arXiv subjects

Christian Wolf

Publications and source records attributed to Christian Wolf.

At least 109 records · Page 6Linked to original sources

arXiv2020

Multiple phase transitions on compact symbolic systems

Let $ϕ:X\to \mathbb R$ be a continuous potential associated with a symbolic dynamical system $T:X\to X$ over a finite alphabet. Introducing a parameter $β>0$ (interpreted as the inverse temperature) we study the regularity of the pressure function $β\mapsto P_{\rm top}(βϕ)$ on an interval $[α,\infty)$ with $α>0$. We say that $ϕ$ has a phase transition at $β_0$ if the pressure function $P_{\rm top}(βϕ)$ is not differentiable at $β_0$. This is equivalent to the condition that the potential $β_0ϕ$ has two (ergodic) equilibrium states with distinct entropies. For any $α>0$ and any increasing sequence of real numbers $(β_n)$ contained in $[α,\infty)$, we construct a potential $ϕ$ whose phase transitions in $[α,\infty)$ occur precisely at the $β_n$'s. In particular, we obtain a potential which has a countably infinite set of phase transitions.

math.DS↗

arXiv2020

Testing gravity using galaxy-galaxy lensing and clustering amplitudes in KiDS-1000, BOSS and 2dFLenS

The physics of gravity on cosmological scales affects both the rate of assembly of large-scale structure, and the gravitational lensing of background light through this cosmic web. By comparing the amplitude of these different observational signatures, we can construct tests that can distinguish general relativity from its potential modifications. We used the latest weak gravitational lensing dataset from the Kilo-Degree Survey, KiDS-1000, in conjunction with overlapping galaxy spectroscopic redshift surveys BOSS and 2dFLenS, to perform the most precise existing amplitude-ratio test. We measured the associated E_G statistic with 15-20% errors, in five dz = 0.1 tomographic redshift bins in the range 0.2 < z < 0.7, on projected scales up to 100 Mpc/h. The scale-independence and redshift-dependence of these measurements are consistent with the theoretical expectation of general relativity in a Universe with matter density Omega_m = 0.27 +/- 0.04. We demonstrate that our results are robust against different analysis choices, including schemes for correcting the effects of source photometric redshift errors, and compare the performance of angular and projected galaxy-galaxy lensing statistics.

astro-ph.CO↗

arXiv2020

SkyMapper Southern Survey: Second Data Release (DR2)

We present the second data release (DR2) of the SkyMapper Southern Survey, a hemispheric survey carried out with the SkyMapper Telescope at Siding Spring Observatory in Australia, using six optical filters: $u,v,g,r,i,z$. DR2 is the first release to go beyond the $\sim$18mag (10$σ$) limit of the Shallow Survey released in DR1, and includes portions of the sky at full survey depth that reach >21mag in $g$ and $r$ filters. The DR2 photometry has a precision as measured by internal reproducibility of 1% in $u$ and $v$, and 0.7% in $griz$. More than 21 000 deg$^2$ have data in some filters (at either Shallow or Main Survey depth) and over 7 000 deg$^2$ have deep Main Survey coverage in all six filters. Finally, about 18 000 deg$^2$ have Main Survey data in $i$ and $z$ filters, albeit not yet at full depth. The release contains over 120 000 images, as well as catalogues with over 500 million unique astrophysical objects and nearly 5 billion individual detections. It also contains cross-matches with a range of external catalogues such as Gaia DR2, Pan-STARRS1 DR1, GALEX GUVcat, 2MASS, and AllWISE, as well as spectroscopic surveys such as 2MRS, GALAH, 6dFGS, and 2dFLenS.

astro-ph.IM↗

arXiv2020

Learning to plan with uncertain topological maps

We train an agent to navigate in 3D environments using a hierarchical strategy including a high-level graph based planner and a local policy. Our main contribution is a data driven learning based approach for planning under uncertainty in topological maps, requiring an estimate of shortest paths in valued graphs with a probabilistic structure. Whereas classical symbolic algorithms achieve optimal results on noise-less topologies, or optimal results in a probabilistic sense on graphs with probabilistic structure, we aim to show that machine learning can overcome missing information in the graph by taking into account rich high-dimensional node features, for instance visual information available at each location of the map. Compared to purely learned neural white box algorithms, we structure our neural model with an inductive bias for dynamic programming based shortest path algorithms, and we show that a particular parameterization of our neural model corresponds to the Bellman-Ford algorithm. By performing an empirical analysis of our method in simulated photo-realistic 3D environments, we demonstrate that the inclusion of visual features in the learned neural planner outperforms classical symbolic solutions for graph based planning.

cs.LG↗

arXiv2020

A Thirty-Four Billion Solar Mass Black Hole in SMSS J2157-3602, the Most Luminous Known Quasar

From near-infrared spectroscopic measurements of the MgII emission line doublet, we estimate the black hole (BH) mass of the quasar, SMSS J215728.21-360215.1, as being (3.4 +/- 0.6) x 10^10 M_sun and refine the redshift of the quasar to be z=4.692. SMSS J2157 is the most luminous known quasar, with a 3000A luminosity of (4.7 +/- 0.5) x 10^47 erg/s and an estimated bolometric luminosity of 1.6 x 10^48 erg/s, yet its Eddington ratio is only ~0.4. Thus, the high luminosity of this quasar is a consequence of its extremely large BH -- one of the most massive BHs at z > 4.

astro-ph.GA↗

arXiv2020

DRLViz: Understanding Decisions and Memory in Deep Reinforcement Learning

We present DRLViz, a visual analytics interface to interpret the internal memory of an agent (e.g. a robot) trained using deep reinforcement learning. This memory is composed of large temporal vectors updated when the agent moves in an environment and is not trivial to understand due to the number of dimensions, dependencies to past vectors, spatial/temporal correlations, and co-correlation between dimensions. It is often referred to as a black box as only inputs (images) and outputs (actions) are intelligible for humans. Using DRLViz, experts are assisted to interpret decisions using memory reduction interactions, and to investigate the role of parts of the memory when errors have been made (e.g. wrong direction). We report on DRLViz applied in the context of video games simulators (ViZDoom) for a navigation scenario with item gathering tasks. We also report on experts evaluation using DRLViz, and applicability of DRLViz to other scenarios and navigation problems beyond simulation games, as well as its contribution to black box models interpretability and explainability in the field of visual analytics.

cs.LG↗

arXiv2020

Evolutionary entropy and the Second Law of Thermodynamics

The dynamics of molecular collisions in a macroscopic body are encoded by the parameter Thermodynamic entropy - a statistical measure of the number of molecular configurations that correspond to a given macrostate. Directionality in the flow of energy in macroscopic bodies is described by the Second Law of Thermodynamics: In isolated systems, that is systems closed to the input of energy and matter, thermodynamic entropy increases. The dynamics of the lower level interactions in populations of replicating organisms is encoded by the parameter Evolutionary entropy, a statistical measure which describes the number and diversity of metabolic cycles in a population of replicating organisms. Directionality in the transformation of energy in populations of organisms is described by the Fundamental Theorem of Evolution: In systems open to the input of energy and matter, Evolutionary entropy increases, when the energy source is scarce and diverse, and decreases when the energy source is abundant and singular. This article shows that when rho to 0, and N to infinity, where rho is the production rate of the external energy source, and N denote the number of replicating units, evolutionary entropy, an organized state of energy; and thermodynamic entropy, a randomized state of energy, coincide. Accordingly, the Fundamental Theorem of Evolution, is a generalization of the Second Law of Thermodynamics.

q-bio.PE↗

arXiv2020

CoPhy: Counterfactual Learning of Physical Dynamics

Understanding causes and effects in mechanical systems is an essential component of reasoning in the physical world. This work poses a new problem of counterfactual learning of object mechanics from visual input. We develop the CoPhy benchmark to assess the capacity of the state-of-the-art models for causal physical reasoning in a synthetic 3D environment and propose a model for learning the physical dynamics in a counterfactual setting. Having observed a mechanical experiment that involves, for example, a falling tower of blocks, a set of bouncing balls or colliding objects, we learn to predict how its outcome is affected by an arbitrary intervention on its initial conditions, such as displacing one of the objects in the scene. The alternative future is predicted given the altered past and a latent representation of the confounders learned by the model in an end-to-end fashion with no supervision. We compare against feedforward video prediction baselines and show how observing alternative experiences allows the network to capture latent physical properties of the environment, which results in significantly more accurate predictions at the level of super human performance.

cs.CV↗

arXiv2020

Suppressed or enhanced central star formation rates in late-type barred galaxies

Bars in disc-dominated galaxies are able to drive gas inflow inside the corotation radius, thus enhancing the central star formation rate (SFR). Previous work, however, has found that disc-dominated galaxies with centrally suppressed SFR frequently host a bar. Here we investigate possible causes for the suppression of central SFR in such cases. We compare physical properties of a sample of disc-dominated barred galaxies with high central SFR (HC galaxies) with those of a sample of disc-dominated barred galaxies with low central SFR (LC galaxies). We find that the two samples have on average similar HI content and bars of similar strength. But we also find that the HCs have bluer colors than LCs, and that outside the bar region they host stronger spiral arms than the LCs where closed rings are more often seen. We discuss and evaluate the possible causes for the suppression of the central SFR in the LC galaxies as opposed to its enhancement in the HC galaxies.

astro-ph.GA↗

arXiv2020

EgoMap: Projective mapping and structured egocentric memory for Deep RL

Tasks involving localization, memorization and planning in partially observable 3D environments are an ongoing challenge in Deep Reinforcement Learning. We present EgoMap, a spatially structured neural memory architecture. EgoMap augments a deep reinforcement learning agent's performance in 3D environments on challenging tasks with multi-step objectives. The EgoMap architecture incorporates several inductive biases including a differentiable inverse projection of CNN feature vectors onto a top-down spatially structured map. The map is updated with ego-motion measurements through a differentiable affine transform. We show this architecture outperforms both standard recurrent agents and state of the art agents with structured memory. We demonstrate that incorporating these inductive biases into an agent's architecture allows for stable training with reward alone, circumventing the expense of acquiring and labelling expert trajectories. A detailed ablation study demonstrates the impact of key aspects of the architecture and through extensive qualitative analysis, we show how the agent exploits its structured internal memory to achieve higher performance.

cs.LG↗

arXiv2020

Probing the extragalactic fast transient sky at minute timescales with DECam

Searches for optical transients are usually performed with a cadence of days to weeks, optimised for supernova discovery. The optical fast transient sky is still largely unexplored, with only a few surveys to date having placed meaningful constraints on the detection of extragalactic transients evolving at sub-hour timescales. Here, we present the results of deep searches for dim, minute-timescale extragalactic fast transients using the Dark Energy Camera, a core facility of our all-wavelength and all-messenger Deeper, Wider, Faster programme. We used continuous 20s exposures to systematically probe timescales down to 1.17 minutes at magnitude limits $g > 23$ (AB), detecting hundreds of transient and variable sources. Nine candidates passed our strict criteria on duration and non-stellarity, all of which could be classified as flare stars based on deep multi-band imaging. Searches for fast radio burst and gamma-ray counterparts during simultaneous multi-facility observations yielded no counterparts to the optical transients. Also, no long-term variability was detected with pre-imaging and follow-up observations using the SkyMapper optical telescope. We place upper limits for minute-timescale fast optical transient rates for a range of depths and timescales. Finally, we demonstrate that optical $g$-band light curve behaviour alone cannot discriminate between confirmed extragalactic fast transients such as prompt GRB flashes and Galactic stellar flares.

astro-ph.HE↗

arXiv2019

Weak Supervision helps Emergence of Word-Object Alignment and improves Vision-Language Tasks

The large adoption of the self-attention (i.e. transformer model) and BERT-like training principles has recently resulted in a number of high performing models on a large panoply of vision-and-language problems (such as Visual Question Answering (VQA), image retrieval, etc.). In this paper we claim that these State-Of-The-Art (SOTA) approaches perform reasonably well in structuring information inside a single modality but, despite their impressive performances , they tend to struggle to identify fine-grained inter-modality relationships. Indeed, such relations are frequently assumed to be implicitly learned during training from application-specific losses, mostly cross-entropy for classification. While most recent works provide inductive bias for inter-modality relationships via cross attention modules, in this work, we demonstrate (1) that the latter assumption does not hold, i.e. modality alignment does not necessarily emerge automatically, and (2) that adding weak supervision for alignment between visual objects and words improves the quality of the learned models on tasks requiring reasoning. In particular , we integrate an object-word alignment loss into SOTA vision-language reasoning models and evaluate it on two tasks VQA and Language-driven Comparison of Images. We show that the proposed fine-grained inter-modality supervision significantly improves performance on both tasks. In particular, this new learning signal allows obtaining SOTA-level performances on GQA dataset (VQA task) with pre-trained models without finetuning on the task, and a new SOTA on NLVR2 dataset (Language-driven Comparison of Images). Finally, we also illustrate the impact of the contribution on the models reasoning by visualizing attention distributions.

cs.CV↗

arXiv2019

Photometric Flaring Fraction of M dwarf Stars from the SkyMapper Southern Survey

We present our search for flares from M dwarf stars in the SkyMapper Southern Survey DR1, which covers nearly the full Southern hemisphere with six-filter sequences that are repeatedly observed in the passbands $uvgriz$. This allows us to identify bona-fide flares in single-epoch observations on timescales of less than four minutes. Using a correlation-based outlier search algorithm we find 254 flare events in the amplitude range of $Δu \sim 0.1$ to 5~mag. In agreement with previous work, we observe the flaring fraction of M dwarfs to increase from $\sim 30$ to $\sim $1,000 per million stars for spectral types M0 to M5. We also confirm the decrease in flare fraction with larger vertical distance from the Galactic plane that is expected from declining stellar activity with age. Based on precise distances from Gaia DR2, we find a steep decline in the flare fraction from the plane to 150~pc vertical distance and a significant flattening towards larger distances. We then reassess the strong type dependence in the flaring fraction with a volume-limited sample within a distance of 50~pc from the Sun: in this sample the trend disappears and we find instead a constant fraction of $\sim$1,650 per million stars for spectral types M1 to M5. Finally, large-amplitude flares with $Δi > 1$ mag are very rare with a fraction of $\sim 0.5$ per million M dwarfs. Hence, we expect that M-dwarf flares will not confuse SkyMapper's search for kilonovae from gravitational-wave events.

astro-ph.SR↗

arXiv2019

A horseshoe with a discontinuous entropy spectrum

We study the regularity of the entropy spectrum of the Lyapunov exponents for hyperbolic maps on surfaces. It is well-known that the entropy spectrum is a concave upper semi-continuous function which is analytic on the interior of the set Lyapunov exponents. In this paper we construct a family of horseshoes with a discontinuous entropy spectrum at the boundary of the set of Lyapunov exponents.

math.DS↗

arXiv2019

Ultra-luminous quasars at redshift $z>4.5$ from SkyMapper

The most luminous quasars at high redshift harbour the fastest-growing and most massive black holes in the early Universe. They are exceedingly rare and hard to find. Here, we present our search for the most luminous quasars in the redshift range from $z=4.5$ to $5$ using data from SkyMapper, Gaia and WISE. We use colours to select likely high-redshift quasars and reduce the stellar contamination of the candidate set with parallax and proper motion data. In $\sim$12,500~deg$^2$ of Southern sky, we find 92 candidates brighter than $R_p=18.2$. Spectroscopic follow-up has revealed 21 quasars at $z\ge 4$ (16 of which are within $z=[4.5,5]$), as well as several red quasars, BAL quasars and objects with unusual spectra, which we tentatively label OFeLoBALQSOs at redshifts of $z\approx 1$ to $2$. This work lifts the number of known bright $z\ge 4.5$ quasars in the Southern hemisphere from 10 to 26 and brings the total number of quasars known at $R_p<18.2$ and $z\ge 4.5$ to 42.

astro-ph.GA↗

arXiv2019

Learning 3D Navigation Protocols on Touch Interfaces with Cooperative Multi-Agent Reinforcement Learning

Using touch devices to navigate in virtual 3D environments such as computer assisted design (CAD) models or geographical information systems (GIS) is inherently difficult for humans, as the 3D operations have to be performed by the user on a 2D touch surface. This ill-posed problem is classically solved with a fixed and handcrafted interaction protocol, which must be learned by the user. We propose to automatically learn a new interaction protocol allowing to map a 2D user input to 3D actions in virtual environments using reinforcement learning (RL). A fundamental problem of RL methods is the vast amount of interactions often required, which are difficult to come by when humans are involved. To overcome this limitation, we make use of two collaborative agents. The first agent models the human by learning to perform the 2D finger trajectories. The second agent acts as the interaction protocol, interpreting and translating to 3D operations the 2D finger trajectories from the first agent. We restrict the learned 2D trajectories to be similar to a training set of collected human gestures by first performing state representation learning, prior to reinforcement learning. This state representation learning is addressed by projecting the gestures into a latent space learned by a variational auto encoder (VAE).

cs.LG↗

arXiv2019

SkyMapper SEDs of nearby galaxies: quenching and bursting probed by a change index for star formation

The wish list of astronomers includes a tool that reveals quenching of star formation in galaxies directly as it proceeds. Here, we present a proof-of-concept for a new quenching-and-bursting diagnostic, a "change index" for star formation, that requires only photometric data, provided they include filters such as the violet $uv$ bands used by SkyMapper. The index responds mostly to changes in star-formation rate on a timescale of 20 to 500 Myr and is nearly insensitive to dust extinction. It works effectively to distances of 100 to 150 Mpc. We explore its application to eight example galaxies in SkyMapper DR2, including known E+A and Seyfert-1 galaxies. Owing to the degeneracies inherent in broad-band photometry, the change index can only be a qualitative indicator of changes in star-formation rate. But once the SkyMapper Southern Survey is complete, the change index will be available for every spatial resolution element of every galaxy in the Southern sky within its working distance range.

astro-ph.GA↗

arXiv2019

Deep Reinforcement Learning on a Budget: 3D Control and Reasoning Without a Supercomputer

An important goal of research in Deep Reinforcement Learning in mobile robotics is to train agents capable of solving complex tasks, which require a high level of scene understanding and reasoning from an egocentric perspective. When trained from simulations, optimal environments should satisfy a currently unobtainable combination of high-fidelity photographic observations, massive amounts of different environment configurations and fast simulation speeds. In this paper we argue that research on training agents capable of complex reasoning can be simplified by decoupling from the requirement of high fidelity photographic observations. We present a suite of tasks requiring complex reasoning and exploration in continuous, partially observable 3D environments. The objective is to provide challenging scenarios and a robust baseline agent architecture that can be trained on mid-range consumer hardware in under 24h. Our scenarios combine two key advantages: (i) they are based on a simple but highly efficient 3D environment (ViZDoom) which allows high speed simulation (12000fps); (ii) the scenarios provide the user with a range of difficulty settings, in order to identify the limitations of current state of the art algorithms and network architectures. We aim to increase accessibility to the field of Deep-RL by providing baselines for challenging scenarios where new ideas can be iterated on quickly. We argue that the community should be able to address challenging problems in reasoning of mobile agents without the need for a large compute infrastructure.

cs.LG↗