Search

In the future, powerful AI systems may be deployed in high-stakes settings, where a single failure could be catastrophic. One technique for improving AI safety in high-stakes settings is adversarial training, which uses an adversary to generate examples to train on in order to achieve better worst-case performance. In this work, we used a safe language generation task (``avoid injuries'') as a testbed for achieving high reliability through adversarial training. We created a series of adversarial training techniques -- including a tool that assists human adversaries -- to find and eliminate failures in a classifier that filters text completions suggested by a generator. In our task, we determined that we can set very conservative classifier thresholds without significantly impacting the quality of the filtered outputs. We found that adversarial training increased robustness to the adversarial attacks that we trained on -- doubling the time for our contractors to find adversarial examples both with our tool (from 13 to 26 minutes) and without (from 20 to 44 minutes) -- without affecting in-distribution performance. We hope to see further work in the high-stakes reliability setting, including more powerful tools for enhancing human adversaries and better ways to measure high levels of reliability, until we can confidently rule out the possibility of catastrophic deployment-time failures of powerful models., Comment: 30 pages, 7 figures, NeurIPS camera-ready

This paper studies the trade-off between two different kinds of pure exploration: breadth versus depth. The most biased coin problem asks how many total coin flips are required to identify a "heavy" coin from an infinite bag containing both "heavy" coins with mean $\theta_1 \in (0,1)$, and "light" coins with mean $\theta_0 \in (0,\theta_1)$, where heavy coins are drawn from the bag with probability $\alpha \in (0,1/2)$. The key difficulty of this problem lies in distinguishing whether the two kinds of coins have very similar means, or whether heavy coins are just extremely rare. This problem has applications in crowdsourcing, anomaly detection, and radio spectrum search. Chandrasekaran et. al. (2014) recently introduced a solution to this problem but it required perfect knowledge of $\theta_0,\theta_1,\alpha$. In contrast, we derive algorithms that are adaptive to partial or absent knowledge of the problem parameters. Moreover, our techniques generalize beyond coins to more general instances of infinitely many armed bandit problems. We also prove lower bounds that show our algorithm's upper bounds are tight up to $\log$ factors, and on the way characterize the sample complexity of differentiating between a single parametric distribution and a mixture of two such distributions. As a result, these bounds have surprising implications both for solutions to the most biased coin problem and for anomaly detection when only partial information about the parameters is known.

Data labeling is a necessary but often slow process that impedes the development of interactive systems for modern data analysis. Despite rising demand for manual data labeling, there is a surprising lack of work addressing its high and unpredictable latency. In this paper, we introduce CLAMShell, a system that speeds up crowds in order to achieve consistently low-latency data labeling. We offer a taxonomy of the sources of labeling latency and study several large crowd-sourced labeling deployments to understand their empirical latency profiles. Driven by these insights, we comprehensively tackle each source of latency, both by developing novel techniques such as straggler mitigation and pool maintenance and by optimizing existing methods such as crowd retainer pools and active learning. We evaluate CLAMShell in simulation and on live workers on Amazon's Mechanical Turk, demonstrating that our techniques can provide an order of magnitude speedup and variance reduction over existing crowdsourced labeling strategies.

The joint JAXA/NASA ASTRO-H mission is the sixth in a series of highly successful X-ray missions developed by the Institute of Space and Astronautical Science (ISAS), with a planned launch in 2015. The ASTRO-H mission is equipped with a suite of sensitive instruments with the highest energy resolution ever achieved at E > 3 keV and a wide energy range spanning four decades in energy from soft X-rays to gamma-rays. The simultaneous broad band pass, coupled with the high spectral resolution of Delta E < 7 eV of the micro-calorimeter, will enable a wide variety of important science themes to be pursued. ASTRO-H is expected to provide breakthrough results in scientific areas as diverse as the large-scale structure of the Universe and its evolution, the behavior of matter in the gravitational strong field regime, the physical conditions in sites of cosmic-ray acceleration, and the distribution of dark matter in galaxy clusters at different redshifts., Comment: 24 pages, 18 figures, Proceedings of the SPIE Astronomical Instrumentation "Space Telescopes and Instrumentation 2014: Ultraviolet to Gamma Ray"

Although the human visual system is surprisingly robust to extreme distortion when recognizing objects, most evaluations of computer object detection methods focus only on robustness to natural form deformations such as people's pose changes. To determine whether algorithms truly mirror the flexibility of human vision, they must be compared against human vision at its limits. For example, in Cubist abstract art, painted objects are distorted by object fragmentation and part-reorganization, to the point that human vision often fails to recognize them. In this paper, we evaluate existing object detection methods on these abstract renditions of objects, comparing human annotators to four state-of-the-art object detectors on a corpus of Picasso paintings. Our results demonstrate that while human perception significantly outperforms current methods, human perception and part-based models exhibit a similarly graceful degradation in object detection performance as the objects become increasingly abstract and fragmented, corroborating the theory of part-based object representation in the brain.

The joint JAXA/NASA ASTRO-H mission is the sixth in a series of highly successful X-ray missions initiated by the Institute of Space and Astronautical Science (ISAS). ASTRO-H will investigate the physics of the high-energy universe via a suite of four instruments, covering a very wide energy range, from 0.3 keV to 600 keV. These instruments include a high-resolution, high-throughput spectrometer sensitive over 0.3-2 keV with high spectral resolution of Delta E < 7 eV, enabled by a micro-calorimeter array located in the focal plane of thin-foil X-ray optics; hard X-ray imaging spectrometers covering 5-80 keV, located in the focal plane of multilayer-coated, focusing hard X-ray mirrors; a wide-field imaging spectrometer sensitive over 0.4-12 keV, with an X-ray CCD camera in the focal plane of a soft X-ray telescope; and a non-focusing Compton-camera type soft gamma-ray detector, sensitive in the 40-600 keV band. The simultaneous broad bandpass, coupled with high spectral resolution, will enable the pursuit of a wide variety of important science themes., Comment: 22 pages, 17 figures, Proceedings of the SPIE Astronomical Instrumentation "Space Telescopes and Instrumentation 2012: Ultraviolet to Gamma Ray"

To fulfill legislative reporting requirements, this report describes accomplishments of federal agencies with archeological programs, as well as the impact of federal projects on the nation's archeological heritage. In 1991, the Secretary of the Interior outlined actions that agencies should take in (1) preserving and researching sites, (2) preventing looting and vandalism, (3) educating the public, and (4) conserving collections and records. This report examines progress in each of these areas and highlights cooperative efforts with American Indians, Alaska Natives, and Native Hawaiians. These efforts have included involvement of Native groups in archeological research; training and technical assistance to Native groups who wish to do their own archeology; contributions by elders to understanding the cultural context of sites; workshops for Native American educators using archeology to build K-12 curricula; and mentoring of Native youth in preservation. Sections on each of the four areas of concern include a short overview of the situation, followed by program profiles and highlights. The section on reaching out to the public describes an Alaska Native initiative to interpret cultural sites to the public, development of a kit and lesson plans on Chinook culture for use in Washington schools, a virtual school "field trip" to American Indian sites in Virginia, and other school and college activities. Contains many photographs and a list of federal agencies that conduct, sponsor, or license archeology. (SV)

The measurements of hadronic structure functions of the photon based on the reaction ee --> ee gamma^(*)(P^2) gamma^*(Q^2)--> ee hadrons are discussed., Comment: 9 pages, 6 figures, Conference contribution to DIS 2000, 25-30 April 2000, Liverpool, England, to appear in the Proceedings