arXiv subjects

Bruno Ribeiro

Publications and source records attributed to Bruno Ribeiro.

At least 73 records · Page 4Linked to original sources

arXiv2020

The evolution of rest-frame UV properties, Lya EWs and the SFR-Stellar mass relation at z~2-6 for SC4K LAEs

We explore deep rest-frame UV to FIR data in the COSMOS field to measure the individual spectral energy distributions (SED) of the ~4000 SC4K (Sobral et al. 2018) Lyman-alpha (Lya) emitters (LAEs) at z~2-6. We find typical stellar masses of 10$^{9.3\pm0.6}$ M$_{\odot}$ and star formation rates (SFR) of SFR$_{SED}=4.4^{+10.5}_{-2.4}$ M$_{\odot}$/yr and SFR$_{Lya}=5.9^{+6.3}_{-2.6}$ M$_{\odot}$/yr, combined with very blue UV slopes of beta=-2.1$^{+0.5}_{-0.4}$, but with significant variations within the population. M$_{UV}$ and beta are correlated in a similar way to UV-selected sources, but LAEs are consistently bluer. This suggests that LAEs are the youngest and/or most dust-poor subset of the UV-selected population. We also study the Lya rest-frame equivalent width (EW$_0$) and find 45 "extreme" LAEs with EW$_0>240$ A (3 $σ$), implying a low number density of $(7\pm1)\times10^{-7}$ Mpc$^{-3}$. Overall, we measure little to no evolution of the Lya EW$_0$ and scale length parameter ($w_0$) which are consistently high (EW$_0=140^{+280}_{-70}$ A, $w_0=129^{+11}_{-11}$ A) from z~6 to z~2 and below. However, $w_0$ is anti-correlated with M$_{UV}$ and stellar mass. Our results imply that sources selected as LAEs have a high Lya escape fraction (f$_{esc, Lya}$) irrespective of cosmic time, but f$_{esc, Lya}$ is still higher for UV-fainter and lower mass LAEs. The least massive LAEs ($<10^{9.5}$ M$_{\odot}$) are typically located above the star formation "Main Sequence" (MS), but the offset from the MS decreases towards z~6 and towards $10^{10}$ M$_{\odot}$. Our results imply a lack of evolution in the properties of LAEs across time and reveals the increasing overlap in properties of LAEs and UV-continuum selected galaxies as typical star-forming galaxies at high redshift effectively become LAEs.

astro-ph.GA↗

arXiv2019

VIS3COS: III. environmental effects on the star formation histories of galaxies at z~0.8 seen in [OII], H$δ$, and Dn4000

[ABRIDGED] We present spectroscopic observations of 466 galaxies in and around a superstructure at $z\sim0.84$ targeted by the VIMOS Spectroscopic Survey of a Supercluster in the COSMOS field (VIS$^{3}$COS). We use [OII]$λ$3727, H$δ$, and $D_n4000$ to trace the recent, mid-, and long-term star formation histories and investigate how stellar mass and the local environment impacts those. By studying trends both in individual and composite galaxy spectra, we find that both stellar mass and environment play a role in the observed galactic properties. We find that the median [OII] equivalent width (|EW$_\mathrm{[OII]}|$) decreases from $27\pm2$ Å to $2.0_{-0.4}^{+0.5}$ Å and $D_n4000$ increases from $1.09\pm0.01$ to $1.56\pm0.03$ with increasing stellar mass (from $\sim10^{9.25}$ to $\sim10^{11.35}\ \mathrm{M_\odot}$). Concerning the dependence on the environment, we find that at fixed stellar mass |EW$_\mathrm{[OII]}|$ is tentatively lower in higher density environments. Regarding $D_n4000$, we find that the increase with stellar mass is sharper in denser environments, hinting that such environments may accelerate galaxy evolution. Moreover, we find larger $D_n4000$ values in denser environments at fixed stellar mass, suggesting that galaxies are on average older and/or more metal-rich in such dense environments. This set of tracers depicts a scenario where the most massive galaxies have, on average, the lowest sSFRs and the oldest stellar populations (age $\gtrsim1$ Gyr, showing a mass-downsizing effect). We also hypothesize that the observed increase in star formation (higher EW$_\mathrm{[OII]|}$, higher sSFR) at intermediate densities may lead to quenching since we find the quenched fraction to increase sharply from the filament to cluster-like regions at similar stellar masses.

astro-ph.GA↗

arXiv2019

Deep Lifetime Clustering

The goal of lifetime clustering is to develop an inductive model that maps subjects into $K$ clusters according to their underlying (unobserved) lifetime distribution. We introduce a neural-network based lifetime clustering model that can find cluster assignments by directly maximizing the divergence between the empirical lifetime distributions of the clusters. Accordingly, we define a novel clustering loss function over the lifetime distributions (of entire clusters) based on a tight upper bound of the two-sample Kuiper test p-value. The resultant model is robust to the modeling issues associated with the unobservability of termination signals, and does not assume proportional hazards. Our results in real and synthetic datasets show significantly better lifetime clusters (as evaluated by C-index, Brier Score, Logrank score and adjusted Rand index) as compared to competing approaches.

cs.LG↗

arXiv2019

VIS3COS: II. Nature and nurture in galaxy structure and morphology

We study the impact of local density and stellar mass on the morphology of ~500 quiescent (Q) and SFGs from the VIS3COS survey. We perform B/D decomposition of the SBPs and find ~41% of 10^10 Msun galaxies to be best fitted with 2 components. We complement our analysis with non-parametric measurements and qualitative visual classifications. We find that galaxy morphology depends on stellar mass and environment for our sample as a whole. We only find an impact of the environment on galaxy sizes for galaxies more massive than 10^11 Msun. We find higher n and B/T in high-density regions when compared to low-density counterparts at similar stellar masses. We also find that higher stellar mass galaxies have steeper light profiles compared to the lower ones. Using visual classifications, we find a morphology-density relation at z~0.84 for galaxies more massive than 10^10 Msun, with elliptical galaxies being dominant at high-density regions and disks more common in low-density regions. However, when splitting the sample into colour-colour selected SF and Q sub-populations, there are no statistically significant differences between low- and high-density regions. We find that Q galaxies are smaller, have higher n, and higher B/T when compared to SF counterparts. We confirm these trends with non-parametric quantities, finding Q galaxies to be smoother and having most of their light over smaller areas than SFGs. Overall, we find that the dependence of morphology on stellar mass is stronger than with local density and these relations are strongly correlated with the quenching fraction. The change in average morphology corresponds to a change in the relative fractions of blue disk-like galaxies and red elliptical galaxies with stellar mass and environment. We hypothesize that the processes responsible for the quenching of SF must also affect the galaxy morphology on similar timescales.

astro-ph.GA↗

arXiv2019

Are Graph Neural Networks Miscalibrated?

Graph Neural Networks (GNNs) have proven to be successful in many classification tasks, outperforming previous state-of-the-art methods in terms of accuracy. However, accuracy alone is not enough for high-stakes decision making. Decision makers want to know the likelihood that a specific GNN prediction is correct. For this purpose, obtaining calibrated models is essential. In this work, we perform an empirical evaluation of the calibration of state-of-the-art GNNs on multiple datasets. Our experiments show that GNNs can be calibrated in some datasets but also badly miscalibrated in others, and that state-of-the-art calibration methods are helpful but do not fix the problem.

cs.LG↗

arXiv2019

Relational Pooling for Graph Representations

This work generalizes graph neural networks (GNNs) beyond those based on the Weisfeiler-Lehman (WL) algorithm, graph Laplacians, and diffusions. Our approach, denoted Relational Pooling (RP), draws from the theory of finite partial exchangeability to provide a framework with maximal representation power for graphs. RP can work with existing graph representation models and, somewhat counterintuitively, can make them even more powerful than the original WL isomorphism test. Additionally, RP allows architectures like Recurrent Neural Networks and Convolutional Neural Networks to be used in a theoretically sound approach for graph classification. We demonstrate improved performance of RP-based graph representations over state-of-the-art methods on a number of tasks.

cs.LG↗

arXiv2019

Janossy Pooling: Learning Deep Permutation-Invariant Functions for Variable-Size Inputs

We consider a simple and overarching representation for permutation-invariant functions of sequences (or multiset functions). Our approach, which we call Janossy pooling, expresses a permutation-invariant function as the average of a permutation-sensitive function applied to all reorderings of the input sequence. This allows us to leverage the rich and mature literature on permutation-sensitive functions to construct novel and flexible permutation-invariant functions. If carried out naively, Janossy pooling can be computationally prohibitive. To allow computational tractability, we consider three kinds of approximations: canonical orderings of sequences, functions with $k$-order interactions, and stochastic optimization algorithms with random permutations. Our framework unifies a variety of existing work in the literature, and suggests possible modeling and algorithmic extensions. We explore a few in our experiments, which demonstrate improved performance over current state-of-the-art methods.

cs.LG↗

arXiv2018

Feedforward Neural Networks for Caching: Enough or Too Much?

We propose a caching policy that uses a feedforward neural network (FNN) to predict content popularity. Our scheme outperforms popular eviction policies like LRU or ARC, but also a new policy relying on the more complex recurrent neural networks. At the same time, replacing the FNN predictor with a naive linear estimator does not degrade caching performance significantly, questioning then the role of neural networks for these applications.

cs.NI↗

arXiv2018

Characterizing Directed and Undirected Networks via Multidimensional Walks with Jumps

Estimating distributions of node characteristics (labels) such as number of connections or citizenship of users in a social network via edge and node sampling is a vital part of the study of complex networks. Due to its low cost, sampling via a random walk (RW) has been proposed as an attractive solution to this task. Most RW methods assume either that the network is undirected or that walkers can traverse edges regardless of their direction. Some RW methods have been designed for directed networks where edges coming into a node are not directly observable. In this work, we propose Directed Unbiased Frontier Sampling (DUFS), a sampling method based on a large number of coordinated walkers, each starting from a node chosen uniformly at random. It is applicable to directed networks with invisible incoming edges because it constructs, in real-time, an undirected graph consistent with the walkers trajectories, and due to the use of random jumps which prevent walkers from being trapped. DUFS generalizes previous RW methods and is suited for undirected networks and to directed networks regardless of in-edges visibility. We also propose an improved estimator of node label distributions that combines information from the initial walker locations with subsequent RW observations. We evaluate DUFS, compare it to other RW methods, investigate the impact of its parameters on estimation accuracy and provide practical guidelines for choosing them. In estimating out-degree distributions, DUFS yields significantly better estimates of the head of the distribution than other methods, while matching or exceeding estimation accuracy of the tail. Last, we show that DUFS outperforms uniform node sampling when estimating distributions of node labels of the top 10% largest degree nodes, even when sampling a node uniformly has the same cost as RW steps.

cs.SI↗

arXiv2018

VIS3COS: I. survey overview and the role of environment and stellar mass on star formation

We present the VIMOS Spectroscopic Survey of a Supercluster in the COSMOS field (VIS3COS) at z~0.84. We use VIMOS high-resolution spectra (GG475 filter) to spectroscopically select 490 galaxies in and around the super-structure and an additional 481 galaxies in the line of sight. We present the redshift distribution, the catalog to be made public and the first results on the properties of individual galaxies and stacked spectra (3500Å<λ<4200Å rest-frame). We probe a wide range of densities and environments (from low-density field to clusters and rich groups). We find a decrease in the median star formation rate from low to high density environments in all bins of stellar mass and a sharp rise of the quenched fraction (from ~10% to ~40-60%) of intermediate stellar mass galaxies $(10<\log_{10}\left(M_\star/\mathrm{M_\odot}\right)<10.75)$ from filaments to clusters. The quenched fraction for massive galaxies shows little dependence on the environment being constant at ~30-40%. We find a break in the strength of the [OII] emission, with nearly constant line equivalent widths at lower densities (~-11Å) and then a drop to ~-2.5Å towards higher densities. The break in the [OII] line strength happens at similar densities ($\log_{10}(1+δ)\sim0.0-0.5$) as the observed rise in the quenched fraction. Our results may provide further clues regarding the different environmental processes affecting galaxies with different stellar masses and highlight the advantages of a single dataset in the COSMOS field probing a wide range of stellar masses and environments. We hypothesize that quenching mechanisms are enhanced in high-density regions.

astro-ph.GA↗

arXiv2018

On the UV compactness and morphologies of typical Lyman-a emitters from z~2 to z~6

We investigate the rest-frame UV morphologies of a large sample of Lyman-a emitters (LAEs) from z~2 to z~6, selected in a uniform way with 16 different narrow- and medium-bands over the full COSMOS field. We use 3045 LAEs with HST coverage in a stacking analysis and find that they have M_UV~-20, below M*_UV at these redshifts. We also focus our analysis on a subsample of 780 individual galaxies with i_AB<25 for which GALFIT converges for 429 of them. The individual median size (re~1 kpc), ellipticities (slightly elongated with (b/a)~0.45), Sérsic index (disk-like with n<2) and light concentration (comparable to that of disk or irregular galaxies, with C~2.7) of LAEs show mild evolution from z~2 to z~6. LAEs with the highest rest-frame equivalent widths (EW) are the smallest/most compact (re~0.8 kpc, compared to re~1.5 kpc for the lower EW LAEs). When stacking our samples in bins of fixed Lya luminosity and Lya EW we find evidence for redshift evolution in n and C, but not in galaxy sizes. The evolution seems to be stronger for LAEs with 25 5 to SFGs being a factor of ~2-4 larger than LAEs for z<2. This means that at the highest redshifts, where typical sizes approach those of LAEs, the fraction of galaxies showing Lya in emission (and with a high Lya escape fraction) should be much higher, consistent with observations.

astro-ph.GA↗

arXiv2018

Slicing COSMOS with SC4K: the evolution of typical Lya emitters and the Lya escape fraction from z~2 to z~6

We present and explore deep narrow- and medium-band data obtained with the Subaru and the Isaac Newton telescopes in the ~2 deg$^2$ COSMOS field. We use these data as an extremely wide, low-resolution (R~20-80) IFU survey to slice through the COSMOS field and obtain a large sample of ~4000 Lyman-$α$ (Lya) emitters from z~2 to z~6 in 16 redshift slices (SC4K). We present new Lya luminosity functions (LFs) covering a co-moving volume of ~10$^8$Mpc$^3$. SC4K extensively complements ultra-deep surveys, jointly covering over 4 dex in Lya luminosity and revealing a global (2.5 3.5, likely linked with the evolution of the AGN population. The Lya luminosity density rises by a factor ~2 from z~2 to z~3 but is then found to be roughly constant (~$1.1\times10^{40}$ erg s$^{-1}$ Mpc$^{-3}$) to z~6, despite the ~0.7 dex drop in UV luminosity density. The Lya/UV luminosity density ratio rises from $4\pm1$% to $30\pm6$% from z~2.2 to z~6. Our results imply a rise of a factor of ~2 in the global ionisation efficiency ($ξ_{\rm ion}$) and a factor ~$4\pm1$ in the Lya escape fraction from z~2 to z~6, hinting for evolution in both the typical burstiness/stellar populations and even more so in the typical ISM conditions allowing Ly$α$ photons to escape.

astro-ph.GA↗

arXiv2017

From Monte Carlo to Las Vegas: Improving Restricted Boltzmann Machine Training Through Stopping Sets

We propose a Las Vegas transformation of Markov Chain Monte Carlo (MCMC) estimators of Restricted Boltzmann Machines (RBMs). We denote our approach Markov Chain Las Vegas (MCLV). MCLV gives statistical guarantees in exchange for random running times. MCLV uses a stopping set built from the training data and has maximum number of Markov chain steps K (referred as MCLV-K). We present a MCLV-K gradient estimator (LVS-K) for RBMs and explore the correspondence and differences between LVS-K and Contrastive Divergence (CD-K), with LVS-K significantly outperforming CD-K training RBMs over the MNIST dataset, indicating MCLV to be a promising direction in learning generative models.

cs.LG↗

arXiv2017

SBG-Sketch: A Self-Balanced Sketch for Labeled-Graph Stream Summarization

Applications in various domains rely on processing graph streams, e.g., communication logs of a cloud-troubleshooting system, road-network traffic updates, and interactions on a social network. A labeled-graph stream refers to a sequence of streamed edges that form a labeled graph. Label-aware applications need to filter the graph stream before performing a graph operation. Due to the large volume and high velocity of these streams, it is often more practical to incrementally build a lossy-compressed version of the graph, and use this lossy version to approximately evaluate graph queries. Challenges arise when the queries are unknown in advance but are associated with filtering predicates based on edge labels. Surprisingly common, and especially challenging, are labeled-graph streams that have highly skewed label distributions that might also vary over time. This paper introduces Self-Balanced Graph Sketch (SBG-Sketch, for short), a graphical sketch for summarizing and querying labeled-graph streams that can cope with all these challenges. SBG-Sketch maintains synopsis for both the edge attributes (e.g., edge weight) as well as the topology of the streamed graph. SBG-Sketch allows efficient processing of graph-traversal queries, e.g., reachability queries. Experimental results over a variety of real graph streams show SBG-Sketch to reduce the estimation errors of state-of-the-art methods by up to 99%.

cs.DB↗

arXiv2017

Stochastic Gradient Descent for Relational Logistic Regression via Partial Network Crawls

Research in statistical relational learning has produced a number of methods for learning relational models from large-scale network data. While these methods have been successfully applied in various domains, they have been developed under the unrealistic assumption of full data access. In practice, however, the data are often collected by crawling the network, due to proprietary access, limited resources, and privacy concerns. Recently, we showed that the parameter estimates for relational Bayes classifiers computed from network samples collected by existing network crawlers can be quite inaccurate, and developed a crawl-aware estimation method for such models (Yang, Ribeiro, and Neville, 2017). In this work, we extend the methodology to learning relational logistic regression models via stochastic gradient descent from partial network crawls, and show that the proposed method yields accurate parameter estimates and confidence intervals.

stat.ML↗

arXiv2017

On the Duration and Intensity of Competitions in Nonlinear Pólya Urn Processes with Fitness

Cumulative advantage (CA) refers to the notion that accumulated resources foster the accumulation of further resources in competitions, a phenomenon that has been empirically observed in various contexts. The oldest and arguably simplest mathematical model that embodies this general principle is the Pólya urn process, which finds applications in a myriad of problems. The original model captures the dynamics of competitions between two equally fit agents under linear CA effects, which can be readily generalized to incorporate different fitnesses and nonlinear CA effects. We study two statistics of competitions under the generalized model, namely duration (i.e., time of the last tie) and intensity (i.e., number of ties). We give rigorous mathematical characterizations of the tail distributions of both duration and intensity under the various regimes for fitness and nonlinearity, which reveal very interesting behaviors. For example, fitness superiority induces much shorter competitions in the sublinear regime while much longer competitions in the superlinear regime. Our findings can shed light on the application of Pólya urn processes in more general contexts where fitness and nonlinearity may be present.

cs.PF↗

arXiv2017

Selective Harvesting over Networks

Active search (AS) on graphs focuses on collecting certain labeled nodes (targets) given global knowledge of the network topology and its edge weights under a query budget. However, in most networks, nodes, topology and edge weights are all initially unknown. We introduce selective harvesting, a variant of AS where the next node to be queried must be chosen among the neighbors of the current queried node set; the available training data for deciding which node to query is restricted to the subgraph induced by the queried set (and their node attributes) and their neighbors (without any node or edge attributes). Therefore, selective harvesting is a sequential decision problem, where we must decide which node to query at each step. A classifier trained in this scenario suffers from a tunnel vision effect: without recourse to independent sampling, the urge to query promising nodes forces classifiers to gather increasingly biased training data, which we show significantly hurts the performance of AS methods and standard classifiers. We find that it is possible to collect a much larger set of targets by using multiple classifiers, not by combining their predictions as an ensemble, but switching between classifiers used at each step, as a way to ease the tunnel vision effect. We discover that switching classifiers collects more targets by (a) diversifying the training data and (b) broadening the choices of nodes that can be queried next. This highlights an exploration, exploitation, and diversification trade-off in our problem that goes beyond the exploration and exploitation duality found in classic sequential decision problems. From these observations we propose D3TS, a method based on multi-armed bandits for non-stationary stochastic processes that enforces classifier diversity, matching or exceeding the performance of competing methods on seven real network datasets in our evaluation.

cs.SI↗

arXiv2017

Identifying User Survival Types via Clustering of Censored Social Network Data

The goal of cluster analysis in survival data is to identify clusters that are decidedly associated with the survival outcome. Previous research has explored this problem primarily in the medical domain with relatively small datasets, but the need for such a clustering methodology could arise in other domains with large datasets, such as social networks. Concretely, we wish to identify different survival classes in a social network by clustering the users based on their lifespan in the network. In this paper, we propose a decision tree based algorithm that uses a global normalization of $p$-values to identify clusters with significantly different survival distributions. We evaluate the clusters from our model with the help of a simple survival prediction task and show that our model outperforms other competing methods.

cs.SI↗