-
Blending Data-Driven Priors in Dynamic Games
Authors:
Justin Lidard,
Haimin Hu,
Asher Hancock,
Zixu Zhang,
Albert Gimó Contreras,
Vikash Modi,
Jonathan DeCastro,
Deepak Gopinath,
Guy Rosman,
Naomi Leonard,
María Santos,
Jaime Fernández Fisac
Abstract:
As intelligent robots like autonomous vehicles become increasingly deployed in the presence of people, the extent to which these systems should leverage model-based game-theoretic planners versus data-driven policies for safe, interaction-aware motion planning remains an open question. Existing dynamic game formulations assume all agents are task-driven and behave optimally. However, in reality, h…
▽ More
As intelligent robots like autonomous vehicles become increasingly deployed in the presence of people, the extent to which these systems should leverage model-based game-theoretic planners versus data-driven policies for safe, interaction-aware motion planning remains an open question. Existing dynamic game formulations assume all agents are task-driven and behave optimally. However, in reality, humans tend to deviate from the decisions prescribed by these models, and their behavior is better approximated under a noisy-rational paradigm. In this work, we investigate a principled methodology to blend a data-driven reference policy with an optimization-based game-theoretic policy. We formulate KLGame, a type of non-cooperative dynamic game with Kullback-Leibler (KL) regularization with respect to a general, stochastic, and possibly multi-modal reference policy. Our method incorporates, for each decision maker, a tunable parameter that permits modulation between task-driven and data-driven behaviors. We propose an efficient algorithm for computing multimodal approximate feedback Nash equilibrium strategies of KLGame in real time. Through a series of simulated and real-world autonomous driving scenarios, we demonstrate that KLGame policies can more effectively incorporate guidance from the reference policy and account for noisily-rational human behaviors versus non-regularized baselines.
△ Less
Submitted 23 February, 2024; v1 submitted 21 February, 2024;
originally announced February 2024.
-
PAC-Bayes Generalization Certificates for Learned Inductive Conformal Prediction
Authors:
Apoorva Sharma,
Sushant Veer,
Asher Hancock,
Heng Yang,
Marco Pavone,
Anirudha Majumdar
Abstract:
Inductive Conformal Prediction (ICP) provides a practical and effective approach for equip** deep learning models with uncertainty estimates in the form of set-valued predictions which are guaranteed to contain the ground truth with high probability. Despite the appeal of this coverage guarantee, these sets may not be efficient: the size and contents of the prediction sets are not directly contr…
▽ More
Inductive Conformal Prediction (ICP) provides a practical and effective approach for equip** deep learning models with uncertainty estimates in the form of set-valued predictions which are guaranteed to contain the ground truth with high probability. Despite the appeal of this coverage guarantee, these sets may not be efficient: the size and contents of the prediction sets are not directly controlled, and instead depend on the underlying model and choice of score function. To remedy this, recent work has proposed learning model and score function parameters using data to directly optimize the efficiency of the ICP prediction sets. While appealing, the generalization theory for such an approach is lacking: direct optimization of empirical efficiency may yield prediction sets that are either no longer efficient on test data, or no longer obtain the required coverage on test data. In this work, we use PAC-Bayes theory to obtain generalization bounds on both the coverage and the efficiency of set-valued predictors which can be directly optimized to maximize efficiency while satisfying a desired test coverage. In contrast to prior work, our framework allows us to utilize the entire calibration dataset to learn the parameters of the model and score function, instead of requiring a separate hold-out set for obtaining test-time coverage guarantees. We leverage these theoretical results to provide a practical algorithm for using calibration data to simultaneously fine-tune the parameters of a model and score function while guaranteeing test-time coverage and efficiency of the resulting prediction sets. We evaluate the approach on regression and classification tasks, and outperform baselines calibrated using a Hoeffding bound-based PAC guarantee on ICP, especially in the low-data regime.
△ Less
Submitted 7 December, 2023;
originally announced December 2023.
-
The KELT Follow-Up Network and Transit False Positive Catalog: Pre-vetted False Positives for TESS
Authors:
Karen A. Collins,
Kevin I. Collins,
Joshua Pepper,
Jonathan Labadie-Bartz,
Keivan Stassun,
B. Scott Gaudi,
Daniel Bayliss,
Joao Bento,
Knicole D. Colón,
Dax Feliz,
David James,
Marshall C. Johnson,
Rudolf B. Kuhn,
Michael B. Lund,
Matthew T. Penny,
Joseph E. Rodriguez,
Robert J. Siverd,
Daniel J. Stevens,
Xinyu Yao,
George Zhou,
Mundra Akshay,
Giulio F. Aldi,
Cliff Ashcraft,
Supachai Awiphan,
Özgür Baştürk
, et al. (86 additional authors not shown)
Abstract:
The Kilodegree Extremely Little Telescope (KELT) project has been conducting a photometric survey for transiting planets orbiting bright stars for over ten years. The KELT images have a pixel scale of ~23"/pixel---very similar to that of NASA's Transiting Exoplanet Survey Satellite (TESS)---as well as a large point spread function, and the KELT reduction pipeline uses a weighted photometric apertu…
▽ More
The Kilodegree Extremely Little Telescope (KELT) project has been conducting a photometric survey for transiting planets orbiting bright stars for over ten years. The KELT images have a pixel scale of ~23"/pixel---very similar to that of NASA's Transiting Exoplanet Survey Satellite (TESS)---as well as a large point spread function, and the KELT reduction pipeline uses a weighted photometric aperture with radius 3'. At this angular scale, multiple stars are typically blended in the photometric apertures. In order to identify false positives and confirm transiting exoplanets, we have assembled a follow-up network (KELT-FUN) to conduct imaging with higher spatial resolution, cadence, and photometric precision than the KELT telescopes, as well as spectroscopic observations of the candidate host stars. The KELT-FUN team has followed-up over 1,600 planet candidates since 2011, resulting in more than 20 planet discoveries. Excluding ~450 false alarms of non-astrophysical origin (i.e., instrumental noise or systematics), we present an all-sky catalog of the 1,128 bright stars (6<V<10) that show transit-like features in the KELT light curves, but which were subsequently determined to be astrophysical false positives (FPs) after photometric and/or spectroscopic follow-up observations. The KELT-FUN team continues to pursue KELT and other planet candidates and will eventually follow up certain classes of TESS candidates. The KELT FP catalog will help minimize the duplication of follow-up observations by current and future transit surveys such as TESS.
△ Less
Submitted 19 September, 2018; v1 submitted 5 March, 2018;
originally announced March 2018.
-
KELT-21b: A Hot Jupiter Transiting the Rapidly-Rotating Metal-Poor Late-A Primary of a Likely Hierarchical Triple System
Authors:
Marshall C. Johnson,
Joseph E. Rodriguez,
George Zhou,
Erica J. Gonzales,
Phillip A. Cargile,
Justin R. Crepp,
Kaloyan Penev,
Keivan G. Stassun,
B. Scott Gaudi,
Knicole D. Colón,
Daniel J. Stevens,
Klaus G. Strassmeier,
Ilya Ilyin,
Karen A. Collins,
John F. Kielkopf,
Thomas E. Oberst,
Luke Maritch,
Phillip A. Reed,
Joao Gregorio,
Valerio Bozza,
Sebastiano Calchi Novati,
Giuseppe D'Ago,
Gaetano Scarpetta,
Roberto Zambelli,
David W. Latham
, et al. (43 additional authors not shown)
Abstract:
We present the discovery of KELT-21b, a hot Jupiter transiting the $V=10.5$ A8V star HD 332124. The planet has an orbital period of $P=3.6127647\pm0.0000033$ days and a radius of $1.586_{-0.040}^{+0.039}$ $R_J$. We set an upper limit on the planetary mass of $M_P<3.91$ $M_J$ at $3σ$ confidence. We confirmed the planetary nature of the transiting companion using this mass limit and Doppler tomograp…
▽ More
We present the discovery of KELT-21b, a hot Jupiter transiting the $V=10.5$ A8V star HD 332124. The planet has an orbital period of $P=3.6127647\pm0.0000033$ days and a radius of $1.586_{-0.040}^{+0.039}$ $R_J$. We set an upper limit on the planetary mass of $M_P<3.91$ $M_J$ at $3σ$ confidence. We confirmed the planetary nature of the transiting companion using this mass limit and Doppler tomographic observations to verify that the companion transits HD 332124. These data also demonstrate that the planetary orbit is well-aligned with the stellar spin, with a sky-projected spin-orbit misalignment of $λ=-5.6_{-1.9}^{+1.7 \circ}$. The star has $T_{\mathrm{eff}}=7598_{-84}^{+81}$ K, $M_*=1.458_{-0.028}^{+0.029}$ $M_{\odot}$, $R_*=1.638\pm0.034$ $R_{\odot}$, and $v\sin I_*=146$ km s$^{-1}$, the highest projected rotation velocity of any star known to host a transiting hot Jupiter. The star also appears to be somewhat metal-poor and $α$-enhanced, with [Fe/H]$=-0.405_{-0.033}^{+0.032}$ and [$α$/Fe]$=0.145 \pm 0.053$; these abundances are unusual, but not extraordinary, for a young star with thin-disk kinematics like KELT-21. High-resolution imaging observations revealed the presence of a pair of stellar companions to KELT-21, located at a separation of 1.2" and with a combined contrast of $ΔK_S=6.39 \pm 0.06$ with respect to the primary. Although these companions are most likely physically associated with KELT-21, we cannot confirm this with our current data. If associated, the candidate companions KELT-21 B and C would each have masses of $\sim0.12$ $M_{\odot}$, a projected mutual separation of $\sim20$ AU, and a projected separation of $\sim500$ AU from KELT-21. KELT-21b may be one of only a handful of known transiting planets in hierarchical triple stellar systems.
△ Less
Submitted 17 January, 2018; v1 submitted 8 December, 2017;
originally announced December 2017.
-
A straw drift chamber spectrometer for studies of rare kaon decays
Authors:
K. Lang,
D. Ambrose,
C. Arroyo,
M. Bachman,
D. Connor,
M. Eckhause,
K. M. Ecklund,
S. Graessle,
M. Hamela,
S. Hamilton,
A. D. Hancock,
K. Hartman,
M. Hebert,
C. H. Hoff,
G. W. Hoffmann,
G. M. Irwin,
J. R. Kane,
N. Kanematsu,
Y. Kuang,
R. Lee,
M. Marcin,
R. D. Martin,
J. McDonough,
A. Milder,
W. R. Molzon
, et al. (11 additional authors not shown)
Abstract:
We describe the design, construction, readout, tests, and performance of planar drift chambers, based on 5 mm diameter copperized Mylar and Kapton straws, used in an experimental search for rare kaon decays. The experiment took place in the high-intensity neutral beam at the Alternating Gradient Synchrotron of Brookhaven National Laboratory, using a neutral beam stop, two analyzing dipoles, and…
▽ More
We describe the design, construction, readout, tests, and performance of planar drift chambers, based on 5 mm diameter copperized Mylar and Kapton straws, used in an experimental search for rare kaon decays. The experiment took place in the high-intensity neutral beam at the Alternating Gradient Synchrotron of Brookhaven National Laboratory, using a neutral beam stop, two analyzing dipoles, and redundant particle identification to remove backgrounds.
△ Less
Submitted 12 August, 2003;
originally announced August 2003.
-
First Observation of the Rare Decay Mode K-long -> e+ e-
Authors:
BNL E871 Collaboration,
D. Ambrose,
C. Arroyo,
M. Bachman,
P. de Cecco,
D. Connor,
M. Eckhause,
K. M. Ecklund,
S. Graessle,
A. D. Hancock,
K. Hartman,
M. Hebert,
C. H. Hoff,
G. W. Hoffmann,
G. M. Irwin,
J. R. Kane,
N. Kanematsu,
Y. Kuang,
K. Lang,
R. Lee,
R. D. Martin,
J. McDonough,
A. Milder,
W. R. Molzon,
M. Pommot-Maia
, et al. (9 additional authors not shown)
Abstract:
In an experiment designed to search for and study very rare two-body decay modes of the K-long, we have observed four examples of the decay K-long -> e+ e-, where the expected background is 0.17+-0.10 events. This observation translates into a branching fraction of 8.7^{+5.7}_{-4.1} X 10^{-12}, consistent with recent theoretical predictions. This result represents by far the smallest branching f…
▽ More
In an experiment designed to search for and study very rare two-body decay modes of the K-long, we have observed four examples of the decay K-long -> e+ e-, where the expected background is 0.17+-0.10 events. This observation translates into a branching fraction of 8.7^{+5.7}_{-4.1} X 10^{-12}, consistent with recent theoretical predictions. This result represents by far the smallest branching fraction yet measured in particle physics.
△ Less
Submitted 1 October, 1998;
originally announced October 1998.
-
A Compact Beam Stop for a Rare Kaon Decay Experiment
Authors:
E871 Collaboration,
J. Belz,
M. Diwan,
M. Eckhause,
C. M. Guss,
A. D. Hancock,
A. P. Heinson,
V. L. Highland,
G. W. Hoffmann,
G. M. Irwin,
J. R. Kane,
S. H. Kettell,
Y. Kuang,
K. Lang,
J. McDonough,
W. K. McFarlane,
W. R. Molzon,
P. J. Riley,
J. L. Ritchie,
A. J. Schwartz,
B. Ware,
R. E. Welsh,
R. G. Winter,
M. Witkowski,
S. G. Wojcicki
, et al. (2 additional authors not shown)
Abstract:
We describe the development and testing of a novel beam stop for use in a rare kaon decay experiment at the Brookhaven AGS. The beam stop is located inside a dipole spectrometer magnet in close proximity to straw drift chambers and intercepts a high-intensity neutral hadron beam. The design process, involving both Monte Carlo simulations and beam tests of alternative beam-stop shielding arrangem…
▽ More
We describe the development and testing of a novel beam stop for use in a rare kaon decay experiment at the Brookhaven AGS. The beam stop is located inside a dipole spectrometer magnet in close proximity to straw drift chambers and intercepts a high-intensity neutral hadron beam. The design process, involving both Monte Carlo simulations and beam tests of alternative beam-stop shielding arrangements, had the goal of minimizing the leakage of particles from the beam stop and the resulting hit rates in detectors, while preserving maximum acceptance for events of interest. The beam tests consisted of measurements of rates in drift chambers, scintilation counter hodoscopes, a gas threshold Cherenkov counter, and a lead glass array. Measurements were also made with a set of specialized detectors which were sensitive to low-energy neutrons, photons, and charged particles. Comparisons are made between these measurements and a detailed Monte Carlo simulation.
△ Less
Submitted 31 August, 1998;
originally announced August 1998.
