Publications

Many "big data" applications need to act on data arriving in real time.However, current programming models for distributed stream processing are relatively low-level, often leaving the user to worry about consistency of state across the system and fault recovery.Furthermore, the models that provide fault recovery do so in an expensive manner, requiring either hot replication or long recovery times.We propose a new programming model, discretized streams (D-Streams), that offers a high-level functional API, strong consistency, and efficient fault recovery.D-Streams support a new recovery mechanism that improves efficiency over the traditional replication and upstream backup schemes in streaming databasesparallel recovery of lost state-and unlike previous systems, also mitigate stragglers.We implement D-Streams as an extension to the Spark cluster computing engine that lets users seamlessly intermix streaming, batch and interactive queries.Our system can process over 60 million records/second at sub-second latency on 100 nodes.

Density functional theory (DFT) calculations are performed to investigate the energetics of the CO2 electrochemical reduction on metal (M) porphyrin-like motifs incorporated into graphene layers. The objective is to develop strategies that enhance CO2 reduction while suppressing the competitive hydrogen evolution reaction (HER). We find that there exists a scaling relation between the binding energy of the catalyst to hydrogen and that to COOH, a key intermediate in the reduction of CO2 to CO; however, the M-H bond is stronger than the M-COOH bond, driving the reaction toward the HER rather than the reduction of CO2 to CO. This scaling relation holds even with axial ligation to the metal cation coordinated to the porphyrin ring. When 4f lanthanide or 5f actinide elements are used as the reactive center, the scaling relation still holds but the M-COOH bond is stronger than the M-H bond, and the reaction favors the reduction of CO2 to CO. By contrast, there is no scaling relation between the binding energy of the catalyst to H and that to OCHO, the key intermediate in CO2 reduction to formic acid. Interestingly, we find that coordination of a ligand to an unoccupied axial site can make the M-OCHO bond stronger than the M-H bond, resulting in preferential formic acid formation. This means that the axial ligand effectively enhances CO2 reduction to formic acid and suppresses the HER. Our DFT calculations have also identified several promising electrocatalysts for CO2 reduction to HCOOH with almost zero overpotentials.

<p>PDF file - 53K</p>

Poly(N-vinylpyrrolidone) (PVP) capped platinum and rhodium nanoparticles (7-12 nm) have been studied with UV-VIS, FTIR and Raman spectroscopy. The absorption bands in the region 190-900 nm are shown to be sensitive to the electronic structure of surface Rh and Pt atoms as well as to the aggregation of the nanoparticles. In-situ FTIR-DRIFT spectroscopy of the thermal decay of PVP stabilized Rh and Pt nanoparticles in H{sub 2} and O{sub 2} atmospheres in temperatures ranging from 30 C-350 C reveal that decomposition of PVP above 200 C, PVP transforms into a 'polyamidpolyene' - like material that is in turn converted into a thin layer of amorphous carbon above 300 C. Adsorbed carbon monoxide was used as a probing molecule to monitor changes of electronic structure of surface Rh and Pt atoms and accessible surface area. The behavior of surface Rh and Pt atoms with ligated CO and amide groups of pyrrolidones resemble that of surface coordination compounds.

The design and development of metal-cluster-based heterogeneous catalysts with high activity, selectivity, and stability under solution-phase reaction conditions will enable their applications as recyclable catalysts in large-scale fine chemicals production. To achieve these required catalytic properties, a heterogeneous catalyst must contain specific catalytically active species in high concentration, and the active species must be stabilized on a solid catalyst support under solution-phase reaction conditions. These requirements pose a great challenge for catalysis research to design metal-cluster-based catalysts for solution-phase catalytic processes. Here, we focus on a silica-supported, polymer-encapsulated Pt catalyst for an electrophilic hydroalkoxylation reaction in toluene, which exhibits superior selectivity and stability against leaching under mild reaction conditions. We unveil the key factors leading to the observed superior catalytic performance by combining X-ray absorption spectroscopy (XAS) and reaction kinetic studies. On the basis of the mechanistic understandings obtained in this work, we also provide useful guidelines for designing metal-cluster-based catalyst for a broader range of reactions in the solution phase.

Human trajectory forecasting is an inherently multi-modal problem. Uncertainty in future trajectories stems from two sources: (a) sources that are known to the agent but unknown to the model, such as long term goals and (b)sources that are unknown to both the agent & the model, such as intent of other agents & irreducible randomness indecisions. We propose to factorize this uncertainty into its epistemic & aleatoric sources. We model the epistemic un-certainty through multimodality in long term goals and the aleatoric uncertainty through multimodality in waypoints& paths. To exemplify this dichotomy, we also propose a novel long term trajectory forecasting setting, with prediction horizons upto a minute, an order of magnitude longer than prior works. Finally, we presentY-net, a scene com-pliant trajectory forecasting network that exploits the pro-posed epistemic & aleatoric structure for diverse trajectory predictions across long prediction horizons.Y-net significantly improves previous state-of-the-art performance on both (a) The well studied short prediction horizon settings on the Stanford Drone & ETH/UCY datasets and (b) The proposed long prediction horizon setting on the re-purposed Stanford Drone & Intersection Drone datasets.

We present Doppler and modulation tomography of the X-ray nova XTE J1118+480\nwith data obtained during quiescence using the 10-m Keck II telescope. The hot\nspot where the gas stream hits the accretion disc is seen in H-Alpha, H-Beta,\nHe I Lambda-5876, and Ca II Lambda-8662, thus verifying the presence of\ncontinued mass transfer within the system. The disc is clearly seen in H-Alpha\nand Ca II Lambda-8662. We image the mass-donor star in narrow absorption lines\nof Na I Lambda-Lambda-5890, 5896, 8183, 8195 and Ca II Lambda-8662, implying an\norigin from the secondary itself rather than the interstellar medium. We also\ndetect deviations in the centroid of the double peak of H-Alpha akin to those\nfound by Zurita et al. 2002 suggesting disc eccentricity.\n

The solid-state portion of the W-Mo-Cr phase diagram is computed without the use of empirically adjustable parameters. The phase diagram is calculated using the bcc tetrahedron approximation of the cluster variation method (CVM) formulated with an independent set of multisite correlation functions. A set of volume-independent effective cluster interactions (ECI's) are used to parametrize the configurational energetics of the system; they are obtained using the method of direct configurational averaging based on a tight-binding, linearized muffin-tin-orbital Hamiltonian. The nearest- and next-nearest-neighbor pair ECI's are found to be dominant and indicate clustering tendencies for all binary systems (W-Mo, W-Cr, and Mo-Cr), consistent with experiment; triplet and higher-order pair interactions are significantly smaller. The ternary phase diagram for W-Mo-Cr is analyzed using twelve ECI's within the bcc tetrahedron, and while there is no experimental data for this ternary system, the results are quite encouraging even though transition temperatures on the binary edges are overestimated. The latter discrepancy is most likely due to the neglect of elastic relaxations driven by the large size mismatch of the constituent elements or to the neglect of vibrational contributions to the free energy.