arXiv subjects

Nikolay Nikolov

Publications and source records attributed to Nikolay Nikolov.

At least 55 records · Page 3Linked to original sources

arXiv2021

JWST Noise Floor II: Systematic Error Sources in JWST NIRCam Time Series

JWST holds great promise in characterizing atmospheres of transiting exoplanets, potentially providing insights into Earth-sized planets within the habitable zones of M dwarf host stars if photon-limited performance can be achieved. Here, we discuss the systematic error sources that are expected to be present in grism time series observations with the NIRCam instrument. We find that pointing jitter and high gain antenna moves on top of the detectors' subpixel crosshatch patterns will produce relatively small variations (less than 6 parts per million, ppm). The time-dependent aperture losses due to thermal instabilities in the optics can also be kept to below 2 ppm. To achieve these low noise sources, it is important to employ a sufficiently large (more than 1.1 arcseconds) extraction aperture. Persistence due to charge trapping will have a minor (less than 3 ppm) effect on time series 20 minutes into an exposure and is expected to play a much smaller role than it does for the HST WFC3 detectors. We expect temperature fluctuations to be less than 3 ppm. In total, our estimated noise floor from known systematic error sources is only 9 ppm per visit. We do however urge caution as unknown systematic error sources could be present in flight and will only be measurable on astrophysical sources like quiescent stars. We find that reciprocity failure may introduce a perennial instrument offset at the 40 ppm level, so corrections may be needed when stitching together a multi-instrument multi-observatory spectrum over wide wavelength ranges.

astro-ph.IM↗

arXiv2020

Evidence of a Clear Atmosphere for WASP-62b: the Only Known Transiting Gas Giant in the JWST Continuous Viewing Zone

Exoplanets with cloud-free, haze-free atmospheres at the pressures probed by transmission spectroscopy represent a valuable opportunity for detailed atmospheric characterization and precise chemical abundance constraints. We present the first optical to infrared (0.3-5 microns) transmission spectrum of the hot Jupiter WASP-62b, measured with Hubble/STIS and Spitzer/IRAC. The spectrum is characterized by a 5.1-sigma detection of Na I absorption at 0.59 microns, in which the pressure-broadened wings of the Na D-lines are observed from space for the first time. A spectral feature at 0.4 microns is tentatively attributed to SiH at 2.1-sigma confidence. Our retrieval analyses are consistent with a cloud-free atmosphere without significant contamination from stellar heterogeneities. We simulate James Webb Space Telescope (JWST) observations, for a combination of instrument modes, to assess the atmospheric characterization potential of WASP-62b. We demonstrate that JWST can conclusively detect Na, H2O, FeH, and SiH within the scope of its Early Release Science (ERS) program. As the only transiting giant planet currently known in the JWST Continuous Viewing Zone, WASP-62b could prove a benchmark giant exoplanet for detailed atmospheric characterization in the James Webb era.

astro-ph.EP↗

arXiv2020

The Hubble PanCET Program: A Metal-rich Atmosphere for the Inflated Hot Jupiter HAT-P-41b

We present a comprehensive analysis of the 0.3--5\,$μ$m transit spectrum for the inflated hot Jupiter HAT-P-41b. The planet was observed in transit with Hubble STIS and WFC3 as part of the Hubble Panchromatic Comparative Exoplanet Treasury (PanCET) program, and we combine those data with warm \textit{Spitzer} transit observations. We extract transit depths from each of the data sets, presenting the STIS transit spectrum (0.29--0.93\,$μ$m) for the first time. We retrieve the transit spectrum both with a free-chemistry retrieval suite (AURA) and a complementary chemical equilibrium retrieval suite (PLATON) to constrain the atmospheric properties at the day-night terminator. Both methods provide an excellent fit to the observed spectrum. Both AURA and PLATON retrieve a metal-rich atmosphere for almost all model assumptions (most likely O/H ratio of $\log_{10}{Z/Z_{\odot}} = 1.46^{+0.53}_{-0.68}$ and $\log_{10}{Z/Z_{\odot}} = 2.33^{+0.23}_{-0.25}$, respectively); this is driven by a 4.9-$σ$ detection of H$_2$O as well as evidence of gas absorption in the optical ($>$2.7-$σ$ detection) due to Na, AlO and/or VO/TiO, though no individual species is strongly detected. Both retrievals determine the transit spectrum to be consistent with a clear atmosphere, with no evidence of haze or high-altitude clouds. Interior modeling constraints on the maximum atmospheric metallicity ($\log_{10}{Z/Z_{\odot}} < 1.7$) favor the AURA results. The inferred elemental oxygen abundance suggests that HAT-P-41b has one of the most metal-rich atmospheres of any hot Jupiters known to date. Overall, the inferred high metallicity and high inflation make HAT-P-41b an interesting test case for planet formation theories.

astro-ph.EP↗

arXiv2020

Into the UV: The Atmosphere of the Hot Jupiter HAT-P-41b Revealed

For solar-system objects, ultraviolet spectroscopy has been critical in identifying sources for stratospheric heating and measuring the abundances of a variety of hydrocarbon and sulfur-bearing species, produced via photochemical mechanisms, as well as oxygen and ozone. To date, less than 20 exoplanets have been probed in this critical wavelength range (0.2-0.4 um). Here we use data from Hubble's newly implemented WFC3 UVIS G280 grism to probe the atmosphere of the hot Jupiter HAT-P-41b in the ultraviolet through optical in combination with observations at infrared wavelengths. We analyze and interpret HAT-P-41b's 0.2-5.0 um transmission spectrum using a broad range of methodologies including multiple treatments of data systematics as well as comparisons with atmospheric forward, cloud microphysical, and multiple atmospheric retrieval models. Although some analysis and interpretation methods favor the presence of clouds or potentially a combination of Na, VO, AlO, and CrH to explain the ultraviolet through optical portions of HAT-P-41b's transmission spectrum, we find that the presence of a significant H- opacity provides the most robust explanation. We obtain a constraint for the abundance of H-, log(H-) = -8.65 +/- 0.62 in HAT-P-41b's atmosphere, which is several orders of magnitude larger than predictions from equilibrium chemistry for a 1700 - 1950 K hot Jupiter. We show that a combination of photochemical and collisional processes on hot hydrogen-dominated exoplanets can readily supply the necessary amount of H- and suggest that such processes are at work in HAT-P-41b and many other hot Jupiter atmospheres.

astro-ph.EP↗

arXiv2020

WASP-127b transmission spectrum]{Abundance measurements of H$_{2}$O and carbon-bearing species in the atmosphere of WASP-127b confirm its super-solar metallicity

The chemical abundances of exoplanet atmospheres may provide valuable information about the bulk compositions, formation pathways, and evolutionary histories of planets. Exoplanets with large, relatively cloud-free atmospheres, and which orbit bright stars provide the best opportunities for accurate abundance measurements. For this reason, we measured the transmission spectrum of the bright (V~10.2), large (1.37 R$_{J}$), sub-Saturn mass (0.19 M$_{J}$) exoplanet WASP-127b across the near-UV to near-infrared wavelength range (0.3 - 5 $μ$m), using the Hubble and Spitzer Space Telescopes. Our results show a feature-rich transmission spectrum, with absorption from Na, H$_{2}$O, and CO$_{2}$, and wavelength-dependent scattering from small-particle condensates. We ran two types of atmospheric retrieval models: one enforcing chemical equilibrium, and the other which fit the abundances freely. Our retrieved abundances at chemical equilibrium for Na, O and C are all super-solar, with abundances relative to solar values of 9$^{+15}_{-6}$, 16$^{+7}_{-5}$, and 26$^{+12}_{-9}$ respectively. Despite giving conflicting C/O ratios, both retrievals gave super-solar CO$_{2}$ volume mixing ratios, which adds to the likelihood that WASP-127b's bulk metallicity is super-solar, since CO$_{2}$ abundance is highly sensitive to atmospheric metallicity. We detect water at a significance of 13.7 $σ$. Our detection of Na is in agreement with previous ground-based detections, though we find a much lower abundance, and we also do not find evidence for Li or K despite increased sensitivity. In the future, spectroscopy with JWST will be able to constrain WASP-127b's C/O ratio, and may reveal the formation history of this metal-enriched, highly observable exoplanet.

astro-ph.EP↗

arXiv2020

A library of ATMO forward model transmission spectra for hot Jupiter exoplanets

We present a grid of forward model transmission spectra, adopting an isothermal temperature-pressure profile, alongside corresponding equilibrium chemical abundances for 117 observationally significant hot exoplanets (Equilibrium Temperatures of 547-2710 K). This model grid has been developed using a 1D radiative-convective-chemical equilibrium model termed ATMO, with up-to-date high temperature opacities. We present an interpretation of observations of ten exoplanets, including best fit parameters and $χ^{2}$ maps. In agreement with previous works, we find a continuum from clear to hazy/cloudy atmospheres for this sample of hot Jupiters. The data for all the 10 planets are consistent with sub-solar to solar C/O ratio, 0.005 to 10 times solar metallicity and water rather than a methane dominated infrared spectra. We then explore the range of simulated atmospheric spectra for different exoplanets, based on characteristics such as temperature, metallicity, C/O-ratio, haziness and cloudiness. We find a transition value for the metallicity between 10 and 50 times solar, which leads to substantial changes in the transmission spectra. We also find a transition value of C/O ratio, from water to carbon species dominated infrared spectra, as found by previous works, revealing a temperature dependence of this transition point ranging from $\sim$0.56 to $\sim$1-1.3 for equilibrium temperatures from $\sim$900 to $\sim$2600 K. We highlight the potential of the spectral features of HCN and C$_2$H$_2$ to constrain the metallicities and C/O ratios of planets, using JWST observations. Finally, our entire grid ($\sim$460,000 simulations) is publicly available and can be used directly with the JWST simulator PandExo for planning observations.

astro-ph.EP↗

arXiv2020

The HST PanCET Program: An Optical to Infrared Transmission Spectrum of HAT-P-32Ab

We present a 0.3-5 micron transmission spectrum of the hot Jupiter HAT-P-32Ab observed with the Space Telescope Imaging Spectrograph (STIS) and Wide Field Camera 3 (WFC3) instruments mounted on the Hubble Space Telescope, combined with Spitzer Infrared Array Camera (IRAC) photometry. The spectrum is composed of 51 spectrophotometric bins with widths ranging between 150 and 400 Å, measured to a median precision of 215 ppm. Comparisons of the observed transmission spectrum to a grid of 1D radiative-convective equilibrium models indicate the presence of clouds/hazes, consistent with previous transit observations and secondary eclipse measurements. To provide more robust constraints on the planet's atmospheric properties, we perform the first full optical to infrared retrieval analysis for this planet. The retrieved spectrum is consistent with a limb temperature of 1248$\pm$92 K, a thick cloud deck, enhanced Rayleigh scattering, and $\sim$10x solar H2O abundance. We find log($Z/Z_{\odot}$) = 2.41$_{-0.07}^{+0.06}$, in agreement with the mass-metallicity relation derived for the Solar System.

astro-ph.EP↗

arXiv2020

Detection of Na, K and H$_2$O in the hazy atmosphere of WASP-6b

We present new observations of the transmission spectrum of the hot Jupiter WASP-6b both from the ground with the Very Large Telescope (VLT) FOcal Reducer and Spectrograph (FORS2) from 0.45-0.83 $μ$m, and space with the Transiting Exoplanet Survey Satellite (TESS) from 0.6-1.0 $μ$m and the Hubble Space Telescope (HST) Wide Field Camera 3 from 1.12-1.65 $μ$m. Archival data from the HST Space Telescope Imaging Spectrograph (STIS) and Spitzer is also reanalysed on a common Gaussian process framework, of which the STIS data show a good overall agreement with the overlapping FORS2 data. We also explore the effects of stellar heterogeneity on our observations and its resulting implications towards determining the atmospheric characteristics of WASP-6b. Independent of our assumptions for the level of stellar heterogeneity we detect Na I, K I and H$_2$O absorption features and constrain the elemental oxygen abundance to a value of [O/H] $\simeq -0.9\pm0.3$ relative to solar. In contrast, we find that the stellar heterogeneity correction can have significant effects on the retrieved distributions of the [Na/H] and [K/H] abundances, primarily through its degeneracy with the sloping optical opacity of scattering haze species within the atmosphere. Our results also show that despite this presence of haze, WASP-6b remains a favourable object for future atmospheric characterisation with upcoming missions such as the James Webb Space Telescope.

astro-ph.EP↗

arXiv2020

Optical to near-infrared transmission spectrum of the warm sub-Saturn HAT-P-12b

We present the transmission spectrum of HAT-P-12b through a joint analysis of data obtained from the Hubble Space Telescope Space Telescope Imaging Spectrograph (STIS) and Wide Field Camera 3 (WFC3) and Spitzer, covering the wavelength range 0.3-5.0 $μ$m. We detect a muted water vapor absorption feature at 1.4 $μ$m attenuated by clouds, as well as a Rayleigh scattering slope in the optical indicative of small particles. We interpret the transmission spectrum using both the state-of-the-art atmospheric retrieval code SCARLET and the aerosol microphysics model CARMA. These models indicate that the atmosphere of HAT-P-12b is consistent with a broad range of metallicities between several tens to a few hundred times solar, a roughly solar C/O ratio, and moderately efficient vertical mixing. Cloud models that include condensate clouds do not readily generate the sub-micron particles necessary to reproduce the observed Rayleigh scattering slope, while models that incorporate photochemical hazes composed of soot or tholins are able to match the full transmission spectrum. From a complementary analysis of secondary eclipses by Spitzer, we obtain measured depths of $0.042\%\pm0.013\%$ and $0.045\%\pm0.018\%$ at 3.6 and 4.5 $μ$m, respectively, which are consistent with a blackbody temperature of $890^{+60}_{-70}$ K and indicate efficient day-night heat recirculation. HAT-P-12b joins the growing number of well-characterized warm planets that underscore the importance of clouds and hazes in our understanding of exoplanet atmospheres.

astro-ph.EP↗

arXiv2020

Efficient Generation of Low Autocorrelation Binary Sequences

Simple and efficient algorithm based on heuristic search by shotgun hill climbing to construct binary sequences with small peak sidelobe levels (PSL) is suggested. The algorithm is applied for generation of binary sequences of lengths between 106 and 300. Improvements are obtained in almost half of the considered lengths while for the rest of the lengths, binary sequences with the same PSL values as reported in the state-of-the-art publications are found.

eess.SP↗

arXiv2020

Non-detection of TiO and VO in the atmosphere of WASP-121b using high-resolution spectroscopy

Thermal inversions have long been predicted to exist in the atmospheres of ultra-hot Jupiters. However, detection of two species thought to be responsible -- TiO and VO -- remain elusive. We present a search for TiO and VO in the atmosphere of the ultra-hot Jupiter WASP-121b ($T_\textrm{eq} \gtrsim 2400$ K), an exoplanet already known to show water features in its dayside spectrum characteristic of a temperature inversion as well as tentative evidence for VO at low-resolution. We observed its transmission spectrum with UVES/VLT and used the cross-correlation method -- a powerful tool for the unambiguous identification of the presence of atomic and molecular species -- in an effort to detect whether TiO or VO were responsible for the observed temperature inversion. No evidence for the presence of TiO or VO was found at the terminator of WASP-121b. By injecting signals into our data at varying abundance levels, we set rough detection limits of $[\text{VO}] \lesssim -7.9$ and $[\text{TiO}] \lesssim -9.3$. However, these detection limits are largely degenerate with scattering properties and the position of the cloud deck. Our results may suggest that neither TiO or VO are the main drivers of the thermal inversion in WASP-121b, but until a more accurate line list is developed for VO, we cannot conclusively rule out its presence. Future work will search for finding other strong optically-absorbing species that may be responsible for the excess absorption in the red-optical.

astro-ph.EP↗

arXiv2020

Detection of Fe I in the atmosphere of the ultra-hot Jupiter WASP-121b, and a new likelihood-based approach for Doppler-resolved spectroscopy

High-resolution Doppler-resolved spectroscopy has opened up a new window into the atmospheres of both transiting and non-transiting exoplanets. Here, we present VLT/UVES observations of a transit of WASP-121b, an 'ultra-hot' Jupiter previously found to exhibit a temperature inversion and detections of multiple species at optical wavelengths. We present initial results using the blue arm of UVES ($\approx$3700-5000A), recovering a clear signal of neutral Fe in the planet's atmosphere at >8$σ$, which could contribute to (or even fully explain) the temperature inversion in the stratosphere. However, using standard cross-correlation methods, it is difficult to extract physical parameters such as temperature and abundances. Recent pioneering efforts have sought to develop likelihood `mappings' that can be used to directly fit models to high-resolution datasets. We introduce a new framework that directly computes the likelihood of the model fit to the data, and can be used to explore the posterior distribution of parameterised model atmospheres via MCMC techniques. Our method also recovers the physical extent of the atmosphere, as well as account for time- and wavelength-dependent uncertainties. We measure a temperature of $3710^{+490}_{-510}$K, indicating a higher temperature in the upper atmosphere when compared to low-resolution observations. We also show that the Fe I signal is physically separated from the exospheric Fe II. However, the temperature measurements are highly degenerate with aerosol properties; detection of additional species, using more sophisticated atmospheric models, or combining these methods with low-resolution spectra should help break these degeneracies.

astro-ph.EP↗

arXiv2019

Urban Driving with Conditional Imitation Learning

Hand-crafting generalised decision-making rules for real-world urban autonomous driving is hard. Alternatively, learning behaviour from easy-to-collect human driving demonstrations is appealing. Prior work has studied imitation learning (IL) for autonomous driving with a number of limitations. Examples include only performing lane-following rather than following a user-defined route, only using a single camera view or heavily cropped frames lacking state observability, only lateral (steering) control, but not longitudinal (speed) control and a lack of interaction with traffic. Importantly, the majority of such systems have been primarily evaluated in simulation - a simple domain, which lacks real-world complexities. Motivated by these challenges, we focus on learning representations of semantics, geometry and motion with computer vision for IL from human driving demonstrations. As our main contribution, we present an end-to-end conditional imitation learning approach, combining both lateral and longitudinal control on a real vehicle for following urban routes with simple traffic. We address inherent dataset bias by data balancing, training our final policy on approximately 30 hours of demonstrations gathered over six months. We evaluate our method on an autonomous vehicle by driving 35km of novel routes in European urban streets.

cs.CV↗

arXiv2019

Transmission Spectroscopy of WASP-79b from 0.6 to 5.0 $μ$m

As part of the PanCET program, we have conducted a spectroscopic study of WASP-79b, an inflated hot Jupiter orbiting an F-type star in Eridanus with a period of 3.66 days. Building on the original WASP and TRAPPIST photometry of Smalley et al (2012), we examine HST/WFC3 (1.125 - 1.650 $μ$m), Magellan/LDSS-3C (0.6 - 1 $μ$m) data, and Spitzer data (3.6 and 4.5 $μ$m). Using data from all three instruments, we constrain the water abundance to be --2.20 $\leq$ log(H$_2$O) $\leq$ --1.55. We present these results along with the results of an atmospheric retrieval analysis, which favor inclusion of FeH and H$^-$ in the atmospheric model. We also provide an updated ephemeris based on the Smalley, HST/WFC3, LDSS-3C, Spitzer, and TESS transit times. With the detectable water feature and its occupation of the clear/cloudy transition region of the temperature/gravity phase space, WASP-79b is a target of interest for the approved JWST Director's Discretionary Early Release Science (DD ERS) program, with ERS observations planned to be the first to execute in Cycle 1. Transiting exoplanets have been approved for 78.1 hours of data collection, and with the delay in the JWST launch, WASP-79b is now a target for the Panchromatic Transmission program. This program will observe WASP-79b for 42 hours in 4 different instrument modes, providing substantially more data by which to investigate this hot Jupiter.

astro-ph.EP↗

arXiv2019

A Hubble PanCET Study of HAT-P-11b: A Cloudy Neptune with a Low Atmospheric Metallicity

We present the first comprehensive look at the $0.35-5$ $μ$m transmission spectrum of the warm ($\sim 800$ K) Neptune HAT-P-11b derived from thirteen individual transits observed using the Hubble and Spitzer Space Telescopes. Along with the previously published molecular absorption feature in the $1.1-1.7$ $μ$m bandpass, we detect a distinct absorption feature at 1.15 $μ$m and a weak feature at 0.95 $μ$m, indicating the presence of water and/or methane with a combined significance of 4.4 $σ$. We find that this planet's nearly flat optical transmission spectrum and attenuated near-infrared molecular absorption features are best-matched by models incorporating a high-altitude cloud layer. Atmospheric retrievals using the combined $0.35-1.7$ $μ$m HST transmission spectrum yield strong constraints on atmospheric cloud-top pressure and metallicity, but we are unable to match the relatively shallow Spitzer transit depths without under-predicting the strength of the near-infrared molecular absorption bands. HAT-P-11b's HST transmission spectrum is well-matched by predictions from our microphysical cloud models. Both forward models and retrievals indicate that HAT-P-11b most likely has a relatively low atmospheric metallicity ($<4.6 \; Z_{\odot}$ and $<86 \; Z_{\odot}$ at the $2 σ$ and $3 σ$ levels respectively), in contrast to the expected trend based on the solar system planets. Our work also demonstrates that the wide wavelength coverage provided by the addition of the HST STIS data is critical for making these inferences.

astro-ph.EP↗

arXiv2019

The HST PanCET Program: Exospheric Mg II and Fe II in the Near-UV transmission spectrum of WASP-121b using Jitter Decorrelation

We present HST near-ultraviolet (NUV) transits of the hot Jupiter WASP-121b, acquired as part of the PanCET program. Time series spectra during two transit events were used to measure the transmission spectra between 2280 and 3070 Angstroms at a resolution of 30,000. Using HST data from 61 STIS visits, we show that data from HST's Pointing Control System can be used to decorrelate the instrument systematic errors (Jitter Decorrelation), which we used to fit the WASP-121b light curves. The NUV spectrum show very strong absorption features, with the NUV white light curve found to be larger than the average optical and near-infrared value at 6-$σ$ confidence. We identify and spectrally resolve absorption from the Mg ii doublet in the planetary exosphere at a 5.9-$σ$ confidence level. The Mg ii doublet is observed to reach altitudes of $R_{pl}/R_{star}=0.284\pm0.037$ for the 2796 Angstrom line and $0.242\pm0.0431$ in the 2804 Angstrom line, which exceeds the Roche lobe size as viewed in transit geometry ($R_{\rm eqRL}/R_{star}$ = 0.158). We also detect and resolve strong features of the Fe ii UV1 and UV2 multiplets, and observe the lines reaching altitudes of $R_{pl}/R_{star}\approx0.3$. At these high altitudes, the atmospheric Mg ii and Fe ii gas is not gravitationally bound to the planet, and these ionized species may be hydrodynamically escaping or could be magnetically confined. Refractory Mg and Fe atoms at high altitudes also indicates that these species are not trapped into condensate clouds at depth, which places constraints on the deep interior temperature.

astro-ph.EP↗

arXiv2019

An emission spectrum for WASP-121b measured across the 0.8-1.1 micron wavelength range using the Hubble Space Telescope

WASP-121b is a transiting gas giant exoplanet orbiting close to its Roche limit, with an inflated radius nearly double that of Jupiter and a dayside temperature comparable to a late M dwarf photosphere. Secondary eclipse observations covering the 1.1-1.6 micron wavelength range have revealed an atmospheric thermal inversion on the dayside hemisphere, likely caused by high altitude absorption at optical wavelengths. Here we present secondary eclipse observations made with the Hubble Space Telescope Wide Field Camera 3 spectrograph that extend the wavelength coverage from 1.1 micron down to 0.8 micron. To determine the atmospheric properties from the measured eclipse spectrum, we performed a retrieval analysis assuming chemical equilibrium, with the effects of thermal dissociation and ionization included. Our best-fit model provides a good fit to the data with reduced chi^2=1.04. The data diverge from a blackbody spectrum and instead exhibit emission due to H- shortward of 1.1 micron. The best-fit model does not reproduce a previously reported bump in the spectrum at 1.25 micron, possibly indicating this feature is a statistical fluctuation in the data rather than a VO emission band as had been tentatively suggested. We estimate an atmospheric metallicity of [M/H]=1.09(-0.69,+0.57), and fit for the carbon and oxygen abundances separately, obtaining [C/H]=-0.29(-0.48,+0.61) and [O/H]=0.18(-0.60,+0.64). The corresponding carbon-to-oxygen ratio is C/O=0.49(-0.37,+0.65), which encompasses the solar value of 0.54, but has a large uncertainty.

astro-ph.EP↗

arXiv2019

Information-Directed Exploration for Deep Reinforcement Learning

Efficient exploration remains a major challenge for reinforcement learning. One reason is that the variability of the returns often depends on the current state and action, and is therefore heteroscedastic. Classical exploration strategies such as upper confidence bound algorithms and Thompson sampling fail to appropriately account for heteroscedasticity, even in the bandit setting. Motivated by recent findings that address this issue in bandits, we propose to use Information-Directed Sampling (IDS) for exploration in reinforcement learning. As our main contribution, we build on recent advances in distributional reinforcement learning and propose a novel, tractable approximation of IDS for deep Q-learning. The resulting exploration strategy explicitly accounts for both parametric uncertainty and heteroscedastic observation noise. We evaluate our method on Atari games and demonstrate a significant improvement over alternative approaches.

cs.LG↗