Industrial recommender systems deal with extremely large action spaces -many millions of items to recommend. Moreover, they need to serve billions of users, who are unique at any point in time, making a complex user state space. Luckily, huge quantities of logged implicit feedback (e.g., user clicks, dwell time) are available for learning. Learning from the logged feedback is however subject to biases caused by only observing feedback on recommendations selected by the previous versions of the recommender. In this work, we present a general recipe of addressing such biases in a production top-K recommender system at YouTube, built with a policy-gradient-based algorithm, i.e. REINFORCE [48]. The contributions of the paper are: (1) scaling REINFORCE to a production recommender system with an action space on the orders of millions;(2) applying off-policy correction to address data biases in learning from logged feedback collected from multiple behavior policies; (3) proposing a novel top-K off-policy correction to account for our policy recommending multiple items at a time; (4) showcasing the value of exploration. We demonstrate the efficacy of our approaches through a series of simulations and multiple live experiments on YouTube.
B cells often constitute abundant cellular components in human tumors. Regulatory B cells that are functionally defi ned by their ability to produce IL10 downregulate infl ammation and control T-cell immunity. Here, we identifi ed a protumorigenic subset of B cells that constitutively expressed higher levels of programmed cell death-1 (PD-1) and constituted ∼ 10% of all B cells in advanced-stage hepatocellular carcinoma (HCC).
This review is focused on molecular momentum transport at fluid-solid interfaces mainly related to microfluidics and nanofluidics in micro-/nano-electro-mechanical systems (MEMS/NEMS). This broad subject covers molecular dynamics behaviors, boundary conditions, molecular momentum accommodations, theoretical and phenomenological models in terms of gas-solid and liquid-solid interfaces affected by various physical factors, such as fluid and solid species, surface roughness, surface patterns, wettability, temperature, pressure, fluid viscosity and polarity. This review offers an overview of the major achievements, including experiments, theories and molecular dynamics simulations, in the field with particular emphasis on the effects on microfluidics and nanofluidics in nanoscience and nanotechnology. In Section 1 we present a brief introduction on the backgrounds, history and concepts. Sections 2 and 3 are focused on molecular momentum transport at gas-solid and liquid-solid interfaces, respectively. Summary and conclusions are finally presented in Section 4.
Herein, we show that coupling boron with cobalt oxide tunes its structure and significantly boost its electrocatalytic performance for the oxygen evolution reaction (OER). Through a simple precipitation and thermal treatment process, a series of Co−B oxides with tunable morphologies and textural parameters were prepared. Detailed structural analysis supported first the formation of an disordered and partially amorphous material with nanosized Co3BO5 and/or Co2B2O6 being present on the local atomic scale. The boron modulation resulted in a superior OER reactivity by delivering a large current and an overpotential of 338 mV to reach a current density of 10 mA cm−2 in 1 M KOH electrolyte. Identical location transmission electron microscopy and in situ electrochemical Raman spectroscopy studies revealed alteration and surface re‐construction of materials, and formation of CoO2 and (oxy)hydroxide intermediate, which were found to be highly dependent on crystallinity of the samples.
The influence of iron on nanocasting of cobalt oxide nanowires and the performance of these materials for the oxygen evolution reaction (OER) are investigated. Pristine Co 3 O 4 and mixed cobalt iron oxide nanowires with a diameter of 7 nm have been synthesized via a nanocasting route by using SBA-15 silica as a template. A small amount of iron added during the synthesis results in a decrease in the nanowires' array length and induces the formation of a bimodal pore size distribution. Raman spectroscopy, X-ray emission, and high-energy resolution X-ray absorption spectroscopies further show that Fe incorporation alters the electronic structure by increasing the average distortion around the cobalt centers and the amount of Co 2+ in tetrahedral sites. These affect the OER activity significantly; the overpotential of pristine Co 3 O 4 at 10 mA/cm 2 decreases from 398 to 378 mV, and the current density at 1.7 V increases from 107 to 150 mA/cm 2 with the addition of iron at the Co/Fe atomic ratio of 32. Furthermore, post-reaction characterization confirmed that both the morphology and electronic structure of nanowires remain intact after a long-term stability test.
Decreasing Li/Ni disorder has been a challenging problem for layered oxide materials, where disorder seriously restricts their electrochemical performances for lithium-ion batteries (LIBs). Element doping is a great strategy that has been widely used to stabilize the structure of the cathode material of an LIB and improve its electrochemical performance. On the basis of the results of previous studies, we hypothesized that the element of Ca, which has a lower valence state and larger radius compared to Ni, would be an ideal doping element to decrease the Li/Ni disorder of LiMO materials and enhance their electrochemical performances. A Ni-rich LiNiMnCoO cathode material was selected as the bare material, which usually shows severe Li/Ni disorder and serious capacity attenuation at a high cutoff voltage. So, a series of Ca-doped LiNiCoMnCaO (x = 0-8%) samples were synthesized by a traditional solid-state method. As hypothesized, neutron diffraction showed that Ca-doped LiNiCoMnO possessed a lower degree of Li/Ni disorder, and potentiostatic intermittent titration results showed a faster diffusion coefficient of Li compared with that of LiNiMnCoO. The Ca-doped LiNiMnCoO samples exhibited higher discharge capacities and better cycle stabilities and rate capabilities, especially under a high cutoff voltage with 4.5 V. In addition, the problems of polarization and voltage reduction of LiNiMnCoO were also alleviated by doping with Ca. More importantly, we infer that it is crucial to choose an appropriate doping element and our findings will help in the research of other layered oxide materials.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.