Searcharxiv⌕ Search

arXiv subjects

Michael Zhang

Publications and source records attributed to Michael Zhang.

At least 73 records · Page 4Linked to original sources
arXiv2024

JWST transmission spectroscopy of HD 209458b: a super-solar metallicity, a very low C/O, and no evidence of CH4, HCN, or C2H2

We present the transmission spectrum of the original transiting hot Jupiter HD\,209458b from 2.3 -- 5.1 $μ$m as observed with the NIRCam instrument on the James Webb Space Telescope (JWST). Previous studies of HD 209458b's atmosphere have given conflicting results on the abundance of H$_2$O and the presence of carbon- and nitrogen-bearing species, which have significant ramifications on the inferences of the planet's metallicity (M/H) and carbon-to-oxygen (C/O) ratio. We detect strong features of H$_2$O and CO$_2$ in the JWST transmission spectrum, which when interpreted using a retrieval that assumes thermochemical equilibrium and fractional grey cloud opacity yields $3^{+4}_{-1}$ $\times$ solar metallicity and C/O = $0.11^{+0.12}_{-0.06}$. The derived metallicity is consistent with the atmospheric metallicity-planet mass trend observed in solar gas giants. The low C/O ratio suggests that this planet has undergone significant contamination by evaporating planetesimals while migrating inward. We are also able to place upper limits on the abundances of CH$_4$, C$_2$H$_2$ and HCN of log($χ_{\mathrm{CH}_4}$) = -5.6, log($χ_{\mathrm{C}_2\mathrm{H}_2}$) = -5.7, and log($χ_{\mathrm{HCN}}$) = -5.1, which are in tension with the recent claim of a detection of these species using ground-based cross-correlation spectroscopy. We find that HD\,209458b has a weaker CO$_2$ feature size than WASP-39b when comparing their scale-height-normalized transmission spectra. On the other hand, the size of HD 209458b's H$_2$O feature is stronger, thus reinforcing the low C/O inference.

astro-ph.EP↗
arXiv2024

The Hedgehog & the Porcupine: Expressive Linear Attentions with Softmax Mimicry

Linear attentions have shown potential for improving Transformer efficiency, reducing attention's quadratic complexity to linear in sequence length. This holds exciting promise for (1) training linear Transformers from scratch, (2) "finetuned-conversion" of task-specific Transformers into linear versions that recover task performance, and (3) "pretrained-conversion" of Transformers such as large language models into linear versions finetunable on downstream tasks. However, linear attentions often underperform standard softmax attention in quality. To close this performance gap, we find prior linear attentions lack key properties of softmax attention tied to good performance: low-entropy (or "spiky") weights and dot-product monotonicity. We further observe surprisingly simple feature maps that retain these properties and match softmax performance, but are inefficient to compute in linear attention. We thus propose Hedgehog, a learnable linear attention that retains the spiky and monotonic properties of softmax attention while maintaining linear complexity. Hedgehog uses simple trainable MLPs to produce attention weights mimicking softmax attention. Experiments show Hedgehog recovers over 99% of standard Transformer quality in train-from-scratch and finetuned-conversion settings, outperforming prior linear attentions up to 6 perplexity points on WikiText-103 with causal GPTs, and up to 8.7 GLUE score points on finetuned bidirectional BERTs. Hedgehog also enables pretrained-conversion. Converting a pretrained GPT-2 into a linear attention variant achieves state-of-the-art 16.7 perplexity on WikiText-103 for 125M subquadratic decoder models. We finally turn a pretrained Llama-2 7B into a viable linear attention Llama. With low-rank adaptation, Hedgehog-Llama2 7B achieves 28.1 higher ROUGE-1 points over the base standard attention model, where prior linear attentions lead to 16.5 point drops.

cs.LG↗
arXiv2024

GJ 367b is a dark, hot, airless sub-Earth

We present the mid-infrared (5-12 $μ$m) phase curve of GJ 367b observed by the Mid-Infrared Instrument (MIRI) on the James Webb Space Telescope (JWST). GJ 367b is a hot (T_eq=1370 K), extremely dense (10.2 +- 1.3 g/cm^3) sub-Earth orbiting a M dwarf on a 0.32 d orbit. We measure an eclipse depth of 79 +- 4 ppm, a nightside planet-to-star flux ratio of 4 +- 8 ppm, and a relative phase amplitude of 0.97 +- 0.10 -- all fully consistent with a zero-albedo planet with no heat recirculation. Su ch a scenario is also consistent with the phase offset of 11 +- 5 degrees East to within 2.2$σ$. The emission spectrum is likewise consistent with a blackbody with no heat redistribution and a low albedo of A_B ~ 0.1, with the exception of one anomalous wavelength bin that we attribute to unexplained systematics. The emission spectrum puts few constraints on the surface composition, but rules out a CO2 atmosphere >~ 1 bar, an outgassed atmosphere >~10 mbar (under heavily reducing conditions), or an outgassed atmosphere >~0.01 mbar (under heavily oxidizing conditions). The lack of day-night heat recirculation implies that 1 bar atmospheres are ruled out for a wide range of compositions, while 0.1 bar atmospheres are consistent with the data. Taken together with the fact that most of the dayside should be molten, our JWST observations suggest the planet must have lost the vast majority of its initial inventory of volatiles.

astro-ph.EP↗
arXiv2024

Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b

Hot Jupiters are among the best-studied exoplanets, but it is still poorly understood how their chemical composition and cloud properties vary with longitude. Theoretical models predict that clouds may condense on the nightside and that molecular abundances can be driven out of equilibrium by zonal winds. Here we report a phase-resolved emission spectrum of the hot Jupiter WASP-43b measured from 5-12 $μ$m with JWST's Mid-Infrared Instrument (MIRI). The spectra reveal a large day-night temperature contrast (with average brightness temperatures of 1524$\pm$35 and 863$\pm$23 Kelvin, respectively) and evidence for water absorption at all orbital phases. Comparisons with three-dimensional atmospheric models show that both the phase curve shape and emission spectra strongly suggest the presence of nightside clouds which become optically thick to thermal emission at pressures greater than ~100 mbar. The dayside is consistent with a cloudless atmosphere above the mid-infrared photosphere. Contrary to expectations from equilibrium chemistry but consistent with disequilibrium kinetics models, methane is not detected on the nightside (2$σ$ upper limit of 1-6 parts per million, depending on model assumptions).

astro-ph.EP↗
arXiv2024

Differential Privacy in Aggregated Mobility Networks: Balancing Privacy and Utility

Location data is collected from users continuously to understand their mobility patterns. Releasing the user trajectories may compromise user privacy. Therefore, the general practice is to release aggregated location datasets. However, private information may still be inferred from an aggregated version of location trajectories. Differential privacy (DP) protects the query output against inference attacks regardless of background knowledge. This paper presents a differential privacy-based privacy model that protects the user's origins and destinations from being inferred from aggregated mobility datasets. This is achieved by injecting Planar Laplace noise to the user origin and destination GPS points. The noisy GPS points are then transformed into a link representation using a link-matching algorithm. Finally, the link trajectories form an aggregated mobility network. The injected noise level is selected using the Sparse Vector Mechanism. This DP selection mechanism considers the link density of the location and the functional category of the localized links. Compared to the different baseline models, including a k-anonymity method, our differential privacy-based aggregation model offers query responses that are close to the raw data in terms of aggregate statistics at both the network and trajectory-levels with maximum 9% deviation from the baseline in terms of network length.

cs.CR↗
arXiv2023

Enabling CMF Estimation in Data-Constrained Scenarios: A Semantic-Encoding Knowledge Mining Model

Precise estimation of Crash Modification Factors (CMFs) is central to evaluating the effectiveness of various road safety treatments and prioritizing infrastructure investment accordingly. While customized study for each countermeasure scenario is desired, the conventional CMF estimation approaches rely heavily on the availability of crash data at given sites. This not only makes the estimation costly, but the results are also less transferable, since the intrinsic similarities between different safety countermeasure scenarios are not fully explored. Aiming to fill this gap, this study introduces a novel knowledge-mining framework for CMF prediction. This framework delves into the connections of existing countermeasures and reduces the reliance of CMF estimation on crash data availability and manual data collection. Specifically, it draws inspiration from human comprehension processes and introduces advanced Natural Language Processing (NLP) techniques to extract intricate variations and patterns from existing CMF knowledge. It effectively encodes unstructured countermeasure scenarios into machine-readable representations and models the complex relationships between scenarios and CMF values. This new data-driven framework provides a cost-effective and adaptable solution that complements the case-specific approaches for CMF estimation, which is particularly beneficial when availability of crash data or time imposes constraints. Experimental validation using real-world CMF Clearinghouse data demonstrates the effectiveness of this new approach, which shows significant accuracy improvements compared to baseline methods. This approach provides insights into new possibilities of harnessing accumulated transportation knowledge in various applications.

cs.LG↗
arXiv2023

Joint Optimization of Traffic Signal Control and Vehicle Routing in Signalized Road Networks using Multi-Agent Deep Reinforcement Learning

Urban traffic congestion is a critical predicament that plagues modern road networks. To alleviate this issue and enhance traffic efficiency, traffic signal control and vehicle routing have proven to be effective measures. In this paper, we propose a joint optimization approach for traffic signal control and vehicle routing in signalized road networks. The objective is to enhance network performance by simultaneously controlling signal timings and route choices using Multi-Agent Deep Reinforcement Learning (MADRL). Signal control agents (SAs) are employed to establish signal timings at intersections, whereas vehicle routing agents (RAs) are responsible for selecting vehicle routes. By establishing relevance between agents and enabling them to share observations and rewards, interaction and cooperation among agents are fostered, which enhances individual training. The Multi-Agent Advantage Actor-Critic algorithm is used to handle multi-agent environments, and Deep Neural Network (DNN) structures are designed to facilitate the algorithm's convergence. Notably, our work is the first to utilize MADRL in determining the optimal joint policy for signal control and vehicle routing. Numerical experiments conducted on the modified Sioux network demonstrate that our integration of signal control and vehicle routing outperforms controlling signal timings or vehicles' routes alone in enhancing traffic efficiency.

eess.SY↗
arXiv2023

Outflowing helium from a mature mini-Neptune

We announce the detection of escaping helium from TOI 2134b, a mini-Neptune a few Gyr old. The average in-transit absorption spectrum shows a peak of 0.37 +- 0.05% and an equivalent width of $W_{\rm avg}=3.3 \pm 0.3$ m$Å$. Among all planets with helium detections, TOI 2134b is the only mature mini-Neptune, has the smallest helium signal, and experiences the lowest XUV flux. Putting TOI 2134b in the context of all other helium detections, we report the detection of a strong (p=3.0e-5) and theoretically expected correlation between $F_{\rm XUV}/ρ_{\rm XUV}$ (proportional to the energy-limited mass loss rate) and $R_* W_{\rm avg}$ (roughly proportional to the observationally inferred mass loss rate). Here, $W_{\rm avg}$ is the equivalent width of the helium absorption and $ρ_{\rm XUV}$ is the density of the planet within the XUV photosphere, but the correlation is similarly strong if we use the optical photosphere. TOI 2134b anchors the relation, having the lowest value on both axes. We encourage further observations to fill in missing regions of this parameter space and improve estimates of $F_{\rm XUV}$.

astro-ph.EP↗
arXiv2023

Confirmation of sub-solar metallicity for WASP-77Ab from JWST thermal emission spectroscopy

We present the dayside thermal emission spectrum of WASP-77Ab from 2.8 -- 5.2 $μ$m as observed with the NIRSpec instrument on the James Webb Space Telescope (JWST). WASP-77Ab was previously found to have a sub-solar metallicity and a solar carbon-to-oxygen (C/O) ratio from H$_2$O and CO absorption lines detected using high-resolution spectroscopy. By performing atmospheric retrievals on the JWST spectrum assuming chemical equilibrium, we find a sub-solar metallicity [M/H]=$-0.91^{+0.24}_{-0.16}$ and C/O ratio $0.36^{+0.10}_{-0.09}$. We identify H$_2$O and CO and constrain their abundances, and we find no CO$_2$ in the spectrum. The JWST and high-resolution spectroscopy results agree within $\sim1σ$ for the metallicity and within 1.8$σ$ for the C/O ratio. However, our results fit less well in the picture painted by the shorter wavelength spectrum measured by HST WFC3. Comparing the JWST thermal emission spectra of WASP-77Ab and HD 149026b shows that both hot Jupiters have nearly identical brightness temperatures in the near-infrared, but distinctly different atmospheric compositions. Our results reaffirm high-resolution spectroscopy as a powerful and reliable method to measure molecular abundances. Our results also highlight the incredible diversity of hot Jupiter atmospheric compositions.

astro-ph.EP↗
arXiv2023

Deep Learning and Symbolic Regression for Discovering Parametric Equations

Symbolic regression is a machine learning technique that can learn the governing formulas of data and thus has the potential to transform scientific discovery. However, symbolic regression is still limited in the complexity and dimensionality of the systems that it can analyze. Deep learning on the other hand has transformed machine learning in its ability to analyze extremely complex and high-dimensional datasets. We propose a neural network architecture to extend symbolic regression to parametric systems where some coefficient may vary but the structure of the underlying governing equation remains constant. We demonstrate our method on various analytic expressions, ODEs, and PDEs with varying coefficients and show that it extrapolates well outside of the training domain. The neural network-based architecture can also integrate with other deep learning architectures so that it can analyze high-dimensional data while being trained end-to-end. To this end we integrate our architecture with convolutional neural networks to analyze 1D images of varying spring systems.

cs.LG↗
arXiv2023

A reflective, metal-rich atmosphere for GJ 1214b from its JWST phase curve

There are no planets intermediate in size between Earth and Neptune in our Solar System, yet these objects are found around a substantial fraction of other stars. Population statistics show that close-in planets in this size range bifurcate into two classes based on their radii. It is hypothesized that the group with larger radii (referred to as "sub-Neptunes") is distinguished by having hydrogen-dominated atmospheres that are a few percent of the total mass of the planets. GJ 1214b is an archetype sub-Neptune that has been observed extensively using transmission spectroscopy to test this hypothesis. However, the measured spectra are featureless, and thus inconclusive, due to the presence of high-altitude aerosols in the planet's atmosphere. Here we report a spectroscopic thermal phase curve of GJ 1214b obtained with JWST in the mid-infrared. The dayside and nightside spectra (average brightness temperatures of 553 $\pm$ 9 and 437 $\pm$ 19 K, respectively) each show >3$σ$ evidence of absorption features, with H$_2$O as the most likely cause in both. The measured global thermal emission implies that GJ 1214b's Bond albedo is 0.51 $\pm$ 0.06. Comparison between the spectroscopic phase curve data and three-dimensional models of GJ 1214b reveal a planet with a high metallicity atmosphere blanketed by a thick and highly reflective layer of clouds or haze.

astro-ph.EP↗
arXiv2023

The Hazy and Metal-Rich Atmosphere of GJ 1214 b Constrained by Near and Mid-Infrared Transmission Spectroscopy

The near-infrared transmission spectrum of the warm sub-Neptune exoplanet GJ 1214 b has been observed to be flat and featureless, implying a high metallicity atmosphere with abundant aerosols. Recent JWST MIRI LRS observations of a phase curve of GJ 1214 b showed that its transmission spectrum is flat out into the mid-infrared. In this paper, we use the combined near- and mid-infrared transmission spectrum of GJ 1214 b to constrain its atmospheric composition and aerosol properties. We generate a grid of photochemical haze models using an aerosol microphysics code for a number of background atmospheres spanning metallicities from 100 to 1000 $\times$ solar, as well as a steam atmosphere scenario. The flatness of the combined data set largely rules out atmospheric metallicities $\leq$300 $\times$ solar due to their large corresponding molecular feature amplitudes, preferring values $\geq$1000 $\times$ solar and column haze production rates $\geq$10$^{-10}$ g cm$^{-2}$ s$^{-1}$. The steam atmosphere scenario with similarly high haze production rates also exhibit sufficiently small molecular features to be consistent with the transmission spectrum. These compositions imply that atmospheric mean molecular weights $\geq$15 g mol$^{-1}$ are needed to fit the data. Our results suggest that haze production is highly efficient on GJ 1214 b and could involve non-hydrocarbon, non-nitrogen haze precursors. Further characterization of GJ 1214 b's atmosphere would likely require multiple transits and eclipses using JWST across the near and mid-infrared, potentially complemented by groundbased high resolution transmission spectroscopy.

astro-ph.EP↗
arXiv2023

EVKG: An Interlinked and Interoperable Electric Vehicle Knowledge Graph for Smart Transportation System

Over the past decade, the electric vehicle industry has experienced unprecedented growth and diversification, resulting in a complex ecosystem. To effectively manage this multifaceted field, we present an EV-centric knowledge graph (EVKG) as a comprehensive, cross-domain, extensible, and open geospatial knowledge management system. The EVKG encapsulates essential EV-related knowledge, including EV adoption, electric vehicle supply equipment, and electricity transmission network, to support decision-making related to EV technology development, infrastructure planning, and policy-making by providing timely and accurate information and analysis. To enrich and contextualize the EVKG, we integrate the developed EV-relevant ontology modules from existing well-known knowledge graphs and ontologies. This integration enables interoperability with other knowledge graphs in the Linked Data Open Cloud, enhancing the EVKG's value as a knowledge hub for EV decision-making. Using six competency questions, we demonstrate how the EVKG can be used to answer various types of EV-related questions, providing critical insights into the EV ecosystem. Our EVKG provides an efficient and effective approach for managing the complex and diverse EV industry. By consolidating critical EV-related knowledge into a single, easily accessible resource, the EVKG supports decision-makers in making informed choices about EV technology development, infrastructure planning, and policy-making. As a flexible and extensible platform, the EVKG is capable of accommodating a wide range of data sources, enabling it to evolve alongside the rapidly changing EV landscape.

cs.AI↗
arXiv2023

High atmospheric metal enrichment for a Saturn-mass planet

Atmospheric metal enrichment (i.e., elements heavier than helium, also called "metallicity") is a key diagnostic of the formation of giant planets. The giant planets of the solar system exhibit an inverse relationship between mass and both their bulk metallicities and atmospheric metallicities. Extrasolar giant planets also display an inverse relationship between mass and bulk metallicity. However, there is significant scatter in the relationship and it is not known how atmospheric metallicity correlates with either planet mass or bulk metallicity. Here we show that the Saturn-mass exoplanet HD 149026b has an atmospheric metallicity 59 - 276 times solar (at 1 $σ$), which is greater than Saturn's atmospheric metallicity of ~7.5 times solar at >4 $σ$ confidence. This result is based on modeling CO$_2$ and H$_2$O absorption features in the thermal emission spectrum of the planet measured by JWST. HD 149026b is the most metal-rich giant planet known, with an estimated bulk heavy element abundance of 66 $\pm$ 2% by mass. We find that the atmospheric metallicities of both HD 149026b and the solar system giant planets are more correlated with bulk metallicity than planet mass.

astro-ph.EP↗
arXiv2023

Effectively Modeling Time Series with Simple Discrete State Spaces

Time series modeling is a well-established problem, which often requires that methods (1) expressively represent complicated dependencies, (2) forecast long horizons, and (3) efficiently train over long sequences. State-space models (SSMs) are classical models for time series, and prior works combine SSMs with deep learning layers for efficient sequence modeling. However, we find fundamental limitations with these prior approaches, proving their SSM representations cannot express autoregressive time series processes. We thus introduce SpaceTime, a new state-space time series architecture that improves all three criteria. For expressivity, we propose a new SSM parameterization based on the companion matrix -- a canonical representation for discrete-time processes -- which enables SpaceTime's SSM layers to learn desirable autoregressive processes. For long horizon forecasting, we introduce a "closed-loop" variation of the companion SSM, which enables SpaceTime to predict many future time-steps by generating its own layer-wise inputs. For efficient training and inference, we introduce an algorithm that reduces the memory and compute of a forward pass with the companion matrix. With sequence length $\ell$ and state-space size $d$, we go from $\tilde{O}(d \ell)$ naïvely to $\tilde{O}(d + \ell)$. In experiments, our contributions lead to state-of-the-art results on extensive and diverse benchmarks, with best or second-best AUROC on 6 / 7 ECG and speech time series classification, and best MSE on 14 / 16 Informer forecasting tasks. Furthermore, we find SpaceTime (1) fits AR($p$) processes that prior deep SSMs fail on, (2) forecasts notably more accurately on longer horizons than prior state-of-the-art, and (3) speeds up training on real-world ETTh1 data by 73% and 80% relative wall-clock time over Transformers and LSTMs.

cs.LG↗
arXiv2023

Simple Hardware-Efficient Long Convolutions for Sequence Modeling

State space models (SSMs) have high performance on long sequence modeling but require sophisticated initialization techniques and specialized implementations for high quality and runtime performance. We study whether a simple alternative can match SSMs in performance and efficiency: directly learning long convolutions over the sequence. We find that a key requirement to achieving high performance is keeping the convolution kernels smooth. We find that simple interventions--such as squashing the kernel weights--result in smooth kernels and recover SSM performance on a range of tasks including the long range arena, image classification, language modeling, and brain data modeling. Next, we develop FlashButterfly, an IO-aware algorithm to improve the runtime performance of long convolutions. FlashButterfly appeals to classic Butterfly decompositions of the convolution to reduce GPU memory IO and increase FLOP utilization. FlashButterfly speeds up convolutions by 2.2$\times$, and allows us to train on Path256, a challenging task with sequence length 64K, where we set state-of-the-art by 29.1 points while training 7.2$\times$ faster than prior work. Lastly, we introduce an extension to FlashButterfly that learns the coefficients of the Butterfly decomposition, increasing expressivity without increasing runtime. Using this extension, we outperform a Transformer on WikiText103 by 0.2 PPL with 30% fewer parameters.

cs.LG↗
arXiv2023

Detection of Atmospheric Escape from Four Young Mini Neptunes

We use Keck/NIRSPEC to survey a sample of of young ($<$1 Gyr), short period mini Neptunes orbiting nearby K dwargs to measure their mass loss via the metastable helium line. We detect helium absorption from all four of the targets in our initial sample. The first detection, around TOI 560b, was announced in a previous paper. We now announce three additional detections around TOI 1430.01, 2076b, and TOI 1683.01. All four planets show an average in-transit excess absorption of 0.7--1.0%. However, the outflows differ in their kinematic properties. TOI 1430b exhibits pre-ingress absorption, while TOI 2076b's outflow is exceptionally optically thick and shows significant post-egress absorption. For all four planets, the width of the measured helium absorption signal is consistent with expectations for a photoevaporative outflow (10--30 km/s, 5000--10,000 K). Unless broadening mechanisms other than thermal velocity and the bulk outflow velocity are significant, our observations disfavor core-powered mass loss models, which predict much slower (1--3 km/s) outflows. We utilize both an isothermal Parker wind model and an order-of-magnitude method to estimate the mass loss timescale, and obtain $\sim$ a few hundred Myr for each planet. We conclude that many, if not all, of these planets will lose their hydrogen-rich envelopes and become super Earths. Our results demonstrate that most mini Neptunes orbiting sun-like stars have primordial atmospheres, and that photoevaporation is an efficient mechanism for stripping these atmospheres and transforming these planets into super Earths.

astro-ph.EP↗
arXiv2022

More Evidence for Variable Helium Absorption from HD 189733b

We present a new Keck/NIRSPEC observation of metastable helium absorption from the upper atmosphere of HD 189733b, a hot Jupiter orbiting a nearby moderately active star. We measure an average helium transit depth of $0.420 \pm 0.013$% integrated over the [-20, 20] km/s velocity range. Comparing this measurement to eight previously published transit observations with different instruments, we find that our depth is 32% (9$σ$) lower than the average of the three CARMENES transits, but only 16% (4.4$σ$) lower than the average of the five GIANO transits. We perform 1D hydrodynamical simulations of the outflow, and find that XUV variability on the order of 33%--common for this star--can change the helium absorption depth by 60%. We conclude that changes in stellar XUV flux can explain the observational variability in helium absorption. 3D models are necessary to explore other sources of variability, such as shear instability and changing stellar wind conditions.

astro-ph.EP↗