1967 Publications

Computational mechanisms of distributed value representations and mixed learning strategies

S. Farashahi, A. Soltani

Learning appropriate representations of the reward environment is challenging in the real world where there are many options, each with multiple attributes or features. Despite existence of alternative solutions for this challenge, neural mechanisms underlying emergence and adoption of value representations and learning strategies remain unknown. To address this, we measure learning and choice during a multi-dimensional probabilistic learning task in humans and trained recurrent neural networks (RNNs) to capture our experimental observations. We find that human participants estimate stimulus-outcome associations by learning and combining estimates of reward probabilities associated with the informative feature followed by those of informative conjunctions. Through analyzing representations, connectivity, and lesioning of the RNNs, we demonstrate this mixed learning strategy relies on a distributed neural code and opponency between excitatory and inhibitory neurons through value-dependent disinhibition. Together, our results suggest computational and neural mechanisms underlying emergence of complex learning strategies in naturalistic settings.

Show Abstract

A Comparison of Circumgalactic Mg ii Absorption between the TNG50 Simulation and the MEGAFLOW Survey

Daniel DeFelippis, Nicolas F. Bouché, S. Genel, G. Bryan, Dylan Nelson, Federico Marinacci, Lars Hernquist

The circumgalactic medium (CGM) contains information on gas flows around galaxies, such as accretion and supernova-driven winds, which are difficult to constrain from observations alone. Here, we use the high-resolution TNG50 cosmological magnetohydrodynamical simulation to study the properties and kinematics of the CGM around star-forming galaxies in 1011.5−1012M⊙ halos at z≃ 1 using mock MgII absorption lines, which we generate by postprocessing halos to account for photoionization in the presence of a UV background. We find that the MgII gas is a very good tracer of the cold CGM, which is accreting inward at inflow velocities of up to 50 km s−1. For sight lines aligned with the galaxy's major axis, we find that MgII absorption lines are kinematically shifted due to the cold CGM's significant corotation at speeds up to 50% of the virial velocity for impact parameters up to 60 kpc. We compare mock MgII spectra to observations from the MusE GAs FLow and Wind (MEGAFLOW) survey of strong MgII absorbers (EW2796Å0>0.5Å). After matching the equivalent-width (EW) selection, we find that the mock MgII spectra reflect the diversity of observed kinematics and EWs from MEGAFLOW, even though the sight lines probe a very small fraction of the CGM. MgII absorption in higher-mass halos is stronger and broader than in lower-mass halos but has qualitatively similar kinematics. The median-specific angular momentum of the MgII CGM gas in TNG50 is very similar to that of the entire CGM and only differs from non-CGM components of the halo by normalization factors of ≲ 1 dex.

Show Abstract

A Late-Time Galaxy-Targeted Search for the Radio Counterpart of GW190814

K. D. Alexander, G. Schroeder, K. Paterson, ..., B. Metzger, et. al.

GW190814 was a compact object binary coalescence detected in gravitational waves by Advanced LIGO and Advanced Virgo that garnered exceptional community interest due to its excellent localization and the uncertain nature of the binary's lighter-mass component (either the heaviest known neutron star, or the lightest known black hole). Despite extensive follow up observations, no electromagnetic counterpart has been identified. Here we present new radio observations of 75 galaxies within the localization volume at Δt≈35−266 days post-merger. Our observations cover ∼32% of the total stellar luminosity in the final localization volume and extend to later timescales than previously-reported searches, allowing us to place the deepest constraints to date on the existence of a radio afterglow from a highly off-axis relativistic jet launched during the merger (assuming that the merger occurred within the observed area). For a viewing angle of ∼46∘ (the best-fit binary inclination derived from the gravitational wave signal) and assumed electron and magnetic field energy fractions of ϵe=0.1 and ϵB=0.01, we can rule out a typical short gamma-ray burst-like Gaussian jet with isotropic-equivalent kinetic energy 2×1051 erg propagating into a constant density medium n≳0.01 cm−3. These are the first limits resulting from a galaxy-targeted search for a radio counterpart to a gravitational wave event, and we discuss the challenges, and possible advantages, of applying similar search strategies to future events using current and upcoming radio facilities.

Show Abstract

Algorithms for solving high dimensional PDEs: from nonlinear Monte Carlo to machine learning

Weinan E, J. Han, Arnulf Jentzen

In recent years, tremendous progress has been made on numerical algorithms for solving partial differential equations (PDEs) in a very high dimension, using ideas from either nonlinear (multilevel) Monte Carlo or deep learning. They are potentially free of the curse of dimensionality for many different applications and have been proven to be so in the case of some nonlinear Monte Carlo methods for nonlinear parabolic PDEs. In this paper, we review these numerical and theoretical advances. In addition to algorithms based on stochastic reformulations of the original problem, such as the multilevel Picard iteration and the deep backward stochastic differential equations method, we also discuss algorithms based on the more traditional Ritz, Galerkin, and least square formulations. We hope to demonstrate to the reader that studying PDEs as well as control and variational problems in very high dimensions might very well be among the most promising new directions in mathematics and scientific computing in the near future.

Show Abstract

Final Targeting Strategy for the Sloan Digital Sky Survey IV Apache Point Observatory Galactic Evolution Experiment 2 North Survey

A. Price-Whelan, M. Ness, et. al.

The Apache Point Observatory Galactic Evolution Experiment 2 (APOGEE-2) is a dual-hemisphere, near-infrared (NIR), spectroscopic survey with the goal of producing a chemodynamical mapping of the Milky Way. The targeting for APOGEE-2 is complex and has evolved with time. In this paper, we present the updates and additions to the initial targeting strategy for APOGEE-2N presented in Zasowski et al. (2017). These modifications come in two implementation modes: (i) "Ancillary Science Programs" competitively awarded to Sloan Digital Sky Survey IV PIs through proposal calls in 2015 and 2017 for the pursuit of new scientific avenues outside the main survey, and (ii) an effective 1.5 yr expansion of the survey, known as the Bright Time Extension (BTX), made possible through accrued efficiency gains over the first years of the APOGEE-2N project. For the 23 distinct ancillary programs, we provide descriptions of the scientific aims, target selection, and how to identify these targets within the APOGEE-2 sample. The BTX permitted changes to the main survey strategy, the inclusion of new programs in response to scientific discoveries or to exploit major new data sets not available at the outset of the survey design, and expansions of existing programs to enhance their scientific success and reach. After describing the motivations, implementation, and assessment of these programs, we also leave a summary of lessons learned from nearly a decade of APOGEE-1 and APOGEE-2 survey operations. A companion paper, F. Santana et al. (submitted; AAS29036), provides a complementary presentation of targeting modifications relevant to APOGEE-2 operations in the Southern Hemisphere.

Show Abstract

A New Classification Model for the ZTF Catalog of Periodic Variable Stars

Siu-Hei Cheung, V. Ashley Villar, Ho-Sang Chan, S. Ho

Using the second data release from the Zwicky Transient Facility (ZTF, Bellm et al. 2019), Chen et al. (2020) created a ZTF Catalog of Periodic Variable Stars (ZTF CPVS) of 781, 602 periodic variables stars (PVSs) with 11 class labels. Here, we provide a new classification model of PVSs in the ZTF CPVS using a convolutional variational autoencoder and hierarchical random forest. We cross-match the sky-coordinate of PVSs in the ZTF CPVS with those presented in the SIMBAD catalog. We identify non-stellar objects that are not previously classified, including extragalactic objects such as Quasi-Stellar Objects, Active Galactic Nuclei, supernovae and planetary nebulae. We then create a new labelled training set with 13 classes in two levels. We obtain a reasonable level of completeness (> 90 %) for certain classes of PVSs, although we have poorer completeness in other classes (~ 40 % in some cases). Our new labels for the ZTF CPVS are available via Zenodo Cheung et al. (2021).

Show Abstract

Rapid build-up of the stellar content in the protocluster core SPT2349−56 at z = 4.3

Ryley Hill, Scott Chapman, Kedar A. Phadke..., C. Hayward, Y. Hezaveh, et. al.

The protocluster SPT2349−56 at z=4.3 contains one of the most actively star-forming cores known, yet constraints on the total stellar mass of this system are highly uncertain. We have therefore carried out deep optical and infrared observations of this system, probing rest-frame ultraviolet to infrared wavelengths. Using the positions of the spectroscopically-confirmed protocluster members, we identify counterparts and perform detailed source deblending, allowing us to fit spectral energy distributions in order to estimate stellar masses. We show that the galaxies in SPT2349−56 have stellar masses proportional to their high star-formation rates, consistent with other protocluster galaxies and field submillimetre galaxies (SMGs) around redshift 4. The galaxies in SPT2349−56 have on average lower molecular gas-to-stellar mass fractions and depletion timescales than field SMGs, although with considerable scatter. We construct the stellar-mass function for SPT2349−56 and compare it to the stellar-mass function of z=1 galaxy clusters, finding consistent shapes between the two. We measure rest-frame galaxy ultraviolet half-light radii from our HST-F160W imaging, finding that on average the galaxies in our sample are similar in size to typical star-forming galaxies at these redshifts. However, the brightest HST-detected galaxy in our sample, found near the luminosity-weighted centre of the protocluster core, remains unresolved at this wavelength. Hydrodynamical simulations predict that the core galaxies will quickly merge into a brightest cluster galaxy, thus our observations provide a direct view of the early formation mechanisms of this class of object.

Show Abstract

The GALAH Survey: chemical tagging and chrono-chemodynamics of accreted halo stars with GALAH+ DR3 and Gaia eDR3

S. Buder, K. Lind, M. Ness, D. K. Feuillet, D. Horta Darrington, S. Monty, T. Buck, T. Nordlander, J. Bland-Hawthorn, A. R. Casey, G. M. De Silva, V. D'Orazi, K. C. Freeman, M. R. Hayden, J. Kos, S. L. Martell, G. F. Lewis, J. Lin, K. J. Schlesinger, S. Sharma, J. D. Simpson, D. Stello, D. B. Zucker, T. Zwitter, I. Ciucă, J. Horner, C. Kobayashi, Y-S. Ting, R. F. G. Wyse

Since the advent of Gaia astrometry, it is possible to identify massive accreted systems within the Galaxy through their unique dynamical signatures. One such system, Gaia-Sausage-Enceladus (GSE), appears to be an early ‘building block’ given its virial mass >1010M⊙ at infall (z ∼ 1−3). In order to separate the progenitor population from the background stars, we investigate its chemical properties with up to 30 element abundances from the GALAH+ Survey Data Release 3 (DR3). To inform our choice of elements for purely chemically selecting accreted stars, we analyse 4164 stars with low-α abundances and halo kinematics. These are most different to the Milky Way stars for abundances of Mg, Si, Na, Al, Mn, Fe, Ni, and Cu. Based on the significance of abundance differences and detection rates, we apply Gaussian mixture models to various element abundance combinations. We find the most populated and least contaminated component, which we confirm to represent GSE, contains 1049 stars selected via [Na/Fe] versus [Mg/Mn] in GALAH+ DR3. We provide tables of our selections and report the chrono-chemodynamical properties (age, chemistry, and dynamics). Through a previously reported clean dynamical selection of GSE stars, including 30

Show Abstract

Neural optimal feedback control with local learning rules

D. Chklovskii, S. Golkar, S. Farashahi, J. Friedrich, A. Genkin, A. Sengupta

A major problem in motor control is understanding how the brain plans and executes proper movements in the face of delayed and noisy stimuli. A prominent framework for addressing such control problems is Optimal Feedback Control (OFC). OFC generates control actions that optimize behaviorally relevant criteria by integrating noisy sensory stimuli and the predictions of an internal model using the Kalman filter or its extensions. However, a satisfactory neural model of Kalman filtering and control is lacking because existing proposals have the following limitations: not considering the delay of sensory feedback, training in alternating phases, and requiring knowledge of the noise covariance matrices, as well as that of systems dynamics. Moreover, the majority of these studies considered Kalman filtering in isolation, and not jointly with control. To address these shortcomings, we introduce a novel online algorithm which combines adaptive Kalman filtering with a model free control approach (i.e., policy gradient algorithm). We implement this algorithm in a biologically plausible neural network with local synaptic plasticity rules. This network performs system identification and Kalman filtering, without the need for multiple phases with distinct update rules or the knowledge of the noise covariances. It can perform state estimation with delayed sensory feedback, with the help of an internal model. It learns the control policy without requiring any knowledge of the dynamics, thus avoiding the need for weight transport. In this way, our implementation of OFC solves the credit assignment problem needed to produce the appropriate sensory-motor control in the presence of stimulus delay.

Show Abstract

Bridging the Gap: Point Clouds for Merging Neurons in Connectomics

J. Berman, D. Chklovskii, J. Wu

In the field of Connectomics, a primary problem is that of 3D neuron segmentation. Although deep learning-based methods have achieved remarkable accuracy, errors still exist, especially in regions with image defects. One common type of defect is that of consecutive missing image sections. Here, data is lost along some axis, and the resulting neuron segmentations are split across the gap. To address this problem, we propose a novel method based on point cloud representations of neurons. We formulate the problem as a classification problem and train CurveNet, a state-of-the-art point cloud classification model, to identify which neurons should be merged. We show that our method not only performs strongly but also scales reasonably to gaps well beyond what other methods have attempted to address. Additionally, our point cloud representations are highly efficient in terms of data, maintaining high performance with an amount of data that would be unfeasible for other methods. We believe that this is an indicator of the viability of using point cloud representations for other proofreading tasks.

Show Abstract
  • Previous Page
  • Viewing
  • Next Page
Advancing Research in Basic Science and MathematicsSubscribe to Flatiron Institute announcements and other foundation updates