Search

Neural Machine Translation models are extremely data and compute-hungry. However, not all data points contribute equally to model training and generalization. Data pruning to remove the low-value data points has the benefit of drastically reducing the compute budget without significant drop in model performance. In this paper, we propose a new data pruning technique: Checkpoints Across Time (CAT), that leverages early model training dynamics to identify the most relevant data points for model performance. We benchmark CAT against several data pruning techniques including COMET-QE, LASER and LaBSE. We find that CAT outperforms the benchmarks on Indo-European languages on multiple test sets. When applied to English-German, English-French and English-Swahili translation tasks, CAT achieves comparable performance to using the full dataset, while pruning up to 50% of training data. We inspect the data points that CAT selects and find that it tends to favour longer sentences and sentences with unique or rare words., Comment: Accepted to ACL 2024 Findings

Active Inference is a recent framework for modeling planning under uncertainty. Empirical and theoretical work have now begun to evaluate the strengths and weaknesses of this approach and how it might be improved. A recent extension - the sophisticated inference (SI) algorithm - improves performance on multi-step planning problems through recursive decision tree search. However, little work to date has been done to compare SI to other established planning algorithms. SI was also developed with a focus on inference as opposed to learning. The present paper has two aims. First, we compare performance of SI to Bayesian reinforcement learning (RL) schemes designed to solve similar problems. Second, we present an extension of SI - sophisticated learning (SL) - that more fully incorporates active learning during planning. SL maintains beliefs about how model parameters would change under the future observations expected under each policy. This allows a form of counterfactual retrospective inference in which the agent considers what could be learned from current or past observations given different future observations. To accomplish these aims, we make use of a novel, biologically inspired environment designed to highlight the problem structure for which SL offers a unique solution. Here, an agent must continually search for available (but changing) resources in the presence of competing affordances for information gain. Our simulations show that SL outperforms all other algorithms in this context - most notably, Bayes-adaptive RL and upper confidence bound algorithms, which aim to solve multi-step planning problems using similar principles (i.e., directed exploration and counterfactual reasoning). These results provide added support for the utility of Active Inference in solving this class of biologically-relevant problems and offer added tools for testing hypotheses about human cognition., Comment: 31 pages, 5 figures

We consider hate speech detection through keyword spotting on radio broadcasts. One approach is to build an automatic speech recognition (ASR) system for the target low-resource language. We compare this to using acoustic word embedding (AWE) models that map speech segments to a space where matching words have similar vectors. We specifically use a multilingual AWE model trained on labelled data from well-resourced languages to spot keywords in data in the unseen target language. In contrast to ASR, the AWE approach only requires a few keyword exemplars. In controlled experiments on Wolof and Swahili where training and test data are from the same domain, an ASR model trained on just five minutes of data outperforms the AWE approach. But in an in-the-wild test on Swahili radio broadcasts with actual hate speech keywords, the AWE model (using one minute of template data) is more robust, giving similar performance to an ASR system trained on 30 hours of labelled data., Comment: Accepted to Interspeech 2023

Radio interferometry calibration and Radio Frequency Interference (RFI) removal are usually done separately. Here we show that jointly modelling the antenna gains and RFI has significant benefits when the RFI follows precise trajectories, such as for satellites. One surprising benefit is improved calibration solutions, by leveraging the RFI signal itself. We present tabascal (TrAjectory BAsed RFI Subtraction and CALibration), a new algorithm that jointly models the RFI and calibration parameters in visibilities. We test tabascal on simulated MeerKAT calibration observations contaminated by satellite-based RFI. We obtain gain estimates that are both unbiased and up to an order of magnitude better constrained compared to uncontaminated data. When combined with an ad hoc RFI subtraction scheme, tabascal solutions can be further applied to an adjacent target observation: 5 minutes of calibration data results in an image with about a third the noise achieved when using flagging alone. The recovered flux distribution of RFI subtracted data was on par with uncontaminated data. In contrast, RFI flagging alone resulted in a higher detection threshold and consistent underestimation of source fluxes. For a mean RFI amplitude of 17 Jy, using RFI subtraction leads to less than 1% loss of data compared to 75% data loss from an ideal $3\sigma$ flagging algorithm, a very significant increase in data available for science analysis. Although we have examined the case of satellite RFI, tabascal should work for any RFI moving on parameterizable trajectories, relative to the phase centre, such as planes and/or objects fixed to the ground., Comment: 21 pages, 17 figures, 4 tables

Radio Frequency Interference (RFI) is unwanted noise that swamps the desired astronomical signal. Radio astronomers have always had to deal with RFI detection and excision around telescope sites, but little has been done to understand the full scope, nature and evolution of RFI in a unified way. We undertake this for the MeerKAT array using a probabilistic multidimensional framework approach focussing on UHF-band and L-band data. In the UHF- band, RFI is dominated by the allocated Global System for Mobile (GSM) Communications, flight Distance Measuring Equipment (DME), and UHF-TV bands. The L-band suffers from known RFI sources such as DMEs, GSM, and the Global Positioning System (GPS) satellites. In the "clean" MeerKAT band, we noticed the RFI occupancy changing with time and direction for both the L-band and UHF band. For example, we saw a significant increase (300% increase) in the fraction of L-band flagged data in November 2018 compared to June 2018. This increase seems to correlate with construction activity on site. In the UHF-band, we found that the early morning is least impacted by RFI and other outliers. We also found a dramatic decrease in DME RFI during the hard lockdown due to the COVID-19 pandemic. The work presented here allows us to characterise the evolution of RFI at the MeerKAT site. Any observatory can adopt it to understand the behaviour of RFI within its surroundings.

Language-agnostic sentence embeddings generated by pre-trained models such as LASER and LaBSE are attractive options for mining large datasets to produce parallel corpora for low-resource machine translation. We test LASER and LaBSE in extracting bitext for two related low-resource African languages: Luhya and Swahili. For this work, we created a new parallel set of nearly 8000 Luhya-English sentences which allows a new zero-shot test of LASER and LaBSE. We find that LaBSE significantly outperforms LASER on both languages. Both LASER and LaBSE however perform poorly at zero-shot alignment on Luhya, achieving just 1.5% and 22.0% successful alignments respectively (P@1 score). We fine-tune the embeddings on a small set of parallel Luhya sentences and show significant gains, improving the LaBSE alignment accuracy to 53.3%. Further, restricting the dataset to sentence embedding pairs with cosine similarity above 0.7 yielded alignments with over 85% accuracy., Comment: Accepted to LoResMT 2022

Anomaly detection algorithms are typically applied to static, unchanging, data features hand-crafted by the user. But how does a user systematically craft good features for anomalies that have never been seen? Here we couple deep learning with active learning -- in which an Oracle iteratively labels small amounts of data selected algorithmically over a series of rounds -- to automatically and dynamically improve the data features for efficient outlier detection. This approach, AHUNT, shows excellent performance on MNIST, CIFAR10, and Galaxy-DESI data, significantly outperforming both standard anomaly detection and active learning algorithms with static feature spaces. Beyond improved performance, AHUNT also allows the number of anomaly classes to grow organically in response to Oracle's evaluations. Extensive ablation studies explore the impact of Oracle question selection strategy and loss function on performance. We illustrate how the dynamic anomaly class taxonomy represents another step towards fully personalized rankings of different anomaly classes that reflect a user's interests, allowing the algorithm to learn to ignore statistically significant but uninteresting outliers (e.g., noise). This should prove useful in the era of massive astronomical datasets serving diverse sets of users who can only review a tiny subset of the incoming data., Comment: 10 pages, 7 figures

Active learning aims to deliver maximum benefit when resources are scarce. We use COMET-QE, a reference-free evaluation metric, to select sentences for low-resource neural machine translation. Using Swahili, Kinyarwanda and Spanish for our experiments, we show that COMET-QE significantly outperforms two variants of Round Trip Translation Likelihood (RTTL) and random sentence selection by up to 5 BLEU points for 20k sentences selected by Active Learning on a 30k baseline. This suggests that COMET-QE is a powerful tool for sentence selection in the very low-resource limit., Comment: Accepted to Findings of EMNLP 2022

The next generation of telescopes such as the SKA and the Rubin Observatory will produce enormous data sets, requiring automated anomaly detection to enable scientific discovery. Here, we present an overview and friendly user guide to the Astronomaly framework for active anomaly detection in astronomical data. Astronomaly uses active learning to combine the raw processing power of machine learning with the intuition and experience of a human user, enabling personalised recommendations of interesting anomalies. It makes use of a Python backend to perform data processing, feature extraction and machine learning to detect anomalous objects; and a JavaScript frontend to allow interaction with the data, labelling of interesting anomalous and active learning. Astronomaly is designed to be modular, extendable and run on almost any type of astronomical data. In this paper, we detail the structure of the Astronomaly code and provide guidelines for basic usage., Comment: 10 pages, 3 figures, to be published in Proc. ADASS XXXI (2021), coincides with v1.2 release of Astronomaly (https://github.com/MichelleLochner/astronomaly)

The Hydrogen Intensity and Real-time Analysis eXperiment (HIRAX) is a radio interferometer array currently in development, with an initial 256-element array to be deployed at the South African Radio Astronomy Observatory (SARAO) Square Kilometer Array (SKA) site in South Africa. Each of the 6m, $f/0.23$ dishes will be instrumented with dual-polarisation feeds operating over a frequency range of 400-800 MHz. Through intensity mapping of the 21 cm emission line of neutral hydrogen, HIRAX will provide a cosmological survey of the distribution of large-scale structure over the redshift range of $0.775 < z < 2.55$ over $\sim$15,000 square degrees of the southern sky. The statistical power of such a survey is sufficient to produce $\sim$7 percent constraints on the dark energy equation of state parameter when combined with measurements from the Planck satellite. Additionally, HIRAX will provide a highly competitive platform for radio transient and HI absorber science while enabling a multitude of cross-correlation studies. In this paper, we describe the science goals of the experiment, overview of the design and status of the sub-components of the telescope system, and describe the expected performance of the initial 256-element array as well as the planned future expansion to the final, 1024-element array., Comment: 24 pages, 6 figures. Updated to match published version

Low-resource African languages have not fully benefited from the progress in neural machine translation because of a lack of data. Motivated by this challenge we compare zero-shot learning, transfer learning and multilingual learning on three Bantu languages (Shona, isiXhosa and isiZulu) and English. Our main target is English-to-isiZulu translation for which we have just 30,000 sentence pairs, 28% of the average size of our other corpora. We show the importance of language similarity on the performance of English-to-isiZulu transfer learning based on English-to-isiXhosa and English-to-Shona parent models whose BLEU scores differ by 5.2. We then demonstrate that multilingual learning surpasses both transfer learning and zero-shot learning on our dataset, with BLEU score improvements relative to the baseline English-to-isiZulu model of 9.9, 6.1 and 2.0 respectively. Our best model also improves the previous SOTA BLEU score by more than 10.

Survey telescopes such as the Vera C. Rubin Observatory and the Square Kilometre Array will discover billions of static and dynamic astronomical sources. Properly mined, these enormous datasets will likely be wellsprings of rare or unknown astrophysical phenomena. The challenge is that the datasets are so large that most data will never be seen by human eyes; currently the most robust instrument we have to detect relevant anomalies. Machine learning is a useful tool for anomaly detection in this regime. However, it struggles to distinguish between interesting anomalies and irrelevant data such as instrumental artefacts or rare astronomical sources that are simply not of interest to a particular scientist. Active learning combines the flexibility and intuition of the human brain with the raw processing power of machine learning. By strategically choosing specific objects for expert labelling, it minimises the amount of data that scientists have to look through while maximising potential scientific return. Here we introduce Astronomaly: a general anomaly detection framework with a novel active learning approach designed to provide personalised recommendations. Astronomaly can operate on most types of astronomical data, including images, light curves and spectra. We use the Galaxy Zoo dataset to demonstrate the effectiveness of Astronomaly, as well as simulated data to thoroughly test our new active learning approach. We find that for both datasets, Astronomaly roughly doubles the number of interesting anomalies found in the first 100 objects viewed by the user. Astronomaly is easily extendable to include new feature extraction techniques, anomaly detection algorithms and even different active learning approaches. The code is publicly available at https://github.com/MichelleLochner/astronomaly., Comment: 26 pages, 11 figures. Matches published version. 2 minute video introducing Astronomaly: https://youtu.be/vyU0gJFiFxA

Deep facial expression recognition faces two challenges that both stem from the large number of trainable parameters: long training times and a lack of interpretability. We propose a novel method based on evolutionary algorithms, that deals with both challenges by massively reducing the number of trainable parameters, whilst simultaneously retaining classification performance, and in some cases achieving superior performance. We are robustly able to reduce the number of parameters on average by 95% (e.g. from 2M to 100k parameters) with no loss in classification accuracy. The algorithm learns to choose small patches from the image, relative to the nose, which carry the most important information about emotion, and which coincide with typical human choices of important features. Our work implements a novel form attention and shows that evolutionary algorithms are a valuable addition to machine learning in the deep learning era, both for reducing the number of parameters for facial expression recognition and for providing interpretable features that can help reduce bias., Comment: Conference of the South African Institute of Computer Scientists and Information Technologists 2020

Radio Frequency Interference (RFI) has historically plagued radio astronomy, worsening with the rapid spread of electronics and increasing telescope sensitivity. We present a multi-dimensional probabilistic framework for characterising the RFI environment around a radio astronomy site that uses automatically flagged data from the array itself. We illustrate the framework using about 1500 hours of commissioning data from the MeerKAT radio telescope; producing a 6-dimensional array that yields both average RFI occupancy as well as confidence intervals around the mean as a function of key variables (frequency, direction, baseline, time). Our results provide the first detailed view of the MeerKAT RFI environment at high sensitivity as a function of direction, frequency, time of day and baseline. They allow us to track the historical evolution of the RFI and to quantify fluctuations which can be used for alerting on new RFI. As expected we find the major RFI contributors for MeerKAT site are from Global Positioning System (GPS) satellites, flight Distance Measurement Equipment (DME) and the Global System for Mobile (GSM) Communications. Beyond characterising RFI environments our approach allows observers access to the prior probability of RFI in any combination of tracked variables, allowing for more efficient observation planning and data excision.

Multiple studies have suggested the spread of COVID-19 is affected by factors such as climate, BCG vaccinations, pollution and blood type. We perform a joint study of these factors using the death growth rates of 40 regions worldwide with both machine learning and Bayesian methods. We find weak, non-significant (< 3$\sigma$) evidence for temperature and relative humidity as factors in the spread of COVID-19 but little or no evidence for BCG vaccination prevalence or $\text{PM}_{2.5}$ pollution. The only variable detected at a statistically significant level (>3$\sigma$) is the rate of positive COVID-19 tests, with higher positive rates correlating with higher daily growth of deaths., Comment: 17 pages, 10 figures, 6 tables

Flagging of Radio Frequency Interference (RFI) is an increasingly important challenge in radio astronomy. We present R-Net, a deep convolutional ResNet architecture that significantly outperforms existing algorithms -- including the default MeerKAT RFI flagger, and deep U-Net architectures -- across all metrics including AUC, F1-score and MCC. We demonstrate the robustness of this improvement on both single dish and interferometric simulations and, using transfer learning, on real data. Our R-Net model's precision is approximately $90\%$ better than the current MeerKAT flagger at $80\%$ recall and has a 35\% higher F1-score with no additional performance cost. We further highlight the effectiveness of transfer learning from a model initially trained on simulated MeerKAT data and fine-tuned on real, human-flagged, KAT-7 data. Despite the wide differences in the nature of the two telescope arrays, the model achieves an AUC of 0.91, while the best model without transfer learning only reaches an AUC of 0.67. We consider the use of phase information in our models but find that without calibration the phase adds almost no extra information relative to amplitude data only. Our results strongly suggest that deep learning on simulations, boosted by transfer learning on real data, will likely play a key role in the future of RFI flagging of radio astronomy data., Comment: 12 pages, 15 figures. Published on MNRAS

Anomaly detection is challenging, especially for large datasets in high dimensions. Here we explore a general anomaly detection framework based on dimensionality reduction and unsupervised clustering. We release DRAMA, a general python package that implements the general framework with a wide range of built-in options. We test DRAMA on a wide variety of simulated and real datasets, in up to 3000 dimensions, and find it robust and highly competitive with commonly-used anomaly detection algorithms, especially in high dimensions. The flexibility of the DRAMA framework allows for significant optimization once some examples of anomalies are available, making it ideal for online anomaly detection, active learning and highly unbalanced datasets., Comment: 6 pages

Statistical uncertainties are rarely incorporated in machine learning algorithms, especially for anomaly detection. Here we present the Bayesian Anomaly Detection And Classification (BADAC) formalism, which provides a unified statistical approach to classification and anomaly detection within a hierarchical Bayesian framework. BADAC deals with uncertainties by marginalising over the unknown, true, value of the data. Using simulated data with Gaussian noise, BADAC is shown to be superior to standard algorithms in both classification and anomaly detection performance in the presence of uncertainties, though with significantly increased computational cost. Additionally, BADAC provides well-calibrated classification probabilities, valuable for use in scientific pipelines. We show that BADAC can work in online mode and is fairly robust to model errors, which can be diagnosed through model-selection methods. In addition it can perform unsupervised new class detection and can naturally be extended to search for anomalous subsets of data. BADAC is therefore ideal where computational cost is not a limiting factor and statistical rigour is important. We discuss approximations to speed up BADAC, such as the use of Gaussian processes, and finally introduce a new metric, the Rank-Weighted Score (RWS), that is particularly suited to evaluating the ability of algorithms to detect anomalies., Comment: 29 pages, 13 figures, Demo available: https://github.com/ethyroberts/BADAC

With the advent of powerful telescopes such as the Square Kilometer Array and the Vera C. Rubin Observatory, we are entering an era of multiwavelength transient astronomy that will lead to a dramatic increase in data volume. Machine learning techniques are well suited to address this data challenge and rapidly classify newly detected transients. We present a multiwavelength classification algorithm consisting of three steps: (1) interpolation and augmentation of the data using Gaussian processes; (2) feature extraction using wavelets; and (3) classification with random forests. Augmentation provides improved performance at test time by balancing the classes and adding diversity into the training set. In the first application of machine learning to the classification of real radio transient data, we apply our technique to the Green Bank Interferometer and other radio light curves. We find we are able to accurately classify most of the 11 classes of radio variables and transients after just eight hours of observations, achieving an overall test accuracy of 78 percent. We fully investigate the impact of the small sample size of 82 publicly available light curves and use data augmentation techniques to mitigate the effect. We also show that on a significantly larger simulated representative training set that the algorithm achieves an overall accuracy of 97 percent, illustrating that the method is likely to provide excellent performance on future surveys. Finally, we demonstrate the effectiveness of simultaneous multiwavelength observations by showing how incorporating just one optical data point into the analysis improves the accuracy of the worst performing class by 19 percent., Comment: 16 pages, 12 figures

Point source detection at low signal-to-noise is challenging for astronomical surveys, particularly in radio interferometry images where the noise is correlated. Machine learning is a promising solution, allowing the development of algorithms tailored to specific telescope arrays and science cases. We present DeepSource - a deep learning solution - that uses convolutional neural networks to achieve these goals. DeepSource enhances the Signal-to-Noise Ratio (SNR) of the original map and then uses dynamic blob detection to detect sources. Trained and tested on two sets of 500 simulated 1 deg x 1 deg MeerKAT images with a total of 300,000 sources, DeepSource is essentially perfect in both purity and completeness down to SNR = 4 and outperforms PyBDSF in all metrics. For uniformly-weighted images it achieves a Purity x Completeness (PC) score at SNR = 3 of 0.73, compared to 0.31 for the best PyBDSF model. For natural-weighting we find a smaller improvement of ~40% in the PC score at SNR = 3. If instead we ask where either of the purity or completeness first drop to 90%, we find that DeepSource reaches this value at SNR = 3.6 compared to the 4.3 of PyBDSF (natural-weighting). A key advantage of DeepSource is that it can learn to optimally trade off purity and completeness for any science case under consideration. Our results show that deep learning is a promising approach to point source detection in astronomical images., Comment: 15 pages, 13 figures, submitted to MNRAS

The ability to identify sentiment in text, referred to as sentiment analysis, is one which is natural to adult humans. This task is, however, not one which a computer can perform by default. Identifying sentiments in an automated, algorithmic manner will be a useful capability for business and research in their search to understand what consumers think about their products or services and to understand human sociology. Here we propose two new Genetic Algorithms (GAs) for the task of automated text sentiment analysis. The GAs learn whether words occurring in a text corpus are either sentiment or amplifier words, and their corresponding magnitude. Sentiment words, such as 'horrible', add linearly to the final sentiment. Amplifier words in contrast, which are typically adjectives/adverbs like 'very', multiply the sentiment of the following word. This increases, decreases or negates the sentiment of the following word. The sentiment of the full text is then the sum of these terms. This approach grows both a sentiment and amplifier dictionary which can be reused for other purposes and fed into other machine learning algorithms. We report the results of multiple experiments conducted on large Amazon data sets. The results reveal that our proposed approach was able to outperform several public and/or commercial sentiment analysis algorithms., Comment: In "2017 Annual Conference of the South African Institute of Computer Scientists and Information"

We develop a machine learning (ML) framework to populate large dark matter-only simulations with baryonic galaxies. Our ML framework takes input halo properties including halo mass, environment, spin, and recent growth history, and outputs central galaxy and halo baryonic properties including stellar mass ($M_*$), star formation rate (SFR), metallicity ($Z$), neutral ($\rm HI$) and molecular ($\rm H_2$) hydrogen mass. We apply this to the MUFASA cosmological hydrodynamic simulation, and show that it recovers the mean trends of output quantities with halo mass highly accurately, including following the sharp drop in SFR and gas in quenched massive galaxies. However, the scatter around the mean relations is under-predicted. Examining galaxies individually, at $z=0$ the stellar mass and metallicity are accurately recovered ($\sigma\lesssim 0.2$~dex), but SFR and $\rm HI$ show larger scatter ($\sigma\gtrsim 0.3$~dex); these values improve somewhat at $z=1,2$. Remarkably, ML quantitatively recovers second parameter trends in galaxy properties, e.g. that galaxies with higher gas content and lower metallicity have higher SFR at a given $M_*$. Testing various ML algorithms, we find that none perform significantly better than the others, nor does ensembling improve performance, likely because none of the algorithms reproduce the large observed scatter around the mean properties. For the random forest algorithm, we find that halo mass and nearby ($\sim 200$~kpc) environment are the most important predictive variables followed by growth history, while halo spin and $\sim$Mpc scale environment are not important. Finally we study the impact of additionally inputting key baryonic properties $M_*$, SFR and $Z$, as would be available e.g. from an equilibrium model, and show that particularly providing the SFR enables $\rm HI$ to be recovered substantially more accurately., Comment: 15 pages, 10 figures, 1 table, accepted version from MNRAS

Deep neural networks continue to show improved performance with increasing depth, an encouraging trend that implies an explosion in the possible permutations of network architectures and hyperparameters for which there is little intuitive guidance. To address this increasing complexity, we propose Evolutionary DEep Networks (EDEN), a computationally efficient neuro-evolutionary algorithm which interfaces to any deep neural network platform, such as TensorFlow. We show that EDEN evolves simple yet successful architectures built from embedding, 1D and 2D convolutional, max pooling and fully connected layers along with their hyperparameters. Evaluation of EDEN across seven image and sentiment classification datasets shows that it reliably finds good networks -- and in three cases achieves state-of-the-art results -- even on a single GPU, in just 6-24 hours. Our study provides a first attempt at applying neuro-evolution to the creation of 1D convolutional networks for sentiment analysis including the optimisation of the embedding layer., Comment: 7 pages, 3 figures, 3 tables and see video https://vimeo.com/234510097

Can textual data be compressed intelligently without losing accuracy in evaluating sentiment? In this study, we propose a novel evolutionary compression algorithm, PARSEC (PARts-of-Speech for sEntiment Compression), which makes use of Parts-of-Speech tags to compress text in a way that sacrifices minimal classification accuracy when used in conjunction with sentiment analysis algorithms. An analysis of PARSEC with eight commercial and non-commercial sentiment analysis algorithms on twelve English sentiment data sets reveals that accurate compression is possible with (0%, 1.3%, 3.3%) loss in sentiment classification accuracy for (20%, 50%, 75%) data compression with PARSEC using LingPipe, the most accurate of the sentiment algorithms. Other sentiment analysis algorithms are more severely affected by compression. We conclude that significant compression of text data is possible for sentiment analysis depending on the accuracy demands of the specific application and the specific sentiment analysis algorithm used., Comment: 8 pages, 2 figures, 8 tables

We discuss the ground-breaking science that will be possible with a wide area survey, using the MeerKAT telescope, known as MeerKLASS (MeerKAT Large Area Synoptic Survey). The current specifications of MeerKAT make it a great fit for science applications that require large survey speeds but not necessarily high angular resolutions. In particular, for cosmology, a large survey over $\sim 4,000 \, {\rm deg}^2$ for $\sim 4,000$ hours will potentially provide the first ever measurements of the baryon acoustic oscillations using the 21cm intensity mapping technique, with enough accuracy to impose constraints on the nature of dark energy. The combination with multi-wavelength data will give unique additional information, such as exquisite constraints on primordial non-Gaussianity using the multi-tracer technique, as well as a better handle on foregrounds and systematics. Such a wide survey with MeerKAT is also a great match for HI galaxy studies, providing unrivalled statistics in the pre-SKA era for galaxies resolved in the HI emission line beyond local structures at z > 0.01. It will also produce a large continuum galaxy sample down to a depth of about 5\,$\mu$Jy in L-band, which is quite unique over such large areas and will allow studies of the large-scale structure of the Universe out to high redshifts, complementing the galaxy HI survey to form a transformational multi-wavelength approach to study galaxy dynamics and evolution. Finally, the same survey will supply unique information for a range of other science applications, including a large statistical investigation of galaxy clusters as well as produce a rotation measure map across a huge swathe of the sky. The MeerKLASS survey will be a crucial step on the road to using SKA1-MID for cosmological applications and other commensal surveys, as described in the top priority SKA key science projects (abridged)., Comment: Larger version of the paper submitted to the Proceedings of Science, "MeerKAT Science: On the Pathway to the SKA", Stellenbosch, 25-27 May 2016

The MIGHTEE large survey project will survey four of the most well-studied extragalactic deep fields, totalling 20 square degrees to $\mu$Jy sensitivity at Giga-Hertz frequencies, as well as an ultra-deep image of a single ~1 square degree MeerKAT pointing. The observations will provide radio continuum, spectral line and polarisation information. As such, MIGHTEE, along with the excellent multi-wavelength data already available in these deep fields, will allow a range of science to be achieved. Specifically, MIGHTEE is designed to significantly enhance our understanding of, (i) the evolution of AGN and star-formation activity over cosmic time, as a function of stellar mass and environment, free of dust obscuration; (ii) the evolution of neutral hydrogen in the Universe and how this neutral gas eventually turns into stars after moving through the molecular phase, and how efficiently this can fuel AGN activity; (iii) the properties of cosmic magnetic fields and how they evolve in clusters, filaments and galaxies. MIGHTEE will reach similar depth to the planned SKA all-sky survey, and thus will provide a pilot to the cosmology experiments that will be carried out by the SKA over a much larger survey volume., Comment: Proceedings of Science, "MeerKAT Science: On the Pathway to the SKA", Stellenbosch, 25-27 May 2016

Regression or classification? This is perhaps the most basic question faced when tackling a new supervised learning problem. We present an Evolutionary Deep Learning (EDL) algorithm that automatically solves this by identifying the question type with high accuracy, along with a proposed deep architecture. Typically, a significant amount of human insight and preparation is required prior to executing machine learning algorithms. For example, when creating deep neural networks, the number of parameters must be selected in advance and furthermore, a lot of these choices are made based upon pre-existing knowledge of the data such as the use of a categorical cross entropy loss function. Humans are able to study a dataset and decide whether it represents a classification or a regression problem, and consequently make decisions which will be applied to the execution of the neural network. We propose the Automated Problem Identification (API) algorithm, which uses an evolutionary algorithm interface to TensorFlow to manipulate a deep neural network to decide if a dataset represents a classification or a regression problem. We test API on 16 different classification, regression and sentiment analysis datasets with up to 10,000 features and up to 17,000 unique target values. API achieves an average accuracy of $96.3\%$ in identifying the problem type without hardcoding any insights about the general characteristics of regression or classification problems. For example, API successfully identifies classification problems even with 1000 target values. Furthermore, the algorithm recommends which loss function to use and also recommends a neural network architecture. Our work is therefore a step towards fully automated machine learning., Comment: 9 pages, 6 figures, 4 tables

Supernova cosmology without spectra will be an important component of future surveys such as LSST. This lack of supernova spectra results in uncertainty in the redshifts which, if ignored, leads to significantly biased estimates of cosmological parameters. Here we present a hierarchical Bayesian formalism -- zBEAMS -- that addresses this problem by marginalising over the unknown or uncertain supernova redshifts to produce unbiased cosmological estimates that are competitive with supernova data with spectroscopically confirmed redshifts. zBEAMS provides a unified treatment of both photometric redshifts and host galaxy misidentification (occurring due to chance galaxy alignments or faint hosts), effectively correcting the inevitable contamination in the Hubble diagram. Like its predecessor BEAMS, our formalism also takes care of non-Ia supernova contamination by marginalising over the unknown supernova type. We illustrate this technique with simulations of supernovae with photometric redshifts and host galaxy misidentification. A novel feature of the photometric redshift case is the important role played by the redshift distribution of the supernovae., Comment: Matches published version, 18 pages, 6 figures, Code available: https://github.com/MichelleLochner/zBEAMS

We outline a new method to compute the Bayes Factor for model selection which bypasses the Bayesian Evidence. Our method combines multiple models into a single, nested, Supermodel using one or more hyperparameters. Since the models are now nested the Bayes Factors between the models can be efficiently computed using the Savage-Dickey Density Ratio (SDDR). In this way model selection becomes a problem of parameter estimation. We consider two ways of constructing the supermodel in detail: one based on combined models, and a second based on combined likelihoods. We report on these two approaches for a Gaussian linear model for which the Bayesian evidence can be calculated analytically and a toy nonlinear problem. Unlike the combined model approach, where a standard Monte Carlo Markov Chain (MCMC) struggles, the combined-likelihood approach fares much better in providing a reliable estimate of the log-Bayes Factor. This scheme potentially opens the way to computationally efficient ways to compute Bayes Factors in high dimensions that exploit the good scaling properties of MCMC, as compared to methods such as nested sampling that fail for high dimensions., Comment: 24 pages, 11 Figures

Bayesian graphical models are an efficient tool for modelling complex data and derive self-consistent expressions of the posterior distribution of model parameters. We apply Bayesian graphs to perform statistical analyses of Type Ia supernova (SN Ia) luminosity distance measurements from the joint light-curve analysis (JLA) data set. In contrast to the $\chi^2$ approach used in previous studies, the Bayesian inference allows us to fully account for the standard-candle parameter dependence of the data covariance matrix. Comparing with $\chi^2$ analysis results, we find a systematic offset of the marginal model parameter bounds. We demonstrate that the bias is statistically significant in the case of the SN Ia standardization parameters with a maximal 6 $\sigma$ shift of the SN light-curve colour correction. In addition, we find that the evidence for a host galaxy correction is now only 2.4 $\sigma$. Systematic offsets on the cosmological parameters remain small, but may increase by combining constraints from complementary cosmological probes. The bias of the $\chi^2$ analysis is due to neglecting the parameter-dependent log-determinant of the data covariance, which gives more statistical weight to larger values of the standardization parameters. We find a similar effect on compressed distance modulus data. To this end, we implement a fully consistent compression method of the JLA data set that uses a Gaussian approximation of the posterior distribution for fast generation of compressed data. Overall, the results of our analysis emphasize the need for a fully consistent Bayesian statistical approach in the analysis of future large SN Ia data sets., Comment: 14 pages, 13 figures, 5 tables. Submitted to MNRAS. Compression utility available at https://gitlab.com/congma/libsncompress/ and example cosmology code with machine-readable version of Tables A1 & A2 at https://gitlab.com/congma/sn-bayesian-model-example/ v2: corrected typo in author's name. v3: 15 pages, incl. corrections, matches the accepted version

Radio interferometers suffer from the problem of missing information in their data, due to the gaps between the antennas. This results in artifacts, such as bright rings around sources, in the images obtained. Multiple deconvolution algorithms have been proposed to solve this problem and produce cleaner radio images. However, these algorithms are unable to correctly estimate uncertainties in derived scientific parameters or to always include the effects of instrumental errors. We propose an alternative technique called Bayesian Inference for Radio Observations (BIRO) which uses a Bayesian statistical framework to determine the scientific parameters and instrumental errors simultaneously directly from the raw data, without making an image. We use a simple simulation of Westerbork Synthesis Radio Telescope data including pointing errors and beam parameters as instrumental effects, to demonstrate the use of BIRO., Comment: Conference proceedings of IAU Symposium 306. 3 pages, 4 figures

Classifying transients based on multi band light curves is a challenging but crucial problem in the era of GAIA and LSST since the sheer volume of transients will make spectroscopic classification unfeasible. Here we present a nonparametric classifier that uses the transient's light curve measurements to predict its class given training data. It implements two novel components: the first is the use of the BAGIDIS wavelet methodology - a characterization of functional data using hierarchical wavelet coefficients. The second novelty is the introduction of a ranked probability classifier on the wavelet coefficients that handles both the heteroscedasticity of the data in addition to the potential non-representativity of the training set. The ranked classifier is simple and quick to implement while a major advantage of the BAGIDIS wavelets is that they are translation invariant, hence they do not need the light curves to be aligned to extract features. Further, BAGIDIS is nonparametric so it can be used for blind searches for new objects. We demonstrate the effectiveness of our ranked wavelet classifier against the well-tested Supernova Photometric Classification Challenge dataset in which the challenge is to correctly classify light curves as Type Ia or non-Ia supernovae. We train our ranked probability classifier on the spectroscopically-confirmed subsample (which is not representative) and show that it gives good results for all supernova with observed light curve timespans greater than 100 days (roughly 55% of the dataset). For such data, we obtain a Ia efficiency of 80.5% and a purity of 82.4% yielding a highly competitive score of 0.49 whilst implementing a truly "model-blind" approach to supernova classification. Consequently this approach may be particularly suitable for the classification of astronomical transients in the era of large synoptic sky surveys., Comment: 14 pages, 8 figures. Published in MNRAS

New telescopes like the Square Kilometre Array (SKA) will push into a new sensitivity regime and expose systematics, such as direction-dependent effects, that could previously be ignored. Current methods for handling such systematics rely on alternating best estimates of instrumental calibration and models of the underlying sky, which can lead to inadequate uncertainty estimates and biased results because any correlations between parameters are ignored. These deconvolution algorithms produce a single image that is assumed to be a true representation of the sky, when in fact it is just one realization of an infinite ensemble of images compatible with the noise in the data. In contrast, here we report a Bayesian formalism that simultaneously infers both systematics and science. Our technique, Bayesian Inference for Radio Observations (BIRO), determines all parameters directly from the raw data, bypassing image-making entirely, by sampling from the joint posterior probability distribution. This enables it to derive both correlations and accurate uncertainties, making use of the flexible software MEQTREES to model the sky and telescope simultaneously. We demonstrate BIRO with two simulated sets of Westerbork Synthesis Radio Telescope data sets. In the first, we perform joint estimates of 103 scientific (flux densities of sources) and instrumental (pointing errors, beamwidth and noise) parameters. In the second example, we perform source separation with BIRO. Using the Bayesian evidence, we can accurately select between a single point source, two point sources and an extended Gaussian source, allowing for 'super-resolution' on scales much smaller than the synthesized beam., Comment: Published in MNRAS. See https://vimeo.com/117391380 for a video of MultiNest converging to the correct source model

This paper describes the data release of the Sloan Digital Sky Survey-II (SDSS-II) Supernova Survey conducted between 2005 and 2007. Light curves, spectra, classifications, and ancillary data are presented for 10,258 variable and transient sources discovered through repeat ugriz imaging of SDSS Stripe 82, a 300 deg2 area along the celestial equator. This data release is comprised of all transient sources brighter than r~22.5 mag with no history of variability prior to 2004. Dedicated spectroscopic observations were performed on a subset of 889 transients, as well as spectra for thousands of transient host galaxies using the SDSS-III BOSS spectrographs. Photometric classifications are provided for the candidates with good multi-color light curves that were not observed spectroscopically. From these observations, 4607 transients are either spectroscopically confirmed, or likely to be, supernovae, making this the largest sample of supernova candidates ever compiled. We present a new method for SN host-galaxy identification and derive host-galaxy properties including stellar masses, star-formation rates, and the average stellar population ages from our SDSS multi-band photometry. We derive SALT2 distance moduli for a total of 1443 SN Ia with spectroscopic redshifts as well as photometric redshifts for a further 677 purely-photometric SN Ia candidates. Using the spectroscopically confirmed subset of the three-year SDSS-II SN Ia sample and assuming a flat Lambda-CDM cosmology, we determine Omega_M = 0.315 +/- 0.093 (statistical error only) and detect a non-zero cosmological constant at 5.7 sigmas., Comment: Submitted to ApJS. Full catalogs and datafiles are available here: http://sdssdp62.fnal.gov/sdsssn/DataRelease/index.html

There are two redshifts in cosmology: $z_{obs}$, the observed redshift computed via spectral lines, and the model redshift, $z$, defined by the effective FLRW scale factor. In general these do not coincide. We place observational constraints on the allowed distortions of $z$ away from $z_{obs}$ - a possibility we dub redshift remapping. Remapping is degenerate with cosmic dynamics for either $d_L(z)$ or $H(z)$ observations alone: for example, the simple remapping $z = \alpha_1 z_{obs} +\alpha_2 z_{obs}^2$ allows a decelerating Einstein de Sitter universe to fit the observed supernova Hubble diagram as successfully as $\Lambda$CDM, highlighting that supernova data alone cannot prove that the universe is accelerating. We show however, that redshift remapping leads to apparent violations of cosmic distance duality that can be used to detect its presence even when neither a specific theory of gravity nor the Copernican Principle are assumed. Combining current data sets favours acceleration but does not yet rule out redshift remapping as an alternative to dark energy. Future surveys, however, will provide exquisite constraints on remapping and any models -- such as backreaction -- that predict it., Comment: 12 pages 13 figures. Significantly generalised to discuss the case where distances and Hubble rate are arbitrarily deformed and General Relativity is not assumed

We present the spectroscopy from 5254 galaxies that hosted supernovae (SNe) or other transient events in the Sloan Digital Sky Survey II (SDSS-II). Obtained during SDSS-I, SDSS-II, and the Baryon Oscillation Spectroscopic Survey (BOSS), this sample represents the largest systematic, unbiased, magnitude limited spectroscopic survey of supernova (SN) host galaxies. Using the host galaxy redshifts, we test the impact of photometric SN classification based on SDSS imaging data with and without using spectroscopic redshifts of the host galaxies. Following our suggested scheme, there are a total of 1166 photometrically classified SNe Ia when using a flat redshift prior and 1126 SNe Ia when the host spectroscopic redshift is assumed. For 1024 (87.8%) candidates classified as likely SNe Ia without redshift information, we find that the classification is unchanged when adding the host galaxy redshift. Using photometry from SDSS imaging data and the host galaxy spectra, we also report host galaxy properties for use in future nalysis of SN astrophysics. Finally, we investigate the differences in the interpretation of the light curve properties with and without knowledge of the redshift. When using the SALT2 light curve fitter, we find a 21% increase in the number of fits that converge when using the spectroscopic redshift. Without host galaxy redshifts, we find that SALT2 light curve fits are systematically biased towards lower photometric redshift estimates and redder colors in the limit of low signal-to-noise data. The general improvements in performance of the light curve fitter and the increased diversity of the host galaxy sample highlights the importance of host galaxy spectroscopy for current photometric SN surveys such as the Dark Energy Survey and future surveys such as the Large Synoptic Survey Telescope., Comment: 27 pages, 22 figures, 10 tables

Using a sample of 608 Type Ia supernovae from the SDSS-II and BOSS surveys, combined with a sample of foreground galaxies from SDSS-II, we estimate the weak lensing convergence for each supernova line-of-sight. We find that the correlation between this measurement and the Hubble residuals is consistent with the prediction from lensing (at a significance of 1.7sigma. Strong correlations are also found between the residuals and supernova nuisance parameters after a linear correction is applied. When these other correlations are taken into account, the lensing signal is detected at 1.4sigma. We show for the first time that distance estimates from supernovae can be improved when lensing is incorporated by including a new parameter in the SALT2 methodology for determining distance moduli. The recovered value of the new parameter is consistent with the lensing prediction. Using CMB data from WMAP7, H0 data from HST and SDSS BAO measurements, we find the best-fit value of the new lensing parameter and show that the central values and uncertainties on Omega_m and w are unaffected. The lensing of supernovae, while only seen at marginal significance in this low redshift sample, will be of vital importance for the next generation of surveys, such as DES and LSST, which will be systematics dominated., Comment: Minor changes. Accepted for publication in ApJ

For future surveys, spectroscopic follow-up for all supernovae will be extremely difficult. However, one can use light curve fitters, to obtain the probability that an object is a Type Ia. One may consider applying a probability cut to the data, but we show that the resulting non-Ia contamination can lead to biases in the estimation of cosmological parameters. A different method, which allows the use of the full dataset and results in unbiased cosmological parameter estimation, is Bayesian Estimation Applied to Multiple Species (BEAMS). BEAMS is a Bayesian approach to the problem which includes the uncertainty in the types in the evaluation of the posterior. Here we outline the theory of BEAMS and demonstrate its effectiveness using both simulated datasets and SDSS-II data. We also show that it is possible to use BEAMS if the data are correlated, by introducing a numerical marginalisation over the types of the objects. This is largely a pedagogical introduction to BEAMS with references to the main BEAMS papers., Comment: Replaced under married name Lochner (formally Knights). 3 pages, 2 figures. To appear in the Proceedings of 13th Marcel Grossmann Meeting (MG13), Stockholm, Sweden, 1-7 July 2012

We present the cosmological analysis of 752 photometrically-classified Type Ia Supernovae (SNe Ia) obtained from the full Sloan Digital Sky Survey II (SDSS-II) Supernova (SN) Survey, supplemented with host-galaxy spectroscopy from the SDSS-III Baryon Oscillation Spectroscopic Survey (BOSS). Our photometric-classification method is based on the SN typing technique of Sako et al. (2011), aided by host galaxy redshifts (0.05

We introduce Bayesian Estimation Applied to Multiple Species (BEAMS), an algorithm designed to deal with parameter estimation when using contaminated data. We present the algorithm and demonstrate how it works with the help of a Gaussian simulation. We then apply it to supernova data from the Sloan Digital Sky Survey (SDSS), showing how the resulting confidence contours of the cosmological parameters shrink significantly., Comment: 23 pages, 9 figures. Chapter 4 in "Astrostatistical Challenges for the New Astronomy" (Joseph M. Hilbe, ed., Springer, New York, forthcoming in 2012), the inaugural volume for the Springer Series in Astrostatistics

New supernova surveys such as the Dark Energy Survey, Pan-STARRS and the LSST will produce an unprecedented number of photometric supernova candidates, most with no spectroscopic data. Avoiding biases in cosmological parameters due to the resulting inevitable contamination from non-Ia supernovae can be achieved with the BEAMS formalism, allowing for fully photometric supernova cosmology studies. Here we extend BEAMS to deal with the case in which the supernovae are correlated by systematic uncertainties. The analytical form of the full BEAMS posterior requires evaluating 2^N terms, where N is the number of supernova candidates. This `exponential catastrophe' is computationally unfeasible even for N of order 100. We circumvent the exponential catastrophe by marginalising numerically instead of analytically over the possible supernova types: we augment the cosmological parameters with nuisance parameters describing the covariance matrix and the types of all the supernovae, \tau_i, that we include in our MCMC analysis. We show that this method deals well even with large, unknown systematic uncertainties without a major increase in computational time, whereas ignoring the correlations can lead to significant biases and incorrect credible contours. We then compare the numerical marginalisation technique with a perturbative expansion of the posterior based on the insight that future surveys will have exquisite light curves and hence the probability that a given candidate is a Type Ia will be close to unity or zero, for most objects. Although this perturbative approach changes computation of the posterior from a 2^N problem into an N^2 or N^3 one, we show that it leads to biases in general through a small number of misclassifications, implying that numerical marginalisation is superior., Comment: Resubmitted under married name Lochner (formally Knights). Version 3: major changes, including a large scale analysis with thousands of MCMC chains. Matches version published in JCAP. 23 pages, 8 figures

Supernova cosmology without spectroscopic confirmation is an exciting new frontier which we address here with the Bayesian Estimation Applied to Multiple Species (BEAMS) algorithm and the full three years of data from the Sloan Digital Sky Survey II Supernova Survey (SDSS-II SN). BEAMS is a Bayesian framework for using data from multiple species in statistical inference when one has the probability that each data point belongs to a given species, corresponding in this context to different types of supernovae with their probabilities derived from their multi-band lightcurves. We run the BEAMS algorithm on both Gaussian and more realistic SNANA simulations with of order 10^4 supernovae, testing the algorithm against various pitfalls one might expect in the new and somewhat uncharted territory of photometric supernova cosmology. We compare the performance of BEAMS to that of both mock spectroscopic surveys and photometric samples which have been cut using typical selection criteria. The latter typically are either biased due to contamination or have significantly larger contours in the cosmological parameters due to small data-sets. We then apply BEAMS to the 792 SDSS-II photometric supernovae with host spectroscopic redshifts. In this case, BEAMS reduces the area of the (\Omega_m,\Omega_\Lambda) contours by a factor of three relative to the case where only spectroscopically confirmed data are used (297 supernovae). In the case of flatness, the constraints obtained on the matter density applying BEAMS to the photometric SDSS-II data are \Omega_m(BEAMS)=0.194\pm0.07. This illustrates the potential power of BEAMS for future large photometric supernova surveys such as LSST., Comment: 25 pages, 15 figures, submitted to ApJ

The original formulation of BEAMS - Bayesian Estimation Applied to Multiple Species - showed how to use a dataset contaminated by points of multiple underlying types to perform unbiased parameter estimation. An example is cosmological parameter estimation from a photometric supernova sample contaminated by unknown Type Ibc and II supernovae. Where other methods require data cuts to increase purity, BEAMS uses all of the data points in conjunction with their probabilities of being each type. Here we extend the BEAMS formalism to allow for correlations between the data and the type probabilities of the objects as can occur in realistic cases. We show with simple simulations that this extension can be crucial, providing a 50% reduction in parameter estimation variance when such correlations do exist. We then go on to perform tests to quantify the importance of the type probabilities, one of which illustrates the effect of biasing the probabilities in various ways. Finally, a general presentation of the selection bias problem is given, and discussed in the context of future photometric supernova surveys and BEAMS, which lead to specific recommendations for future supernova surveys., Comment: 10 figures, 14 pages

Using data from the Sloan Digital Sky Supernova Survey-II, we measure the rate of Type Ia Supernovae (SNe Ia) as a function of galaxy properties at intermediate redshift. A sample of 342 SNe Ia with 0.05

We improve estimates of stellar mass and mass-weighted average age of Type Ia supernova (SN Ia) host galaxies by combining UV and near-IR photometry with optical photometry in our analysis. Using 206 SNe Ia drawn from the full three-year SDSS-II Supernova Survey (median redshift of z {\approx} 0.2) and multi-wavelength host-galaxy photometry from SDSS, GALEX, and UKIDSS, we present evidence of a correlation (1.9{\sigma} confidence level) between the residuals of SNe Ia about the best-fit Hubble relation and the mass-weighted average age of their host galaxies. The trend is such that older galaxies host SNe Ia that are brighter than average after standard light-curve corrections are made. We also confirm, at the 3.0{\sigma} level, the trend seen by previous studies that more massive galaxies often host brighter SNe Ia after light-curve correction., Comment: The Astrophysical Journal (in press)

We analyze the three-year SDSS-II Superernova (SN) Survey data and identify a sample of 1070 photometric SN Ia candidates based on their multi-band light curve data. This sample consists of SN candidates with no spectroscopic confirmation, with a subset of 210 candidates having spectroscopic redshifts of their host galaxies measured, while the remaining 860 candidates are purely photometric in their identification. We describe a method for estimating the efficiency and purity of photometric SN Ia classification when spectroscopic confirmation of only a limited sample is available, and demonstrate that SN Ia candidates from SDSS-II can be identified photometrically with ~91% efficiency and with a contamination of ~6%. Although this is the largest uniform sample of SN candidates to date for studying photometric identification, we find that a larger spectroscopic sample of contaminating sources is required to obtain a better characterization of the background events. A Hubble diagram using SN candidates with no spectroscopic confirmation, but with host galaxy spectroscopic redshifts, yields a distance modulus dispersion that is only ~20 - 40% larger than that of the spectroscopically-confirmed SN Ia sample alone with no significant bias. A Hubble diagram with purely photometric classification and redshift-distance measurements, however, exhibit biases that require further investigation for precision cosmology., Comment: The Astrophysical Journal (in press)

We present an analysis of peculiar velocities and their effect on supernova cosmology. In particular, we study (a) the corrections due to our own motion, (b) the effects of correlations in peculiar velocities induced by large-scale structure, and (c) uncertainties arising from a possible local under- or over-density. For all of these effects we present a case study of their impact on the cosmology derived by the Sloan Digital Sky Survey-II Supernova Survey (SDSS-II SN Survey). Correcting supernova redshifts for the CMB dipole slightly over-corrects nearby supernovae that share some of our local motion. We show that while neglecting the CMB dipole would cause a shift in the derived equation of state of Delta w ~ 0.04 (at fixed matter density) the additional local-motion correction is currently negligible (Delta w<0.01). We use a covariance-matrix approach to statistically account for correlated peculiar velocities. This down-weights nearby supernovae and effectively acts as a graduated version of the usual sharp low-redshift cut. Neglecting coherent velocities in the current sample causes a systematic shift of ~2% in the preferred value of w and will therefore have to be considered carefully when future surveys aim for percent-level accuracy. Finally, we perform n-body simulations to estimate the likely magnitude of any local density fluctuation (monopole) and estimate the impact as a function of the low-redshift cutoff. We see that for this aspect the low-z cutoff of z=0.02 is well-justified theoretically, but that living in a putative local density fluctuation leaves an indelible imprint on the magnitude-redshift relation., Comment: 16 pages, 12 figures, accepted by ApJ

We present Keck high-quality rest-frame ultraviolet (UV) through optical spectra of 21 Type Ia supernovae (SNe Ia) in the redshift range 0.11 < z < 0.37 and a mean redshift of 0.22 that were discovered during the Sloan Digital Sky Survey-II (SDSS-II) SN Survey. Using the broad-band photometry of the SDSS survey, we are able to reconstruct the SN host-galaxy spectral energy distributions (SEDs), allowing for a correction for the host-galaxy contamination in the SN Ia spectra. Comparison of composite spectra constructed from a subsample of 17 high-quality spectra to those created from a low-redshift sample with otherwise similar properties shows that the Keck/SDSS SNe Ia have, on average, extremely similar rest-frame optical spectra but show a UV flux excess. This observation is confirmed by comparing synthesized broad-band colors of the individual spectra, showing a difference in mean colors at the 2.4 - 4.4 sigma level for various UV colors. We further see a slight difference in the UV spectral shape between SNe with low-mass and high-mass host galaxies. Additionally, we detect a relationship between the flux ratio at 2770 and 2900 A and peak luminosity that differs from that observed at low redshift. We find that changing the UV SED of an SN Ia within the observed dispersion can change the inferred distance moduli by ~0.1 mag. This effect only occurs when the data probe the rest-frame UV. We suggest that this discrepancy could be due to differences in the host-galaxy population of the two SN samples or to small-sample statistics., Comment: 28 pages, 21 figures, accepted by AJ, spectra are available at http://www.cfa.harvard.edu/~rfoley/data/