Publications

Automatic video captioning aims to train models to generate text descriptions\nfor all segments in a video, however, the most effective approaches require\nlarge amounts of manual annotation which is slow and expensive. Active learning\nis a promising way to efficiently build a training set for video captioning\ntasks while reducing the need to manually label uninformative examples. In this\nwork we both explore various active learning approaches for automatic video\ncaptioning and show that a cluster-regularized ensemble strategy provides the\nbest active learning approach to efficiently gather training sets for video\ncaptioning. We evaluate our approaches on the MSR-VTT and LSMDC datasets using\nboth transformer and LSTM based captioning models and show that our novel\nstrategy can achieve high performance while using up to 60% fewer training data\nthan the strong state of the art baselines.\n

We have discovered 16 Type Ia supernovae (SNe Ia) with the Hubble Space Telescope (HST) and have used them to provide the first conclusive evidence for cosmic deceleration that preceded the current epoch of cosmic acceleration. These objects, discovered during the course of the GOODS ACS Treasury program, include 6 of the 7 highest-redshift SNe Ia known, all at z>1.25, and populate the Hubble diagram in unexplored territory. The luminosity distances to these and 170 previous SNe Ia are provided. A purely kinematic interpretation of the SN Ia sample provides evidence at the > 99% confidence level for a transition from deceleration to acceleration or similarly, strong evidence for a cosmic jerk. Using a simple model of the expansion history, the transition between the two epochs is constrained to be at z=0.46 +/- 0.13. The data are consistent with the cosmic concordance model of Omega_M ~ 0.3, Omega_Lambda~0.7 (chi^2_dof=1.06), and are inconsistent with a simple model of evolution or dust as an alternative to dark energy. For a flat Universe with a cosmological constant. When combined with external flat-Universe constraints we find w=-1.02 + 0.13 - 0.19 (and $<-0.76 at the 95% confidence level) for an assumed static equation of state of dark energy, P = w\rho c^2. Joint constraints on both the recent equation of state of dark energy, $w_0$, and its time evolution, dw/dz, are a factor of ~8 more precise than its first estimate and twice as precise as those without the SNe Ia discovered with HST. Our constraints are consistent with the static nature of and value of w expected for a cosmological constant (i.e., w_0 = -1.0, dw/dz = 0), and are inconsistent with very rapid evolution of dark energy. We address consequences of evolving dark energy for the fate of the Universe.

Memristors were first proposed in 1971 by Leon Chua. These devices are usually regarded as being one of the newest fundamental breakthrough for electronics. Their role in designing new electronic systems is expected to be an important, key-factor. As an example, they already come in many forms: PCA, ReRAM, etc., to mention a few. In any case, since actual memristors have only appeared quite recently, this technology has yet to be mature enough to provide with readily available, off-the-shelf components. This implies that developing and testing new concepts or design architectures based on memristors are performed mainly by the use of numerical simulation. In this paper, we discuss a powerful modeling framework that eases creating and implementing new memristor models, illustrating with some examples of use.

Abstract Cation‐disordered rocksalt (DRX) materials have emerged as a class of novel high‐capacity cathodes for Li‐ion batteries. However, the commercialization of DRX cathodes will require reducing their capacity decay, which has been associated with oxygen loss during cycling. Recent studies show that fluorination of DRX cathodes can effectively reduce oxygen loss and improve cycling stability; however, the underlying atomic‐scale mechanisms remain elusive. Herein, using a combination of electrochemical measurements, scanning transmission electron microscopy, and electron energy loss spectroscopy, the correlation between the electrochemical properties and structural evolution in Mn‐redox‐based DRX cathodes, Li 1.2 Ti 0.4– x Mn 0.4+ x O 2.0‐ x F x ( x = 0 and 0.2) is examined. It is found that fluorination strongly suppresses structural amorphization and void formation initiated from the particle surface, therefore greatly enhancing the cyclability of the cathode. A novel rocksalt‐to‐spinel‐like structural transformation in the DRX bulk is further revealed, which surprisingly contributes to a gradual capacity increase during cycling. The results provide important insight for the design of novel DRX cathodes with high capacity and long cycle life.

While Pseudomonas has the capability for high metabolic activity and the provision of reduced redox cofactors important for biocatalytic applications, exploitation of this characteristic might be hindered by high, constitutive activity of and, consequently, competition with the NADH dehydrogenases of the respiratory chain. The in-depth analysis of NADH dehydrogenase mutants of Pseudomonas taiwanensis VLB120 presented here provides insight into the phenotypic and metabolic response of this strain to these redox metabolism perturbations. This high degree of metabolic flexibility needs to be taken into account for rational engineering of this promising biotechnological workhorse toward a host with a controlled and efficient supply of redox cofactors for product synthesis.

Modern distributed applications run across numerous microservices and components deployed in cloud datacenters, using shared cloud services for computing and storage, edge services such as content distribution networks, network functions such as rate limiters and firewalls, security infrastructures, network routers, and physical links. When a user-visible fault occurs, the first step toward diagnosis is localization to determine where the fault has occurred. However, because application delivery spans different layers and different organizations, no entity has complete visibility or access to the information required to localize faults quickly. This paper proposes a cross-layer, cross-domain, and cross-application fault localization primitive with a simple and standardized information interface for the Internet.