Search for: All records

Artificial intelligence (AI) has impacted human life at many levels, entailing economic and societal changes. AI algorithms are increasingly used by organizations to generate predictions that feed into decisions (e.g., who is eligible for insurance coverage, approved for bank loans, selected for job interviews). Since the data used for developing the algorithms can contain bias such as gender or racial prejudice, AI predictions can become discriminatory. For-profit and not-for-profit organizations face the hurdles of developing, applying, and maintaining governance of AI, making sure that goal optimization responds to ethical and fairness values.

 

Out-of-distribution (OOD) detection is important for machine learning models deployed in the wild. Recent methods use auxiliary outlier data to regularize the model for improved OOD detection. However, these approaches make a strong distributional assumption that the auxiliary outlier data is completely separable from the in-distribution (ID) data. In this paper, we propose a novel framework that leverages wild mixture data -- that naturally consists of both ID and OOD samples. Such wild data is abundant and arises freely upon deploying a machine learning classifier in their \emph{natural habitats}. Our key idea is to formulate a constrained optimization problem and to show how to tractably solve it. Our learning objective maximizes the OOD detection rate, subject to constraints on the classification error of ID data and on the OOD error rate of ID examples. We extensively evaluate our approach on common OOD detection tasks and demonstrate superior performance. 

A bstract We construct a family of non-supersymmetric extremal black holes and their horizonless microstate geometries in four dimensions. The black holes can have finite angular momentum and an arbitrary charge-to-mass ratio, unlike their supersymmetric cousins. These features make them and their microstate geometries astrophysically relevant. Thus, they provide interesting prototypes to study deviations from Kerr solutions caused by new horizon-scale physics. In this paper, we compute the gravitational multipole structure of these solutions and compare them to Kerr black holes. The multipoles of the black hole differ significantly from Kerr as they depend non-trivially on the charge-to-mass ratio. The horizonless microstate geometries (that are comparable in size to a black hole) have a similar multipole structure as their corresponding black hole, with deviations to the black hole multipole values set by the scale of their microstructure. 

Lithium‐rich layered oxides (LRLO) have attracted great interest for high‐energy Li‐ion batteries due to their high theoretical capacity. However, capacity decay and voltage fade during the cycling impede the practical application of LRLO. Herein, the use of lithium bis‐(oxalate)borate (LiBOB) as an electrolyte additive is reported to improve the cycling stability in high voltage LRLO/graphite full cells. The cell with LiBOB‐containing electrolyte delivers 248 mAh g⁻¹initial capacity and shows no capacity decay after 70 cycles as well as 95.5% retention after 150 cycles over 4.5 V cycling. A systematic mechanism study for the LiBOB‐enabled cycling performance improvement is conducted. Analytical electron microscopy under cryo‐condition confirms the formation of a uniform interphase and less phase transformation on the LRLO particle, accompanied by less voltage decay in the cathode. The formation of B‐F species is identified in the cycled electrolyte, elucidating the HF scavenger effect of LiBOB. Due to less HF corrosion on both electrode interphases, a reduced amount of transition metal dissolution and redeposition on the graphite is proved, thereby mitigating the capacity decay in LRLO/graphite full cells. These findings suggest that the borate additive is a promising strategy to optimize high voltage electrolyte for the industrialization of LRLO.

 

The recent proliferation of renewable energy generation offers mankind hope, with regard to combatting global climate change. However, reaping the full benefits of these renewable energy sources requires the ability to store and distribute any renewable energy generated in a cost‐effective, safe, and sustainable manner. As such, sodium‐ion batteries (NIBs) have been touted as an attractive storage technology due to their elemental abundance, promising electrochemical performance and environmentally benign nature. Moreover, new developments in sodium battery materials have enabled the adoption of high‐voltage and high‐capacity cathodes free of rare earth elements such as Li, Co, Ni, offering pathways for low‐cost NIBs that match their lithium counterparts in energy density while serving the needs for large‐scale grid energy storage. In this essay, a range of battery chemistries are discussed alongside their respective battery properties while keeping metrics for grid storage in mind. Matters regarding materials and full cell cost, supply chain and environmental sustainability are discussed, with emphasis on the need to eliminate several elements (Li, Ni, Co) from NIBs. Future directions for research are also discussed, along with potential strategies to overcome obstacles in battery safety and sustainable recyclability.