Search

Cross-lingual summarization aims to bridge language barriers by summarizing documents in different languages. However, ensuring semantic coherence across languages is an overlooked challenge and can be critical in several contexts. To fill this gap, we introduce multi-target cross-lingual summarization as the task of summarizing a document into multiple target languages while ensuring that the produced summaries are semantically similar. We propose a principled re-ranking approach to this problem and a multi-criteria evaluation protocol to assess semantic coherence across target languages, marking a first step that will hopefully stimulate further research on this problem., Comment: Accepted to EMNLP 2024 (Findings)

We develop a non-conventional description of the vacuum energy in quantum field theory in terms of quantum entropy. Precisely, we show that the vacuum energy of any non-interacting quantum field at zero temperature is proportional to the quantum entropy of the qubit degrees of freedom associated with virtual fluctuations. We prove this for fermions first, and then extend the derivation to quanta of any spin. Finally, we use these results to obtain the first law of thermodynamics for a non-interacting quantum vacuum at zero temperature., Comment: 7 pages, typos corrected, small restructure of the main text and conclusions section expanded. Included finite temperature effects and expanded section of virtual qubits

We study how the successive scattering of spin 1/2 particles with a central spin 1/2 target particle can generate entanglement between the helicity degrees of freedom of all scattered particles, effectively producing a multipartite entangled state. We show that the bipartite entanglement between each pair of scattered particles, as quantified by the concurrence, is largest for reflected particles and decreases with the number of scatterings. We study the entanglement generation as a function of the scattered particles momenta, angular distribution and mass ratios, and show that there is always a combination of optimal helicities and momentum which generate the largest amount of bipartite entanglement., Comment: 4 figures

The last few years have seen rapid development of applications of quantum computation to quantum field theory. The first algorithms for quantum simulation of scattering have been proposed in the context of scalar and fermionic theories, requiring thousands of logical qubits. These algorithms are not suitable to simulate scattering of incoming bound states, as the initial-state preparation relies typically on adiabatically transforming wavepackets of the free theory into wavepackets of the interacting theory. In this paper we present a strategy to excite wavepackets of the interacting theory directly from the vacuum of the interacting theory, allowing the preparation of states of composite particles. This is the first step towards digital quantum simulation of scattering of bound states. The approach is based on the Haag-Ruelle scattering theory, which provides a way to construct creation and annihilation operators of a theory in a full, nonperturbative framework. We provide a quantum algorithm requiring a number of ancillary qubits that is logarithmic in the size of the wavepackets, and with a success probability vanishing at most like a polynomial in the lattice parameters and the energy of the wavepacket. The gate complexity for a single iteration of the circuit is equivalent to that of a time evolution for a fixed time. Furthermore, we propose a complete protocol for scattering simulation using this algorithm. We study its efficiency and find improvements with respect to previous algorithms in the literature., Comment: Published version. 22 pages, 3 figures. Key words: quantum simulation, lattice field theory, scattering, bound states

We show that virtual particles, despite being unobservable, can be described by quantum operators which can be interpreted under certain conditions as valid qubit quantum states. For a single virtual fermion, we prove that such a state is a separable mixed 2-qubit state with a well-defined finite temperature. For spin-1 virtual bosons, we find them to be associated to 4-qubit operators which can be interpreted as quantum states for some gauges. We also study the creation of virtual pairs of fermions, where the pair is shown to be associated to an entangled 4-qubit operator, and show the corresponding quantum circuit. Finally, we prove that renormalization does not structurally affect these results. These findings represent new connections between quantum field theory, quantum information and quantum thermodynamics., Comment: Published version in PRA. The paper was restructured with new sections added. The spin 1 boson and fermionic pair creation cases were introduced

We show how a minimization principle of quantum entanglement between the oscillating flavors of a neutrino leads to a unique prediction for the CP-violation phase in the neutrino sector without assuming extra symmetries in the Standard Model. We find a theoretical prediction consistent with either no CP-violation or a very small presence of it., Comment: 5 pages, 3 figures

We show that a single polynomial entanglement measure is enough to verify equivalence between generic $n$-qubit states under Stochastic Local Operations with Classical Communication (SLOCC). SLOCC operations may be represented geometrically by M\"obius transformations on the roots of the entanglement measure on the Bloch sphere. Moreover, we show how the roots of the 3-tangle measure classify 4-qubit generic states, and propose a method to obtain the normal form of a 4-qubit state which bypasses the possibly infinite iterative procedure., Comment: 18 pages, 3 figures

Training neural network models with discrete (categorical or structured) latent variables can be computationally challenging, due to the need for marginalization over large or combinatorial sets. To circumvent this issue, one typically resorts to sampling-based approximations of the true marginal, requiring noisy gradient estimators (e.g., score function estimator) or continuous relaxations with lower-variance reparameterized gradients (e.g., Gumbel-Softmax). In this paper, we propose a new training strategy which replaces these estimators by an exact yet efficient marginalization. To achieve this, we parameterize discrete distributions over latent assignments using differentiable sparse mappings: sparsemax and its structured counterparts. In effect, the support of these distributions is greatly reduced, which enables efficient marginalization. We report successful results in three tasks covering a range of latent variable modeling applications: a semisupervised deep generative model, a latent communication game, and a generative model with a bit-vector latent representation. In all cases, we obtain good performance while still achieving the practicality of sampling-based approximations., Comment: Accepted for spotlight presentation at NeurIPS 2020

Systems at finite temperature make up the vast majority of realistic physical scenarios. Indeed, although zero temperature is often accompanied by simpler mathematics, the richness in physical results is evident when one considers the system to have temperature and even more so if the background geometry is curved. This thesis will be dedicated to the study of this type of physical systems, where thermodynamics and general relativity equally contribute to the dynamics. The first part will be devoted to the study of classical thermodynamic systems in curved spacetime, namely thin matter shells at finite temperature. These objects partition spacetime into separate pieces, and their very existence is conditioned by the so-called junctions conditions. The latter conditions allow us to carefully study both the mechanical and thermodynamics of the shell and, in particular, they give rise to a well-defined notion of entropy. The shell can then be taken to its black hole limit, providing an alternative way to study black hole thermodynamics. We will do this for different geometries, obtaining as byproduct a plausible answer for the debated value of the entropy of an extremal black hole. In the second part we shall review the standard formalisms to study QFT in curved spacetimes, in order to explore quantum properties of thermodynamic systems in the presence of gravity. Massive quantum scalar fields at finite temperature will be the systems of choice, whereby various instances of vacuum polarisation will be calculated in a variety of black hole geometries. Both numerical and analytic results will be obtained, and new addition formulas for a certain class of transcendental functions will be derived. This part will culminate with a careful numerical study of symmetry restoration of a self-interacting scalar field around a charged black hole, where we verify insights present in the literature., Comment: PhD thesis, Instituto Superior Tecnico, December 2018. 189 pages

Attention mechanisms have become ubiquitous in NLP. Recent architectures, notably the Transformer, learn powerful context-aware word representations through layered, multi-headed attention. The multiple heads learn diverse types of word relationships. However, with standard softmax attention, all attention heads are dense, assigning a non-zero weight to all context words. In this work, we introduce the adaptively sparse Transformer, wherein attention heads have flexible, context-dependent sparsity patterns. This sparsity is accomplished by replacing softmax with $\alpha$-entmax: a differentiable generalization of softmax that allows low-scoring words to receive precisely zero weight. Moreover, we derive a method to automatically learn the $\alpha$ parameter -- which controls the shape and sparsity of $\alpha$-entmax -- allowing attention heads to choose between focused or spread-out behavior. Our adaptively sparse Transformer improves interpretability and head diversity when compared to softmax Transformers on machine translation datasets. Findings of the quantitative and qualitative analysis of our approach include that heads in different layers learn different sparsity preferences and tend to be more diverse in their attention distributions than softmax Transformers. Furthermore, at no cost in accuracy, sparsity in attention heads helps to uncover different head specializations., Comment: Conference on Empirical Methods in Natural Language Processing (EMNLP), 2019, Hong Kong, China

In this work, we explore the space of quantum states composed of $N$ particles. To investigate the entanglement resistant to particles loss, we introduce the notion of $m$-resistant states. A quantum state is $m$-resistant if it remains entangled after losing an arbitrary subset of m particles, but becomes separable after losing a number of particles larger than $m$. We establish an analogy to the problem of designing a topological link consisting of $N$ rings such that, after cutting any $(m + 1)$ of them, the remaining rings become disconnected. We present a constructive solution to this problem, which allows us to exhibit several distinguished $N$-particles states with the desired property of entanglement resistance to a particle loss., Comment: 14 pages, 12 figures

Automatic post-editing (APE) seeks to automatically refine the output of a black-box machine translation (MT) system through human post-edits. APE systems are usually trained by complementing human post-edited data with large, artificial data generated through back-translations, a time-consuming process often no easier than training an MT system from scratch. In this paper, we propose an alternative where we fine-tune pre-trained BERT models on both the encoder and decoder of an APE system, exploring several parameter sharing strategies. By only training on a dataset of 23K sentences for 3 hours on a single GPU, we obtain results that are competitive with systems that were trained on 5M artificial sentences. When we add this artificial data, our method obtains state-of-the-art results., Comment: In proceedings of ACL 2019

This paper describes Unbabel's submission to the WMT2019 APE Shared Task for the English-German language pair. Following the recent rise of large, powerful, pre-trained models, we adapt the BERT pretrained model to perform Automatic Post-Editing in an encoder-decoder framework. Analogously to dual-encoder architectures we develop a BERT-based encoder-decoder (BED) model in which a single pretrained BERT encoder receives both the source src and machine translation tgt strings. Furthermore, we explore a conservativeness factor to constrain the APE system to perform fewer edits. As the official results show, when trained on a weighted combination of in-domain and artificial training data, our BED system with the conservativeness penalty improves significantly the translations of a strong Neural Machine Translation system by $-0.78$ and $+1.23$ in terms of TER and BLEU, respectively. Finally, our submission achieves a new state-of-the-art, ex-aequo, in English-German APE of NMT., Comment: Updated sections 2.2 and 4

The thermodynamic properties of self-gravitating spherical thin matter shells an black holes in $d>4$ dimensions are studied, extending previous analysis for $d=4$. The shell joins a Minkowski interior to a Tangherlini exterior, i.e., a Schwarzschild exterior in $d$ dimensions, with $d\geqslant4$, The junction conditions alone together with the first law of thermodynamics enable one to establish that the entropy of the thin shell depends only on its own gravitational radius. Endowing the shell with a power-law temperature equation of state allows to establish a precise form for the entropy and to perform a thermodynamic stability analysis for the shell. An interesting case is when the shell's temperature has the Hawking form, i.e., it is inversely proportional to the shell's gravitational radius. It is shown in this case that the shell's heat capacity is positive, and thus there is stability, for shells with radii in-between their own gravitational radius and the photonic radius, i.e., the radius of circular photon orbits, reproducing unexpectedly York's thermodynamic stability criterion for a $d=4$ black hole in the canonical ensemble. Additionally, the Euler equation for the matter shell is derived, the Bekenstein and holographic entropy bounds are studied, and the large $d$ limit is analyzed. Within this formalism the thermodynamic properties of black holes can be studied too. Putting the shell at its own gravitational radius, i.e., in the black hole situation, obliges one to choose precisely the Hawking temperature for the shell which in turn yields the Bekenstein-Hawking entropy. The stability analysis implies that the black hole is thermodynamically stable substantiating that in this configuration our system and York's canonical ensemble black hole are indeed the same system. Also relevant is the derivation in a surprising way of the Smarr formula for black holes in $d$ dimensions., Comment: 12 pages, 3 figures

We consider the restoration of a spontaneously broken symmetry of an interacting quantum scalar field around neutral, i.e., Schwarzschild, and electrically charged, i.e., Reissner-Nordstr\"om, black holes in four dimensions. This is done through a semiclassical self-consistent procedure, by solving the system of non-linear coupled equations describing the dynamics of the background field and the vacuum polarization. The black hole at its own horizon generates an indefinitely high temperature which decreases to the Hawking temperature at infinity. Due to the high temperature in its vicinity, there forms a bubble around the black hole in which the scalar field can only assume a value equal to zero, a minimum of energy. Thus, in this region the symmetry of the energy and the field is preserved. At the bubble radius, there is a phase transition in the value of the scalar field due to a spontaneous symmetry breaking mechanism. Indeed, outside the bubble radius the temperature is low enough such that the scalar field settles with a nonzero value in a new energy minimum, indicating a breaking of the symmetry in this outer region. Conversely, there is symmetry restoration from the outer region to the inner bubble close to the horizon. Specific properties that emerge from different black hole electric charges are also noteworthy. It is found that colder black holes, i.e., more charged ones, have a smaller bubble length of restored symmetry. In the extremal case the bubble has zero length, i.e., there is no bubble. Additionally, for colder black holes, it becomes harder to excite the quantum field modes, so the vacuum polarization has smaller values. In the extremal case, the black hole temperature is zero and the vacuum polarization is never excited., Comment: 16 pages, 4 figures

We propose a new classification scheme for quantum entanglement based on topological links. This is done by identifying a non-rigid ring to a particle, attributing the act of cutting and removing a ring to the operation of tracing out the particle, and associating linked rings to entangled particles. This analogy naturally leads us to a classification of multipartite quantum entanglement based on all possible distinct links for a given number of rings. To determine all different possibilities, we develop a formalism which associates any link to a polynomial, with each polynomial thereby defining a distinct equivalence class. In order to demonstrate the use of this classification scheme, we choose qubit quantum states as our example of physical system. A possible procedure to obtain qubit states from the polynomials is also introduced, providing an example state for each link class. We apply the formalism for the quantum systems of three and four qubits, and demonstrate the potential of these new tools in a context of qubit networks., Comment: 13 pages, 12 figures

WKB approximation methods are applied to the case of a massive scalar field around a 5-dimensional Reissner-Nordstr\"om black hole. The divergences are explicitly isolated and the cancellation against the Schwinger-De Witt counter-terms proven. The resulting finite quantity is evaluated for different values of the free parameters, namely the black hole mass and charge, and the scalar field mass. We thus extend our previous results on quantum vacuum polarization effects for uncharged asymptotically flat higher dimensional black holes to electrically charged black holes., Comment: 6 pages, 3 figures

The goal of this paper is to extend to higher dimensionality the methods and computations of vacuum polarization effects in black hole spacetimes. We focus our attention on the case of five dimensional Schwarzschild-Tangherlini black holes, for which we adapt the general method initially developed by Candelas and later refined by Anderson and others. We make use of point splitting regularization and of the WKB approximation to extract the divergences occuring in the coincidence limit of the Green function and, after calculating the counter-terms using the Schwinger - De Witt expansion, we explicitly prove the cancellation of the divergences and the regularity of the vacuum polarization once counter-terms are added up. We finally handle numerically the renormalized expression of the vacuum polarization. As a check on the method we also prove the regularity of the vacuum polarization in the six dimensional case in the large mass limit., Comment: 14 pages, 2 Figures

There has been considerable interest in applying the gauge/gravity duality to condensed matter theories with particular attention being devoted to gravity duals (Lifshitz spacetimes) of theories that exhibit anisotropic scaling. In this context, black hole solutions with Lifshitz asymptotics have also been constructed aiming at incorporating finite temperature effects. The goal here is to look at quantum polarization effects in these spacetimes, and to this aim, we develop a way to compute the coincidence limit of the Green's function for massive, non-minimally coupled scalar fields, adapting to the present situation the analysis developed for the case of asymptotically anti de Sitter black holes. The basics are similar to previous calculations, however in the Lifshitz case one needs to extend previous results to include a more general form for the metric and dependence on the dynamical exponent. All formulae are shown to reduce to the AdS case studied before once the value of the dynamical exponent is set to unity and the metric functions are accordingly chosen. The analytical results we present are general and can be applied to a variety of cases, in fact, to all spherically symmetric Lifshitz black hole solutions. We also implement the numerical analysis choosing some known Lifshitz black hole solutions as illustration.

Using a unified approach we study the entropy of extremal black holes through the entropy of an electrically charged thin shell. We encounter three cases in which a shell can be taken to its own gravitational or horizon radius and become an extremal spacetime. In case 1, we use a non extremal shell, calculate all the thermodynamics quantities including the entropy, take it to the horizon radius, and then take the extremal limit. In case 2, we take the extremal limit and the horizon radius limit simultaneously, i.e., as the shell approaches its horizon radius it also approaches extremality. In case 3, we build first an extremal shell, and then take its horizon radius. We find that the thermodynamic quantities in general have different expressions in the three different cases. The entropy is the Bekenstein-Hawking entropy $ S=A_+/4$ (where $A_+$ is the horizon area) in cases 1 and 2, and in case 3 it can be any well-behaved function of $A_+$. The contributions from the various thermodynamic quantities for the entropy in all three cases are distinct. Indeed, in cases 1 and 2 the limits agree in what concerns the entropy but they disagree in the behavior of all other thermodynamic quantities. Cases 2 and 3 disagree in what concerns the entropy but agree in the behavior of the local temperature and electric potential. Case 2 is in a sense intermediate between cases 1 and 3. Our approach sheds light on the extremal black hole entropy issue., Comment: 7 pages

There is a debate as to what is the value of the the entropy $S$ of extremal black holes. There are approaches that yield zero entropy $S=0$, while there are others that yield the Bekenstein-Hawking entropy $S=A_+/4$, in Planck units. There are still other approaches that give that $S$ is proportional to $r_+$ or even that $S$ is a generic well-behaved function of $r_+$. Here $r_+$ is the black hole horizon radius and $A_+=4\pi r_+^2$ is its horizon area. Using a spherically symmetric thin matter shell with extremal electric charge, we find the entropy expression for the extremal thin shell spacetime. When the shell's radius approaches its own gravitational radius, and thus turns into an extremal black hole, we encounter that the entropy is $S=S(r_+)$, i.e., the entropy of an extremal black hole is a function of $r_+$ alone. We speculate that the range of values for an extremal black hole is $0\leq S(r_+) \leq A_+/4$., Comment: 11 pages, minor changes, added references, matches the published version

A static self-gravitating electrically charged spherical thin shell embedded in a (3+1)-dimensional spacetime is used to study the thermodynamic and entropic properties of the corresponding spacetime. Inside the shell, the spacetime is flat, outside it is Reissner-Nordstr\"om, and this establishes the energy density, the pressure, and the electric charge in the shell. Imposing that the shell is at a given local temperature and that the first law of thermodynamics holds on the shell one can find the integrability conditions for the temperature and for the thermodynamic electric potential, the thermodynamic equilibrium states, and the thermodynamic stability conditions. Through the integrability conditions and the first law of thermodynamics an expression for the shell's entropy can be calculated. It is found that the shell's entropy is a function of the shell's gravitational and Cauchy radii alone. A plethora of sets of temperature and electric potential equations of state can be given. One set of equations of state is related to the Hawking temperature and a precisely given electric potential. Then, as one pushes the shell to its own gravitational radius and the temperature is set precisely equal to the Hawking temperature, so that there is a finite quantum backreaction, one finds that the entropy of the shell equals the Bekenstein-Hawking entropy for a black hole. Other set of equations of state are given and the thermodynamic properties analyzed. Whatever the initial equation of state for the temperature, as the shell radius approaches its own gravitational radius, the quantum backreaction imposes the Hawking temperature for the shell in this limit. Thus, when the shell's radius is sent to the shell's own gravitational radius the formalism developed allows to find the precise form of the Bekenstein-Hawking entropy of the correlated black hole., Comment: 15 pages, no figures, minor changes