Publications

An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures.

The safety of autonomous vehicles (AVs) depends on their ability to perform complex computations on high-volume sensor data in a timely manner. Their ability to run these computations with state-of-the-art models is limited by the processing power and slow update cycles of their onboard hardware. In contrast, cloud computing offers the ability to burst computation to vast amounts of the latest generation of hardware. However, accessing these cloud resources requires traversing wireless networks that are often considered to be too unreliable for real-time AV driving applications. Our work seeks to harness this unreliable cloud to enhance the accuracy of an AV's decisions, while ensuring that it can always fall back to its on-board computational capabilities. We identify three mechanisms that can be used by AVs to safely leverage the cloud for accuracy enhancements, and elaborate why current execution systems fail to enable these mechanisms. To address these limitations, we provide a system design based on the speculative execution of an AV's pipeline in the cloud, and show the efficacy of this approach in simulations of complex real-world scenarios that apply these mechanisms.

Nebular-phase spectra of SN 2006aj, which was discovered in coincidence with X-ray flash 060218, were obtained with Keck in 2006 July and the Very Large Telescope in 2006 September. At the latter epoch spectropolarimetry was also attempted, yielding an upper limit of ~ 2% for the polarization. The spectra show strong emission lines of [OI] and MgI], as expected from a Type Ic supernova, but weak CaII lines. The [FeII] lines that were strong in the spectra of SN 1998bw are much weaker in SN 2006aj, consistent with the lower luminosity of this SN. The outer velocity of the line-emitting ejecta is ~ 8000 km/s in July and ~ 7400 km/s in September, consistent with the relatively low kinetic energy of expansion of SN 2006aj. All emission lines have similar width, and the profiles are symmetric, indicating that no major asymmetries are present in the ejecta at the velocities sampled by the nebular lines (v < 8000 km/s), except perhaps in the innermost part. The spectra were modelled with a non-LTE code. The mass of 56Ni required to power the emission spectrum is ~ 0.20 Msun, in excellent agreement with the results of early light curve modelling. The oxygen mass is ~ 1.5 Msun, again much less than in SN 1998bw but larger by ~ 0.7 Msun than the value derived from the early-time modelling. The total ejected mass is ~ 2 Msun below 8000 km/s. This confirms that SN 2006aj was only slightly more massive and energetic than the prototypical Type Ic SN 1994I, but also indicates the presence of a dense inner core, containing ~ 1 Msun of mostly oxygen and carbon. The presence of such a core is inferred for all broad-lined SNe Ic. This core may have the form of an equatorial oxygen-dominated region, but it is too deep to affect the early light curve and too small to affect the late polarization spectrum.

Synthesis prediction is a key accelerator for the rapid design of advanced materials. However, determining synthesis variables such as the choice of precursor materials is challenging for inorganic materials because the sequence of reactions during heating is not well understood. In this work, we use a knowledge base of 29,900 solid-state synthesis recipes, text-mined from the scientific literature, to automatically learn which precursors to recommend for the synthesis of a novel target material. The data-driven approach learns chemical similarity of materials and refers the synthesis of a new target to precedent synthesis procedures of similar materials, mimicking human synthesis design. When proposing five precursor sets for each of 2,654 unseen test target materials, the recommendation strategy achieves a success rate of at least 82%. Our approach captures decades of heuristic synthesis data in a mathematical form, making it accessible for use in recommendation engines and autonomous laboratories.

Summary Contemporary models conceptualize spatial attention as a blinking spotlight that sequentially samples visual space. Hence, behavior fluctuates over time even in states of presumed ‘sustained’ attention. Recent evidence suggested that rhythmic neural activity in the frontoparietal network constitutes the functional basis of rhythmic attentional sampling. However, causal evidence to support this notion remains absent. Using a lateralized spatial attention task, we addressed this issue in patients with focal lesions in the frontoparietal attention network. Our results uncovered that frontoparietal lesions introduce periodic neglect, i.e., temporally-specific behavioral deficits that were aligned with the underlying neural oscillations. Attention-guided perceptual sensitivity was on par with healthy controls during optimal phases but attenuated during the less excitable sub-cycles. Theta-dependent sampling (3 – 8 Hz) was causally dependent on prefrontal cortex, while alpha-band sampling (8 – 14 Hz) emerged from parietal areas. Collectively, our findings reveal that lesion-induced high amplitude, low frequency brain activity is not epiphenomenal, but has immediate behavioral consequences. More generally, these results provide causal evidence for the hypothesis that the functional architecture of attention is inherently rhythmic.

Overfishing and environmental change have triggered many severe and unexpected consequences. As existing communities have collapsed, new ones have become established, fundamentally transforming ecosystems to those that are often less productive for fisheries, more prone to cycles of booms and busts, and thus less manageable. We contend that the failure of fisheries science and management to anticipate these transformations results from a lack of appreciation for the nature, strength, complexity, and outcome of species interactions. Ecologists have come to understand that networks of interacting species exhibit nonlinear dynamics and feedback loops that can produce sudden and unexpected shifts. We argue that fisheries science and management must follow this lead by developing a sharper focus on species interactions and how disrupting these interactions can push ecosystems in which fisheries are embedded past their tipping points.

We show that orbit-superposition dynamical models (Schwarzschild's method) provide reliable estimates of nuclear black hole masses and errors when constructed from adequate orbit libraries and kinematic data. We thus rebut two recent papers that argue that BH masses obtained from this method are unreliable. These papers claim to demonstrate that the range of allowable BH masses derived from a given dataset is artificially too narrow as a result of an inadequate number of orbits in the library used to construct dynamical models. This is an elementary error that is easily avoided. We describe a method to estimate the number and nature of orbits needed for the library. We provide an example that shows that this prescription is adequate, in the sense that the range of allowable BH masses is not artificially narrowed by use of too few orbits. A second point raised by critics is that kinematic data are generally obtained with insufficient spatial resolution to obtain a reliable mass. We make the distinction between unreliable determinations and imprecise ones. We show that there are several different properties of a kinematic dataset that can lead to imprecise BH determinations, but none of the attributes we have investigated leads to an unreliable determination. In short, the degree to which the BH radius of influence is resolved by spectroscopic observations is already reflected in the BH-mass error envelope, and is not a hidden source of error. The BH masses published by our group and the Leiden group are reliable.

An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures.

The Mixture-of-Experts (MoE) architecture has become increasingly popular as a method to scale up large language models (LLMs). To save costs, heterogeneity-aware training solutions have been proposed to utilize GPU clusters made up of both newer and older-generation GPUs. However, existing solutions are agnostic to the performance characteristics of different MoE model components (i.e., attention and expert) and do not fully utilize each GPU's compute capability. In this paper, we introduce HeterMoE, a system to efficiently train MoE models on heterogeneous GPUs. Our key insight is that newer GPUs significantly outperform older generations on attention due to architectural advancements, while older GPUs are still relatively efficient for experts. HeterMoE disaggregates attention and expert computation, where older GPUs are only assigned with expert modules. Through the proposed zebra parallelism, HeterMoE overlaps the computation on different GPUs, in addition to employing an asymmetric expert assignment strategy for fine-grained load balancing to minimize GPU idle time. Our evaluation shows that HeterMoE achieves up to 2.3x speed-up compared to existing MoE training systems, and 1.4x compared to an optimally balanced heterogeneity-aware solution. HeterMoE efficiently utilizes older GPUs by maintaining 95% training throughput on average, even with half of the GPUs in a homogeneous A40 cluster replaced with V100.