi matrix, we have
Appendix B Proofs of Lemmas and Theorem 1
We restate and prove Lemma 1 in the main text. We also mention that a related but more general result on a de Finetti theorem for non-signalling classical conditional probability distributions can be found inChristandl and Toner (2009).
For every inductive learning protocol that assigns labels to test instantes , there exists a set of classifying functions and stochastic maps , such that the inductive protocol
has expected risk for all .
Proof of Lemma 4.
Consider the expected risk of protocol . Let be any permutation of elements, and let the be the accordingly permuted protocol
Furthermore, let be the symmetrized protocol,
It follows trivially that
One can define the marginal maps , which are all equal, so we refer to them as ,
Since is non-signalling, so is , namely satisfies the condition
and so we can simply write . The conditional expected risk can be expressed in terms of ,
Considering fixed, is a stochastic map from to , and thus it is a convex combination of deterministic maps for some set of functions , i.e.
where is a probability measure that depends on . Then
Thus, the stochastic maps and can be combined into the protocol
Proof of Theorem 1.
Let us start by assuming admits a pure state extension . Then
where is the symmetric subspace of . Let also .
Let be a generic element, a reference state in , and the Haar measure on . Let and use .
For any , let be a POVM in , such that
This allows to write
where is the residual state on when measuring
Then is close to a convex combination of separable and -iid states , with a distribution independent of , namely
is close to zero in trace-norm. The operator-valued measure is given by
We now bound , where
One can readily check that
On the other hand,
Notice that this is an operator in . With this we have