Search

We report on observations of the extended environment of the bright Be star $\gamma$-Cas performed using intensity interferometry measurements within its H$\alpha$ emission line. These observations were performed using a modified version of the I2C intensity interferometry instrument installed onto the 1.54 meter M\'{e}O optical metrology telescope and a portable 1-meter telescope (T1M). In order to better constrain the extent of the H$\alpha$ envelope, observations were performed for two different positions of the T1M telescope, corresponding to an intermediate and long baselines in which the extended region was partially and fully resolved. We find that the observed data are consistent with past interferometric observations of $\gamma$-Cas. These observations demonstrate the capability to equip optical telescopes of different optical designs with intensity interferometry capabilities and illustrate the potential to scale a similar system onto many additional telescopes.

We investigate the extent to which a two-level quantum system subjected to an external time-dependent drive can be characterized by supervised learning. We apply this approach to the case of bang-bang control and the estimation of the offset and the final distance to a given target state. For any control protocol, the goal is to find the mapping between the offset and the distance. This mapping is interpolated using a neural network. The estimate is global in the sense that no a priori knowledge is required on the relation to be determined. Different neural network algorithms are tested on a series of data sets. We show that the mapping can be reproduced with very high precision in the direct case when the offset is known, while obstacles appear in the indirect case starting from the distance to the target. We point out the limits of the estimation procedure with respect to the properties of the mapping to be interpolated. We discuss the physical relevance of the different results., Comment: 19 pages, 5 figures

Consider $n$ points independently sampled from a density $p$ of class $\mathcal{C}^2$ on a smooth compact $d$-dimensional sub-manifold $\mathcal{M}$ of $\mathbb{R}^m$, and consider the generator of a random walk visiting these points according to a transition kernel $K$. We study the almost sure uniform convergence of this operator to the diffusive Laplace-Beltrami operator when $n$ tends to infinity. This work extends known results of the past 15 years. In particular, our result does not require the kernel $K$ to be continuous, which covers the cases of walks exploring $k$NN-random and geometric graphs, and convergence rates are given. The distance between the random walk generator and the limiting operator is separated into several terms: a statistical term, related to the law of large numbers, is treated with concentration tools and an approximation term that we control with tools from differential geometry. The convergence of $k$NN Laplacians is detailed., Comment: 32 pages;0 figure

Recently, Person Re-Identification (Re-ID) has received a lot of attention. Large datasets containing labeled images of various individuals have been released, allowing researchers to develop and test many successful approaches. However, when such Re-ID models are deployed in new cities or environments, the task of searching for people within a network of security cameras is likely to face an important domain shift, thus resulting in decreased performance. Indeed, while most public datasets were collected in a limited geographic area, images from a new city present different features (e.g., people's ethnicity and clothing style, weather, architecture, etc.). In addition, the whole frames of the video streams must be converted into cropped images of people using pedestrian detection models, which behave differently from the human annotators who created the dataset used for training. To better understand the extent of this issue, this paper introduces a complete methodology to evaluate Re-ID approaches and training datasets with respect to their suitability for unsupervised deployment for live operations. This method is used to benchmark four Re-ID approaches on three datasets, providing insight and guidelines that can help to design better Re-ID pipelines in the future., Comment: This paper is the extended version of our short paper accepted in VISAPP - 2023

The aim of this article is to establish a concise proof for a stability result of self-similar solutions of the binormal flow, in some more restrictive cases than in [5]. This equation, also known as the Local Induction Approximation, is a standard model for vortex filament dynamics, and its self-similar solution describes the formation of a corner singularity on the filament. Our approach strongly uses the link that Hasimoto pointed out in 1972 between the solution of the binormal flow and the one of the 1-D cubic Schr\"odinger equation, as well as the existence results associated to the latter.

The aim of this article is to prove the existence of a new class of solutions of 1D cubic NLS with an initial data related to a sum of Dirac masses, of critical regularity $F(L^\infty)$, and belonging to $\dot H^s$ for any $s <-1/2$. This problem is motivated by the lack of result for critical regularity initial condition, and also by the study of the vortex filaments dynamics approximated by the binormal flow. Our result is based on a scattering approach, after performing a pseudo-conformal transformation, and on fine estimations of oscillatory integrals.

Photons are the elementary quantum excitations of the electromagnetic field. Quantization is usually constructed on the basis of an expansion in eigenmodes, in the form of plane waves. Since they form a basis, other electromagnetic configurations can be constructed by linear combinations. In this presentation we discuss a formulation constructed in the general formalism of bosonic Fock space, in which the quantum excitation can be constructed directly on localized pulses of arbitrary shape. Although the two formulations are essentially equivalent, the direct formulation in terms of pulses has some conceptual and practical advantages, which we illustrate with some examples. The first one is the passage of a single photon pulse through a beam splitter. The analysis of this formulation in terms of pulses in Fock space shows that there is no need to introduce "vacuum fluctuations entering through the unused port", as is often done in the literature. Another example is the Hong-Ou-Mandel effect. It is described as a time dependent process in the Schr\"odinger representation in Fock space. The analysis shows explicitly how the two essential ingredients of the Hong-Ou-Mandel effect are the same shape of the pulses and the bosonic nature of photons. This formulation shows that all the phenomena involving linear quantum optical devices can be described and calculated on the basis of the time dependent solution of the corresponding classical Maxwell's equations for pulses, from which the quantum dynamics in Fock space can be immediately constructed., Comment: 26 pages, 3 figures

The usage of deep neural networks in safety-critical systems is limited by our ability to guarantee their correct behavior. Runtime monitors are components aiming to identify unsafe predictions and discard them before they can lead to catastrophic consequences. Several recent works on runtime monitoring have focused on out-of-distribution (OOD) detection, i.e., identifying inputs that are different from the training data. In this work, we argue that OOD detection is not a well-suited framework to design efficient runtime monitors and that it is more relevant to evaluate monitors based on their ability to discard incorrect predictions. We call this setting out-ofmodel-scope detection and discuss the conceptual differences with OOD. We also conduct extensive experiments on popular datasets from the literature to show that studying monitors in the OOD setting can be misleading: 1. very good OOD results can give a false impression of safety, 2. comparison under the OOD setting does not allow identifying the best monitor to detect errors. Finally, we also show that removing erroneous training data samples helps to train better monitors.

While the emergence of large administrative claims data provides opportunities for research, their use remains limited by the lack of clinical annotations relevant to disease outcomes, such as recurrence in breast cancer (BC). Several challenges arise from the annotation of such endpoints in administrative claims, including the need to infer both the occurrence and the date of the recurrence, the right-censoring of data, or the importance of time intervals between medical visits. Deep learning approaches have been successfully used to label temporal medical sequences, but no method is currently able to handle simultaneously right-censoring and visit temporality to detect survival events in medical sequences. We propose EDEN (Event DEtection Network), a time-aware Long-Short-Term-Memory network for survival analyses, and its custom loss function. Our method outperforms several state-of-the-art approaches on real-world BC datasets. EDEN constitutes a powerful tool to annotate disease recurrence from administrative claims, thus paving the way for the massive use of such data in BC research., Comment: Extended Abstract presented at Machine Learning for Health (ML4H) symposium 2022, November 28th, 2022, New Orleans, United States & Virtual, http://www.ml4h.cc, 6 pages

Medical dialogue generation aims to generate responses according to a history of dialogue turns between doctors and patients. Unlike open-domain dialogue generation, this requires background knowledge specific to the medical domain. Existing generative frameworks for medical dialogue generation fall short of incorporating domain-specific knowledge, especially with regard to medical terminology. In this paper, we propose a novel framework to improve medical dialogue generation by considering features centered on domain-specific terminology. We leverage an attention mechanism to incorporate terminologically centred features, and fill in the semantic gap between medical background knowledge and common utterances by enforcing language models to learn terminology representations with an auxiliary terminology recognition task. Experimental results demonstrate the effectiveness of our approach, in which our proposed framework outperforms SOTA language models. Additionally, we provide a new dataset with medical terminology annotations to support the research on medical dialogue generation. Our dataset and code are available at https://github.com/tangg555/meddialog., Comment: Accepted by ICASSP 2023

One of the key challenges of automatic story generation is how to generate a long narrative that can maintain fluency, relevance, and coherence. Despite recent progress, current story generation systems still face the challenge of how to effectively capture contextual and event features, which has a profound impact on a model's generation performance. To address these challenges, we present EtriCA, a novel neural generation model, which improves the relevance and coherence of the generated stories through residually mapping context features to event sequences with a cross-attention mechanism. Such a feature capturing mechanism allows our model to better exploit the logical relatedness between events when generating stories. Extensive experiments based on both automatic and human evaluations show that our model significantly outperforms state-of-the-art baselines, demonstrating the effectiveness of our model in leveraging context and event features., Comment: Accepted by EMNLP 2022 Findings

Story generation aims to generate a long narrative conditioned on a given input. In spite of the success of prior works with the application of pre-trained models, current neural models for Chinese stories still struggle to generate high-quality long text narratives. We hypothesise that this stems from ambiguity in syntactically parsing the Chinese language, which does not have explicit delimiters for word segmentation. Consequently, neural models suffer from the inefficient capturing of features in Chinese narratives. In this paper, we present a new generation framework that enhances the feature capturing mechanism by informing the generation model of dependencies between words and additionally augmenting the semantic representation learning through synonym denoising training. We conduct a range of experiments, and the results demonstrate that our framework outperforms the state-of-the-art Chinese generation models on all evaluation metrics, demonstrating the benefits of enhanced dependency and semantic representation learning.

To improve the performance of long text generation, recent studies have leveraged automatically planned event structures (i.e. storylines) to guide story generation. Such prior works mostly employ end-to-end neural generation models to predict event sequences for a story. However, such generation models struggle to guarantee the narrative coherence of separate events due to the hallucination problem, and additionally the generated event sequences are often hard to control due to the end-to-end nature of the models. To address these challenges, we propose NGEP, an novel event planning framework which generates an event sequence by performing inference on an automatically constructed event graph and enhances generalisation ability through a neural event advisor. We conduct a range of experiments on multiple criteria, and the results demonstrate that our graph-based neural framework outperforms the state-of-the-art (SOTA) event planning approaches, considering both the performance of event sequence generation and the effectiveness on the downstream task of story generation.

The ability of pulse-shaping devices to generate accurately quantum optimal control is a strong limitation to the development of quantum technologies. We propose and demonstrate a systematic procedure to design robust digital control processes adapted to such experimental constraints. We show to what extent this digital pulse can be obtained from its continuous-time counterpart. A remarkable efficiency can be achieved even for a limited number of pulse parameters. We experimentally implement the protocols on IBM quantum computers for a single qubit, obtaining an optimal robust transfer in a time T = 382 ns.

We use radiances collected from space by the Infrared Atmospheric Sounder Interferometer (IASI) when looking down at ocean surfaces during the day to remotely determine the probability distribution of wave slopes. This is achieved by using about 300 channels between 3.6 and 4.0 $\mu$m and a physically-based approach which properly takes the contribution of the reflected solar radiation into account. Based on about 150 million observations, the same number of wave-slope probabilities are retrieved for wind speeds up to 15 m/s. We revisit and discuss the methodology proposed by Cox and Munk (CM) to derive their celebrated wave-slope probability distribution function (pdf) from photographs of the sun glitter. We propose an original and robust approach for accurate retrievals of the 7 parameters appearing in the Gram-Charlier representation of the pdf. Our results for the mean square slopes are fully compatible with those of CM, and with the more recent results by Br\'eon and Henriot, but our lower uncertainties enable us to point out departures from the linear wind-speed dependencies and a slight overestimation of the upwind MSS described by the linear fit of CM at moderate wind speed. Our skewness and kurtosis coefficients show clear influences of the wind speed, with a steady decrease of the former and the alongwind kurtosis coefficient being maximal at moderate wind speeds, features that CM could not point out due to the limitations of their measurements. We revisit the renormalization procedure employed by CM to obtain the complete variances from truncated pdf and show that it imposes stringent conditions on the kurtosis coefficients that allow us to determine them accurately, with wind-dependent values specific to the local sea state., Comment: 54 pages, 15 figures, 1 table

Robust time-optimal control is known to feature constant (square) pulses. We analyze fast adiabatic dynamics that preserve robustness by using alternative smooth quasi-square pulses, typically represented by hyper-Gaussian pulses. We here consider the two protocols, robust inverse optimization and time-contracted adiabatic passage, allowing the design of the same pulse shape in both cases. The dynamics and their performance are compared. The superiority of the former protocol is shown.

We demonstrate the reversible in situ photoreduction of molecular junctions of phosphomolybdate [PMo12O40]3- monolayer self-assembled on flat gold electrodes, connected by the tip of a conductive atomic force microscope. The conductance of the one electron reduced [PMo12O40]4- molecular junction is increased by ca. 10, this open-shell state is stable in the junction in air at room temperature. The analysis of a large current-voltage dataset by unsupervised machine learning and clustering algorithms reveals that the electron transport in the pristine phosphomolybdate junctions leads to symmetric current-voltage curves, controlled by the lowest unoccupied molecular orbital (LUMO) at 0.6 - 0.7 eV above the Fermi energy with ca. 25% of the junctions having a better electronic coupling to the electrodes than the main part of the dataset. This analysis also shows that a small fraction (ca. 18% of the dataset) of the molecules is already reduced. The UV light in situ photoreduced phosphomolybdate junctions are systematically featuring slightly asymmetric current - voltage behaviors, which is ascribed to electron transport mediated by the single occupied molecular orbital (SOMO) nearly at resonance with the Fermi energy of the electrode and by a closely located single unoccupied molecular orbital (SUMO) at ca. 0.3 eV above the SOMO with a weak electronic coupling to the electrodes (ca. 50% of the dataset) or at ca. 0.4 eV but with a better electrode coupling (ca. 50% of the dataset). These results shed lights to the electronic properties of reversible switchable redox polyoxometalates, a key point for potential applications in nanoelectronic devices.

We apply the inverse geometric optimization technique to generate an optimal and robust stimulated Raman exact passage (STIREP) considering the loss of the upper state as a characterization parameter. Control fields temporal shapes that are optimal with respect to pulse area, energy, and duration, are found to form a simple sequence with a combination of intuitively (near the beginning and the end) and counter-intuitively ordered pulse pairs. The resulting dynamics produces a loss which is about a third of that of the non-robust optimal STIREP. Alternative optimal solutions featuring lower losses, larger pulse areas, and fully counter-intuitive pulse sequences are derived.

The Siegert relation relates field and intensity temporal correlations. After a historical review of the Siegert relation and the Hanbury Brown and Twiss effect, we discuss the validity of this relation in two different domains. We first show that this relation can be used in astrophysics to determine the fundamental parameters of stars, and that it is especially important for the observation with stellar emission lines. Second, we verify the validity of this relation for moving quantum scatterers illuminated by a strong driving field.

Person Re-Identification (Re-ID) aims to search for a person of interest (query) in a network of cameras. In the classic Re-ID setting the query is sought in a gallery containing properly cropped images of entire bodies. Recently, the live Re-ID setting was introduced to represent the practical application context of Re-ID better. It consists in searching for the query in short videos, containing whole scene frames. The initial live Re-ID baseline used a pedestrian detector to build a large search gallery and a classic Re-ID model to find the query in the gallery. However, the galleries generated were too large and contained low-quality images, which decreased the live Re-ID performance. Here, we present a new live Re-ID approach called TrADe, to generate lower high-quality galleries. TrADe first uses a Tracking algorithm to identify sequences of images of the same individual in the gallery. Following, an Anomaly Detection model is used to select a single good representative of each tracklet. TrADe is validated on the live Re-ID version of the PRID-2011 dataset and shows significant improvements over the baseline., Comment: 6 pages, 4 figures, Accepted on ICMLA 2022

We present the current status of the I2C stellar intensity interferometer used towards high angular resolution observations of stars in visible wavelengths. In these proceedings, we present recent technical improvements to the instrument, and share results from ongoing campaigns using arrays of small diameter optical telescopes. A tip-tilt adaptive optics unit was integrated into the optical system to stabilize light injection into an optical fiber. The setup was successfully tested with several facilities on the Calern Plateau site of the Observatoire de la C\^ote d'Azur. These include one of the 1 m diameter telescopes of the C2PU observatory, a portable 1 m diameter telescope, and also the 1.5 m M\'eO telescope. To better constrain on-sky measurements, the spectral transmission of instrument was characterized in the laboratory using a high resolution spectrograph. The system was also tested with two of the auxiliary telescopes of the VLTI resulting in successful temporal and spatial correlation measurements of three stars.

Super- and subradiance are usually described in the framework of Dicke collective states, which is an ``atomic picture'' in which the electromagnetic field only provides an effective interaction between the atoms. Here, we discuss a complementary picture, in which we describe the propagation and scattering of light in the atomic medium, which provides a complex susceptibility and scatterers. This ``optical picture'' is valid in the linear-optics regime for disordered samples and is mainly relevant at low density, when the susceptibility and scattering cross-section can be computed from simple textbook formulas. In this picture, superradiance is a dispersion effect due to a single scattering event dressed by an effective refractive index, whereas subradiance is due to multiple scattering. We present numerical and experimental data supporting our interpretation.

With the increasing use of Machine Learning (ML) in critical autonomous systems, runtime monitors have been developed to detect prediction errors and keep the system in a safe state during operations. Monitors have been proposed for different applications involving diverse perception tasks and ML models, and specific evaluation procedures and metrics are used for different contexts. This paper introduces three unified safety-oriented metrics, representing the safety benefits of the monitor (Safety Gain), the remaining safety gaps after using it (Residual Hazard), and its negative impact on the system's performance (Availability Cost). To compute these metrics, one requires to define two return functions, representing how a given ML prediction will impact expected future rewards and hazards. Three use-cases (classification, drone landing, and autonomous driving) are used to demonstrate how metrics from the literature can be expressed in terms of the proposed metrics. Experimental results on these examples show how different evaluation choices impact the perceived performance of a monitor. As our formalism requires us to formulate explicit safety assumptions, it allows us to ensure that the evaluation conducted matches the high-level system requirements., Comment: 9 pages, 5 figures, 3 tables, to appear in the proceedings of the 33rd IEEE International Symposium on Software Reliability Engineering (ISSRE 2022)

The future Ricochet experiment aims at searching for new physics in the electroweak sector by providing a high precision measurement of the Coherent Elastic Neutrino-Nucleus Scattering (CENNS) process down to the sub-100 eV nuclear recoil energy range. The experiment will deploy a kg-scale low-energy-threshold detector array combining Ge and Zn target crystals 8.8 meters away from the 58 MW research nuclear reactor core of the Institut Laue Langevin (ILL) in Grenoble, France. Currently, the Ricochet collaboration is characterizing the backgrounds at its future experimental site in order to optimize the experiment's shielding design. The most threatening background component, which cannot be actively rejected by particle identification, consists of keV-scale neutron-induced nuclear recoils. These initial fast neutrons are generated by the reactor core and surrounding experiments (reactogenics), and by the cosmic rays producing primary neutrons and muon-induced neutrons in the surrounding materials. In this paper, we present the Ricochet neutron background characterization using $^3$He proportional counters which exhibit a high sensitivity to thermal, epithermal and fast neutrons. We compare these measurements to the Ricochet Geant4 simulations to validate our reactogenic and cosmogenic neutron background estimations. Eventually, we present our estimated neutron background for the future Ricochet experiment and the resulting CENNS detection significance., Comment: 14 pages, 14 figures, 1 table

This paper considers a setting where embedded devices are used to acquire and classify images. Because of limited computing capacity, embedded devices rely on a parsimonious classification model with uneven accuracy. When local classification is deemed inaccurate, devices can decide to offload the image to an edge server with a more accurate but resource-intensive model. Resource constraints, e.g., network bandwidth, however, require regulating such transmissions to avoid congestion and high latency. The paper investigates this offloading problem when transmissions regulation is through a token bucket, a mechanism commonly used for such purposes. The goal is to devise a lightweight, online offloading policy that optimizes an application-specific metric (e.g., classification accuracy) under the constraints of the token bucket. The paper develops a policy based on a Deep Q-Network (DQN), and demonstrates both its efficacy and the feasibility of its deployment on embedded devices. Of note is the fact that the policy can handle complex input patterns, including correlation in image arrivals and classification accuracy. The evaluation is carried out by performing image classification over a local testbed using synthetic traces generated from the ImageNet image classification benchmark. Implementation of this work is available at https://github.com/qiujiaming315/edgeml-dqn., Comment: 13 pages, 12 figures, to appear at the Proceedings of the 2022 International Conference on Embedded Software, October 07-14, 2022, Shanghai, China

Over the last decades, quantum optics has evolved from high quality factor cavities in the early experiments toward new cavity designs involving leaky modes. Despite very reliable models, in the concepts of cavity quantum electrodynamics, photon leakage is most of the time treated phenomenologically. Here, we take a different approach, and starting from first principles, we define an inside-outside representation which is derived from the original true-mode representation, in which one can determine effective Hamiltonian and Poynting vector. Contrary to the phenomenological model, they allow a full description of a leaking single photon produced in the cavity and propagating in free space. This is applied for a laser-driven atom-cavity system. In addition, we propose an atom-cavity non-resonant scheme for single photon generation, and we rigorously analyze the outgoing single photon in time and frequency domains for different coupling regimes. Finally, we introduce a particular coupling regime ensuring adiabatic elimination for which the pulse shape of the outgoing single photon is tailored using a specifically designed driving field envelope.

We study the diffusion of particles confined close to a single wall and in double-wall planar channel geometries where the local diffusivities depend on the distance to the boundaries. Displacement parallel to the walls is Brownian as characterized by its variance, but it is non-Gaussian having a non-zero fourth cumulant. Establishing a link with Taylor dispersion, we calculate the fourth cumulant and the tails of the displacement distribution for general diffusivity tensors along with potentials generated by either the walls or externally, for instance gravity. Experimental and numerical studies of the motion of a colloid in the direction parallel to the wall give measured fourth cumulants which are correctly predicted by our theory. Interestingly, contrary to models of Brownian-yet-non-Gaussian diffusion, the tails of the displacement distribution are shown to be Gaussian rather than exponential. All together, our results provide additional tests and constraints for the inference of force maps and local transport properties near surfaces.

Visual localization, i.e., camera pose estimation in a known scene, is a core component of technologies such as autonomous driving and augmented reality. State-of-the-art localization approaches often rely on image retrieval techniques for one of two purposes: (1) provide an approximate pose estimate or (2) determine which parts of the scene are potentially visible in a given query image. It is common practice to use state-of-the-art image retrieval algorithms for both of them. These algorithms are often trained for the goal of retrieving the same landmark under a large range of viewpoint changes which often differs from the requirements of visual localization. In order to investigate the consequences for visual localization, this paper focuses on understanding the role of image retrieval for multiple visual localization paradigms. First, we introduce a novel benchmark setup and compare state-of-the-art retrieval representations on multiple datasets using localization performance as metric. Second, we investigate several definitions of "ground truth" for image retrieval. Using these definitions as upper bounds for the visual localization paradigms, we show that there is still sgnificant room for improvement. Third, using these tools and in-depth analysis, we show that retrieval performance on classical landmark retrieval or place recognition tasks correlates only for some but not all paradigms to localization performance. Finally, we analyze the effects of blur and dynamic scenes in the images. We conclude that there is a need for retrieval approaches specifically designed for localization paradigms. Our benchmark and evaluation protocols are available at https://github.com/naver/kapture-localization., Comment: International Journal of Computer Vision (2022). arXiv admin note: text overlap with arXiv:2011.11946

In this paper we report on spatial intensity interferometry measurements within the H$\alpha$ line on two stars: the Luminous Blue Variable supergiant \PCygni\,and the late-type B supergiant Rigel. The experimental setup was upgraded to allow simultaneous measurement of two polarization channels, instead of one in our previous setup, and the zero baseline correlation function on-sky to validate independent estimates obtained from the stellar spectrum and the instrumental spectral throughput. Combined with simultaneous spectra measurements and based on radiative transfer models calculated with the code CMFGEN, we were able to fit our measured visibility curves to extract the stellar distances. Our distance determinations for both \PCygni\ (1.61 $\pm$ 0.18 kpc) and Rigel (0.26 $\pm$ 0.02 kpc) agree very well with the values provided by astrometry with the Gaia and Hipparcos missions, respectively. This result for Rigel was obtained by adopting a stellar luminosity of $L_{\star}$ = 123000 $L_{\odot}$, which is reported in the literature as being consistent with the Hipparcos distance to Rigel. However, due to the lack of consensus on Rigel's luminosity, we also explore how the adoption of the stellar luminosity in our models affects our distance determination for Rigel. In conclusion, we support, in an independent way, the distance to Rigel as the one provided by the Hipparcos mission, when taking the luminosity of 123000 $L_{\odot}$ at face value. This study is the first successful step towards extending the application of the Wind Momentum Luminosity Relation method for distance calibration from an LBV supergiant to a more normal late-type B supergiant., Comment: Monthly Notices of the Royal Astronomical Society: Letters, Oxford Journals, In press